Computer implemented methods and systems for improved data retrieval

BACKGROUND

0001The present invention relates to computer implemented methods and systems for improved data retrieval. In the <patcit id="pcit0001" dnum="US6421662A"><text>United States Patent no. 6 421 662 a</text></patcit> method and apparatus provided for retrieving data from a table in a data base is described. The patent describes generation of an index for processing a query by selecting values as unique keys form one or more columns in a table in a relational database system. Each key corresponds to a row in the table. The condition of whether a row in a table satisfies a search criterion, is represented by a bit in a bit sequence. In an alternative embodiment it is described that a bitstring is associated with multiple search criteria (statuses), each row in the table corresponds to a series of bits in the bitstring (one bit for each status), and each bit in the series of bits that correspond to the row indicates whether the row satisfies a different one of the statuses associated with the bitstring.

SUMMARY

0002It is to be understood that both the following general description and the following detailed description are exemplary and explanatory only and are not restrictive, as claimed. Provided are methods and systems for data management and analysis. The invention is set out in the appended claims.

0003Computer implemented methods are described comprising determining data values of a plurality of unsorted data values stored in one or more databases and generating a data structure. Methods are further described comprising determining data values of a plurality of data values stored in one or more databases that satisfy a first threshold, wherein the plurality of data values are unsorted, representing, in a first data structure, data values of the plurality of data values that satisfy the first threshold with a first symbol, and representing, in the first data structure, data values of the plurality of data values that do not satisfy the first threshold with a second symbol. Methods are further described comprising applying a first filter to a plurality of unsorted data values stored in one or more databases and applying a second filter to the plurality of unsorted data values stored in the one or more databases. Moreover, methods are described including receiving a query, determining that the query is related a filter of the plurality of filters, determining, in the data structure, the column associated with the filter, determining, in the data structure, the rows intersecting the column, wherein the intersection of the rows and the column comprise cells containing the first symbol, and retrieving the data values of the plurality of unsorted data values associated with the rows from the one or more databases. Further described is a data storage and retrieval system for a computer memory, comprising a memory, configured according to a logical table, the logical table comprising, a plurality of logical rows, each said logical row including one or more of a first symbol or a second symbol, wherein the first symbol and the second symbol represent a result of application of a filter to a plurality of unsorted data values stored in one or more databases, each said logical row corresponding to a respective data value of the plurality of unsorted data values, and a plurality of logical columns intersecting said plurality of logical rows to define a plurality of logical cells, each said logical column including a filter identifier to identify each said logical column.

0004One advantage of the data structure is that an amount of data that ordinarily is extracted in order to determine data satisfying a query may be reduced. Moreover, the targeted extraction of data values that satisfy a filter may be achieved may use less time and computational resources.

0005Additional advantages will be set forth in part in the description which follows or may be learned by practice. The advantages will be realized and attained by means of the elements and combinations particularly pointed out in the appended claims.

BRIEF DESCRIPTION OF THE DRAWINGS

0006The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments and together with the description, serve to explain the principles of the methods and systems: <dl id="dl0001" compact="compact"><dt>Figure 1</dt><dd>is a schematic diagram showing an embodiment of a system forming an implementation of the disclosed methods;</dd><dt>Figure 2</dt><dd>is a set of tables showing exemplary Tables 1-5 of a simple database and associations between variables in the tables;</dd><dt>Figure 3</dt><dd>is an overview of the relations between data model, indexes in disk and windowed view of disk indexes in memory;</dd><dt>Figure 4</dt><dd>illustrates an example virtual record;</dd><dt>Figure 5</dt><dd>is a representation of the data structure used for indexlets;</dd><dt>Figure 6</dt><dd>illustrates an example of inter-table inferencing using indexlets;</dd><dt>Figure 7</dt><dd>is a representation of the data structure used for table handling;</dd><dt>Figure 8</dt><dd>is a table tree representation of a data model;</dd><dt>Figure 9</dt><dd>illustrates an example application of bidirectional table indexes and bidirectional association indexes;</dd><dt>Figure 10</dt><dd>illustrates an example of linking indexlets of different tables;</dd><dt>Figure 11</dt><dd>illustrates an example method for database traversals;</dd><dt>Figure 12</dt><dd>illustrates an example system architecture for scalable index generation and chart generation;</dd><dt>Figure 13</dt><dd>illustrates an example indexation service;</dd><dt>Figure 14</dt><dd>illustrates an example global symbol service;</dd><dt>Figure 15</dt><dd>illustrates load distribution in a single table case;</dd><dt>Figure 16</dt><dd>illustrates load distribution in a multiple table case;</dd><dt>Figure 17</dt><dd>illustrates I2I mapping;</dd><dt>Figure 18</dt><dd>illustrates an example distributed system for data model traversal and hypercube evaluation;</dd><dt>Figure 19</dt><dd>is a schematic flowchart showing basic steps performed when extracting information from a database;</dd><dt>Figure 20</dt><dd>is a set of tables showing a final data structure, e.g. a multidimensional cube, created by evaluating mathematical functions;</dd><dt>Figure 21</dt><dd>is a schematic diagram showing how a selection by a user operates on a scope to generate a data subset;</dd><dt>Figure 22</dt><dd>is a schematic graphical presentation showing selections and a diagram of data associated to the selections as received after processing by an external engine;</dd><dt>Figure 23</dt><dd>is a schematic representation of data exchanged with an external engine based on selections in <figref idref="f0022">FIG. 22</figref>;</dd><dt>Figure 24</dt><dd>is a schematic graphical presentation showing selections and a diagram of data associated to the selections as received after second computations from an external engine;</dd><dt>Figure 25</dt><dd>is a schematic representation of data exchanged with an external engine based on selections in <figref idref="f0024">FIG. 24</figref>;</dd><dt>Figure 26</dt><dd>is a schematic graphical presentation showing selections and a diagram of data associated to the selections;</dd><dt>Figure 27</dt><dd>is a schematic representation of data exchanged with an external engine based on selections in <figref idref="f0026">FIG. 26</figref>;</dd><dt>Figure 28</dt><dd>is a schematic graphical presentation showing a further set of selections and a diagram of data associated to the selections as received after third computations from an external engine;</dd><dt>Figure 29</dt><dd>is a table showing results from computations based on different selections in the presentation of <figref idref="f0026">FIG. 26</figref>;</dd><dt>Figure 30</dt><dd>is diagram illustrating an example sorting technique and data structure;</dd><dt>Figure 31</dt><dd>is a flow chart illustrating another example method;</dd><dt>Figure 32</dt><dd>is a flow chart illustrating another example method;</dd><dt>Figure 33</dt><dd>is a flow chart illustrating another example method; and</dd><dt>Figure 34</dt><dd>is an exemplary operating environment for performing the disclosed methods.</dd></dl>

DETAILED DESCRIPTION

0007Various embodiments of the disclosed methods and arrangements are discussed in detail below. While specific implementations are discussed, it should be understood that this is done for illustration purposes only. A person skilled in the relevant art will recognize that other components, configurations, and steps may be used without parting from the spirit and scope of the disclosure.

0008Before the present methods and systems are disclosed and described in more detail, it is to be understood that the methods and systems are not limited to specific steps, processes, components, or structure described, or to the order or particular combination of such steps or components as described. It is also to be understood that the terminology used herein is for the purpose of describing exemplary embodiments only and is not intended to be restrictive or limiting.

0009As used herein the singular forms "a," "an," and "the" include both singular and plural referents unless the context clearly dictates otherwise. Values expressed as approximations, by use of antecedents such as "about" or "approximately," shall include reasonable variations from the referenced values. If such approximate values are included with ranges, not only are the endpoints considered approximations, the magnitude of the range shall also be considered an approximation. Lists are to be considered exemplary and not restricted or limited to the elements comprising the list or to the order in which the elements have been listed unless the context clearly dictates otherwise.

0010Throughout the specification and claims of this disclosure, the following words have the meaning that is set forth: "comprise" and variations of the word, such as "comprising" and "comprises," mean including but not limited to, and are not intended to exclude, for example, other additives, components, integers or steps. "Exemplary" means "an example of", but not essential, necessary, or restricted or limited to, nor does it convey an indication of a preferred or ideal embodiment. "Include" and variations of the word, such as "including" are not intended to mean something that is restricted or limited to what is indicated as being included, or to exclude what is not indicated. "May" means something that is permissive but not restrictive or limiting. "Optional" or "optionally" means something that may or may not be included without changing the result or what is being described. "Prefer" and variations of the word such as "preferred" or "preferably" mean something that is exemplary and more ideal, but not required. "Such as" means something that is exemplary.

0011Steps and components described herein as being used to perform the disclosed methods and construct the disclosed systems are exemplary unless the context clearly dictates otherwise. It is to be understood that when combinations, subsets, interactions, groups, etc. of these steps and components are disclosed, that while specific reference of each various individual and collective combinations and permutation of these may not be explicitly disclosed, each is specifically contemplated and described herein, for all methods and systems. This applies to all aspects of this application including, but not limited to, steps in disclosed methods and/or the components disclosed in the systems. Thus, if there are a variety of additional steps that can be performed or components that can be added, it is understood that each of these additional steps can be performed and components added with any specific embodiment or combination of embodiments of the disclosed systems and methods.

0012The present methods and systems may be understood more readily by reference to the following detailed description of preferred embodiments and the Examples included therein and to the Figures and their previous and following description.

0013As will be appreciated by one skilled in the art, the methods and systems may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the methods and systems may take the form of a computer program product on a computer-readable storage medium having computer-readable program instructions (e.g., computer software) embodied in the storage medium. More particularly, the present methods and systems may take the form of web-implemented computer software. Any suitable computer-readable storage medium may be utilized including hard disks, CD-ROMs, optical storage devices, or magnetic storage devices, whether internal, networked or cloud based.

0014Embodiments of the methods and systems are described below with reference to diagrams, flowcharts and other illustrations of methods, systems, apparatuses and computer program products. It will be understood that each block of the block diagrams and flowchart illustrations, and combinations of blocks in the block diagrams and flowchart illustrations, respectively, can be implemented by computer program instructions. These computer program instructions may be loaded onto a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions which execute on the computer or other programmable data processing apparatus create a means for implementing the functions specified in the flowchart block or blocks.

0015These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including computer-readable instructions for implementing the function specified in the flowchart block or blocks. The computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer-implemented process such that the instructions that execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart block or blocks.

0016Accordingly, blocks of the block diagrams and flowchart illustrations support combinations of means for performing the specified functions, combinations of steps for performing the specified functions and program instruction means for performing the specified functions. It will also be understood that each block of the block diagrams and flowchart illustrations, and combinations of blocks in the block diagrams and flowchart illustrations, can be implemented by special purpose hardware-based computer systems that perform the specified functions or steps, or combinations of special purpose hardware and computer instructions.

0017The present disclosure relates to computer implemented methods and systems for data management, data analysis, and processing. The disclosed methods and systems can incorporate external data analysis into an otherwise closed data analysis environment. A typical environment for the systems and methods described herein is for assisting in a computer implemented method for building and updating a multi-dimensional cube data structure, such as, e.g., the systems and methods described in <patcit id="pcit0002" dnum="US7058621B"><text>U.S. Pat. Nos. 7,058,621</text></patcit>; <patcit id="pcit0003" dnum="US8745099B"><text>8,745,099</text></patcit>; <patcit id="pcit0004" dnum="US8244741B"><text>8,244,741</text></patcit>; and <patcit id="pcit0005" dnum="US054321" dnum-type="L"><text>U.S. Pat. App. No. 14/054,321</text></patcit>.

0018In an aspect, the methods and systems can manage associations among data sets with every data point in the analytic dataset being associated with every other data point in the dataset. Datasets can be larger than hundreds of tables with thousands of fields. A multi-dimensional dataset or array of data is referred to as an OnLine Analytic Processing (OLAP) cube. A cube can be considered a multi-dimensional generalization of a two- or three-dimensional spreadsheet. For example, it may be desired to summarize financial data by product, by time-period, and by city to compare actual and budget expenses. Product, time, city, and scenario (actual and budget) can be referred to as dimensions. A multi-dimensional dataset is normally called a hypercube if the number of dimensions is greater than 3. A hypercube can comprise tuples made of two (or more) dimensions and one or more expressions.

0019<figref idref="f0001"><b>FIG. 1</b></figref> illustrates an associative data indexing engine <b>100</b> with data flowing in from the left and operations starting from a script engine <b>104</b> and going clockwise (indicated by the clockwise arrow) to export features <b>118.</b> Data from a data source <b>102</b> can be extracted by a script engine <b>104.</b> The data source <b>102</b> can comprise any type of known database, such as relational databases, post-relational databases, object-oriented databases, hierarchical databases, flat files, spread sheet, etc. The Internet may also be regarded as a database in the context of the present disclosure. A visual interface can be used as an alternative or combined with a script engine <b>104.</b> The script engine <b>104</b> can read record by record from the data source <b>102</b> and data can be stored or appended to symbol and data tables in an internal database <b>120.</b> Read data can be referred to as a data set.

0020An example data source <b>102</b> (e.g., a database), as shown in <figref idref="f0002"><b>FIG. 2</b></figref><b>,</b> can comprise a number of data tables (Tables 1-5). Each data table can contain data values of a number of data variables. For example, in Table 1 each data record contains data values of the data variables "Product," "Price," and "Part." If there is no specific value in a field of the data record, this field is considered to hold a NULL-value. Similarly, in Table 2 each data record contains values of the variables "Date," "Client," "Product," and "Number." In Table 3 each data record contains values of variable "Date" as "Year," "Month" and "Day." In Table 4 each data record contains values of variables "Client" and "Country," and in Table 5 each data record contains values of variables "Country," "Capital," and "Population." Typically, the data values are stored in the form of ASCII-coded strings, but can be stored in any form.

0021Each of the data elements of the database shown in Tables 1-5 of <figref idref="f0002"><b>FIG. 2</b></figref> has a data element type and a data element value (for example "Client" is the data element type and "Nisse" is the data element value). Multiple records can be stored in different database structures such as data cubes, data arrays, data strings, flat files, lists, vectors, and the like; and the number of database structures can be greater than or equal to one and can comprise multiple types and combinations of database structures. While these and other database structures can be used with, and as part of, the methods and systems disclosed, the remaining description will refer to tables, vectors, strings and data cubes solely for convenience.

0022Additional database structures can be included within the database illustrated as an example herein, with such structures including additional information pertinent to the database such as, in the case of products for example; color, optional packages, etc. Each table can comprise a header row which can identify the various data element types, often referred to as the dimensions or the fields, that are included within the table. Each table can also have one or more additional rows which comprise the various records making up the table. Each of the rows can contain data element values (including null) for the various data element types comprising the record.

0023Returning to <figref idref="f0001"><b>FIG. 1</b></figref><b>,</b> in an aspect, the extraction of the data can comprise extracting an initial data set or scope from the data source <b>102,</b> e.g. by reading the initial data set into the primary memory (e.g. RAM) of the computer. The initial data set can comprise the entire contents of the data source <b>102,</b> or a subset thereof. The internal database <b>120</b> can comprise the extracted data and symbol tables. Symbol tables can be created for each field and, in one aspect, can only contain the distinct field values, each of which can be represented by their clear text meaning and a bit filled pointer. The data tables can contain said bit filled pointers.

0024In the case of a query of the data source <b>102,</b> a scope can be defined by the tables included in a SELECT statement (or equivalent) and how these are joined. In an aspect, the SELECT statement can be SQL (Structured Query Language) based. For an Internet search, the scope can be an index of found web pages, for example, organized as one or more tables. A result of scope definition can be a data set.

0025Once the data has been extracted, a user interface can be generated to facilitate dynamic display of the data. By way of example, a particular view of a particular dataset or data subset generated for a user can be referred to as a state space or a session. The methods and systems can dynamically generate one or more visual representations of the data to present in the state space.

0026A user can make a selection in the data set, causing a logical inference engine <b>106</b> to evaluate a number of filters on the data set. For example, a query on a database that holds data of placed orders, could be requesting results matching an order year of ' 1999' and a client group be 'Nisse.' The selection may thus be uniquely defined by a list of included fields and, for each field, a list of selected values or, more generally, a condition. Based on the selection, the logical inference engine <b>106</b> can generate a data subset that represents a part of the scope. The data subset may thus contain a set of relevant data records from the scope, or a list of references (e.g. indices, pointers, or binary numbers) to these relevant data records. The logical inference engine <b>106</b> can process the selection and can determine what other selections are possible based on the current selections. In an aspect, flags can enable the logical inference engine <b>106</b> to work out the possible selections. By way of example, two flags can be used: the first flag can represent whether a value is selected or not, the second can represent whether or not a value selection is possible. For every click in an application, states and colors for all field values can be calculated. These can be referred to as state vectors, which can allow for state evaluation propagation between tables.

0027The logical inference engine <b>106</b> can utilize an associative model to connect data. In the associative model, all the fields in the data model have a logical association with every other field in the data model. An example data model <b>301</b> is shown in <figref idref="f0003"><b>FIG. 3</b></figref><b>.</b> The data model <b>301</b> illustrates connections between a plurality of tables which represent logical associations. Depending on the amount of data, the data model <b>301</b> can be too large to be loaded into memory. To address this issue, the logical inference engine <b>106</b> can generate one or more indexes for the data model. The one or more indexes can be loaded into memory in lieu of the data model <b>301.</b> The one or more indexes can be used as the associative model. An index is used by database management programs to provide quick and efficient associative access to a table's records. An index is a data structure (for example, a B-tree, a hash table, and the like) that stores attributes (e.g., values) for a specific column in a table. A B-tree is a self-balancing tree data structure that keeps data sorted and allows searches, sequential access, insertions, and deletions in logarithmic time. The B-tree is a generalization of a binary search tree in that a node can have more than two children. A hash table (also referred to as a hash index) can comprise a collection of buckets organized in an array. A hash function maps index keys to corresponding buckets in the hash index.

0028Queries that compare for equality to a string can retrieve values very fast using a hash index. For instance, referring to the tables of <figref idref="f0002"><b>FIG. 2</b></figref><b>,</b> a query of SELECT * FROM Table 2 WHERE Client = 'Kalle') could benefit from a hash index created on the Client column. In this example, the hash index would be configured such that the column value will be the key into the hash index and the actual value mapped to that key would just be a pointer to the row data in Table 2. Since a hash index is an associative array, a typical entry can comprise "Kalle => 0x29838", where 0x29838 is a reference to the table row where Kalle is stored in memory. Thus, looking up a value of "Kalle" in a hash index can return a reference to the row in memory which is faster than scanning Table 2 to find all rows with a value of "Kalle" in the Client column. The pointer to the row data enables retrieval of other values in the row. Disclosed are methods and systems that enable not only efficient processing of single queries, but also one or more batches of queries.

0029Thus, the logical inference engine <b>106</b> can determine a data subset based on user selections. The logical inference engine <b>106</b> automatically maintains associations among every piece of data in the entire data set used in an application. The logical inference engine <b>106</b> can store the binary state of every field and of every data table dependent on user selection (e.g., included or excluded). This can be referred to as a state space and can be updated by the logical inference engine <b>106</b> every time a selection is made. There is one bit in the state space for every value in the symbol table or row in the data table, as such the state space is smaller than the data itself and faster to query. The inference engine will work associating values or binary symbols into the dimension tuples. Dimension tuples are normally needed by a hypercube to produce a result.

0030The associations thus created by the logical inference engine <b>106</b> means that when a user makes a selection, the logical inference engine <b>106</b> can resolve (quickly) which values are still valid (e.g., possible values) and which values are excluded. The user can continue to make selections, clear selections, and make new selections, and the logical inference engine <b>106</b> will continue to present the correct results from the logical inference of those selections. In contrast to a traditional join model database, the associative model provides an interactive associative experience to the user.

0031Based on current selections and possible rows in data tables a calculation/chart engine <b>108</b> can calculate aggregations in objects forming transient hyper cubes in an application. The calculation/chart engine <b>108</b> can further build a virtual temporary table from which aggregations can be made. The calculation/chart engine <b>108</b> can perform a calculation (e.g., evaluate an expression in response to a user selection/de-selection) via a multithreaded operation. The state space can be queried to gather all of the combinations of dimensions and values necessary to perform the calculation. In an aspect, the query can be on one thread per object, one process, one worker, combinations thereof, and the like. The expression can be calculated on multiple threads per object. Results of the calculation can be passed to a rendering engine <b>116</b> and/or optionally to an extension engine <b>110.</b>

0032In an aspect, the chart engine <b>108</b> can receive dimensions, expressions, and sorting parameters and can compute a hypercube data structure containing aggregations along the dimensions. For example, a virtual record can be built with a placeholder for all field values (or indices) needed, as a latch memory location. When all values are assigned, the virtual record can be processed to aggregate the fields needed for computations and save the dimension values in a data structure per row of the resulting hypercube. In such a way, the traversal of the database can be done in an arbitrary way, just depending on requirements provided by memory consumption and indexing techniques used for the particular case at hand. An example virtual record is shown in <figref idref="f0004"><b>FIG. 4</b></figref><b>.</b> This example considers the dimensions Order Date and Country/Region and the expression sum(Unit Price * Quantity).

0033Optionally, the extension engine <b>110</b> can be implemented to communicate data via an interface <b>112</b> to an external engine <b>114.</b> In another aspect, the extension engine <b>110</b> can communicate data, metadata, a script, a reference to one or more artificial neural networks (ANNs), one or more commands to be executed, one or more expressions to be evaluated, combinations thereof, and the like to the external engine <b>114.</b> The interface <b>114</b> can comprise, for example, an Application Programming Interface (API). The external engine <b>114</b> can comprise one or more data processing applications (e.g., simulation applications, statistical applications, mathematical computation applications, database applications, combinations thereof, and the like). The external engine <b>114</b> can be, for example, one or more of MATLAB^®, R, Maple^®, Mathematica^®, combinations thereof, and the like.

0034In an aspect, the external engine <b>114</b> can be local to the associative data indexing engine <b>100</b> or the external engine <b>114</b> can be remote from the associative data indexing engine <b>100.</b> The external engine <b>114</b> can perform additional calculations and transmit the results to the extension engine <b>110</b> via the interface <b>112.</b> A user can make a selection in the data model of data to be sent to the external engine <b>114.</b> The logical inference engine <b>106</b> and/or the extension engine <b>110</b> can generate data to be output to the external engine <b>114</b> in a format to which the external engine <b>114</b> is accustomed to processing. In an example application, tuples forming a hypercube can comprise two dimensions and one expression, such as (Month, Year, Count (ID)), ID being a record identification of one entry. Then said tuples can be exchanged with the external engine <b>114</b> through the interface <b>112</b> as a table. If the data comprise births there can be timestamps of the births and these can be stored as month and year. If a selection in the data model will give a set of month-year values that are to be sent out to an external unit, the logical inference engine <b>106</b> and/or the extension engine <b>110</b> can ripple that change to the data model associatively and produce the data (e.g., set and/or values) that the external engine <b>114</b> needs to work with. The set and/or values can be exchanged through the interface <b>112</b> with the external engine <b>114.</b> The external engine <b>114</b> can comprise any method and/or system for performing an operation on the set and/or values. In an aspect, operations on the set and/or values by the external engine <b>114</b> can be based on tuples (aggregated or not). In an aspect, operations on the set and/or values by the external engine <b>114</b> can comprise a database query based on the tuples. Operations on the set and/or values by the external engine <b>114</b> can be any transformation/operation of the data as long as the cardinality of the result is consonant to the sent tuples/hypercube result.

0035In an aspect, tuples that are transmitted to the external engine <b>114</b> through the interface <b>112</b> can result in different data being received from the external engine <b>114</b> through the interface <b>112.</b> For example, a tuple consisting of (Month, Year, Count (ID)) should return as 1-to-1, m-to-1 (where aggregations are computed externally) or n-to-n values. If data received are not what were expected, association can be lost. Transformation of data by the external engine <b>114</b> can be configured such that cardinality of the results is consonant to the sent tuples and/or hypercube results. The amount of values returned can thus preserve associativity.

0036Results received by the extension engine <b>110</b> from the external engine <b>114</b> can be appended to the data model. In an aspect, the data can be appended to the data model without intervention of the script engine <b>104.</b> Data model enrichment is thus possible "on the fly." A natural work flow is available allowing clicking users to associatively extend the data. The methods and systems disclosed permit incorporation of user implemented functionality into a presently used work flow. Interaction with third party complex computation engines, such as MATLAB^® or R, is thus facilitated.

0037The logical inference engine <b>106</b> can couple associated results to the external engine <b>114</b> within the context of an already processed data model. The context can comprise tuple or tuples defined by dimensions and expressions computed by hypercube routines. Association is used for determination of which elements of the present data model are relevant for the computation at hand. Feedback from the external engine <b>114</b> can be used for further inference inside the inference engine or to provide feedback to the user.

0038A rendering engine <b>116</b> can produce a desired graphical object (charts, tables, etc.) based on selections/calculations. When a selection is made on a rendered object there can be a repetition of the process of moving through one or more of the logical inference engine <b>106,</b> the calculation/chart engine <b>108,</b> the extension engine <b>110,</b> the external engine <b>114,</b> and/or the rendering engine <b>116.</b> The user can explore the scope by making different selections, by clicking on graphical objects to select variables, which causes the graphical object to change. At every time instant during the exploration, there exists a current state space, which is associated with a current selection state that is operated on the scope (which always remains the same).

0039Different export features or tools <b>118</b> can be used to publish, export or deploy any output of the associative data indexing engine <b>100.</b> Such output can be any form of visual representation, including, but not limited to, textual, graphical, animation, audio, tactile, and the like.

0040As described previously with regard to <figref idref="f0001"><b>FIG. 1</b></figref><b>,</b> the logical inference engine <b>106</b> can generate one or more indexes for the data model. The one or more indexes can be loaded into memory in lieu of the data model <b>301</b> shown in <figref idref="f0003"><b>FIG. 3</b></figref><b>.</b> As further shown in <figref idref="f0003"><b>FIG. 3</b></figref><b>,</b> the logical inference engine <b>106</b> can be configured for generating one or more bidirectional table indexes (BTI) <b>302a, 302b, 302c, 302d,</b> and/or <b>302e</b> and one or more bidirectional associative indexes (BAI) <b>303a, 303b, 303c</b> and/or <b>303d</b> based on a data model <b>301.</b> The logical inference engine <b>106</b> can scan each table in the data model <b>301</b> and create the BTI <b>302a, 302b, 302c, 302d,</b> and/or <b>302e.</b> A BTI can be created for each column of each table in the data. The BTI <b>302a, 302b, 302c, 302d,</b> and/or <b>302e</b> can comprise a hash index. The BTI <b>302a, 302b, 302c, 302d,</b> and/or <b>302e</b> can comprise first attributes and pointers to the table rows comprising the first attributes. For example, referring to the tables of <figref idref="f0002"><b>FIG. 2</b></figref><b>,</b> an example BTI <b>302a</b> can comprise "Kalle => 0x29838", where Kalle is an attribute found in Table 2 and 0x29838 is a reference to the row in Table 2 where Kalle is stored in memory. Thus, the BTI <b>302a, 302b, 302c, 302d,</b> and/or <b>302e</b> can be used to determine other attributes in other columns (e.g., second attributes, third attributes, etc...) in table rows comprising the first attributes. Accordingly, the BTI can be used to determine that an association exists between the first attributes and the other attributes.

0041The logical inference engine <b>106</b> can scan one or more of BTI <b>302a, 302b, 302c, 302d,</b> and/or <b>302e</b> and create the BAI <b>303a, 303b, 303c</b> and/or <b>303d.</b> The BAI <b>303a, 303b, 303c</b> and/or <b>303d</b> can comprise a hash index. The BAI <b>303a, 303b, 303c</b> and/or <b>303d</b> can comprise an index configured for connecting attributes in a first table to common columns in a second table. The BAI <b>303a, 303b, 303c</b> and/or <b>303d</b> thus allows for identification of rows in the second table which then permits identification of other attributes in other tables. For example, referring to the tables of <figref idref="f0002"><b>FIG. 2</b></figref><b>,</b> an example BAI <b>303a</b> can comprise "Kalle => 0x39838", where Kalle is an attribute found in Table 2 and 0x39838 is a reference to a row in Table 4 that contains Kalle. In an aspect, the reference can be to a hash that can be in-memory or on disk.

0042Using the BTI <b>302a, 302b, 302c, 302d,</b> and/or <b>302e</b> and the BAI <b>303a, 303b, 303c,</b> and/or <b>303d,</b> the logical inference engine <b>106</b> can generate an index window <b>304</b> by taking a portion of the data model <b>301</b> and mapping it into memory. The portion of the data model <b>301</b> taken into memory can be sequential (e.g., not random). The result is a significant reduction in the size of data required to be loaded into memory.

0043In an aspect, bidirectional indexing using BTIs can have limits as to how much parallelization can be applied when processing the data model <b>301.</b> To improve parallelization applied to the data model <b>301,</b> the logical inference engine <b>106</b> can generate bidirectional indexes for partitions for a table in the data model <b>301.</b> Such bidirectional indexes are hereinafter referred to as "indexlets," and the partitions for the table corresponding to an indexlet can be referred to as a "datalet." In an aspect, the logical inference engine <b>106</b> can generate indexlets for a given table by partitioning the table into blocks of rows (e.g., datalets). In an aspect, the blocks of rows can be of a same size. In an aspect, a last block of rows can be of a size less than the remaining blocks of rows. In an aspect, after partitioning the blocks of rows, the logical inference engine can generate an indexlet for each of the blocks of rows (e.g., datalets). In an aspect, generating an indexlet for a given block of rows (e.g., datalets) comprises generating a bidirectional index as described above, but limited in scope to the given block of rows (e.g., datalet). An indexlet is the smallest computational unit using a single BTI. BTI's can be used globally, but such global usage impairs scale-out possibilities. Using indexlets thus ensures scalability and distributed use of BTI's.

0044A representation of a data structure for indexlets is shown in <figref idref="f0005"><b>FIG. 5</b></figref><b>.</b> Rows of a given table <b>550</b> can be divided into datalets bidirectionally indexed by indexlets <b>552, 554</b> and <b>556,</b> respectively. In the example of <figref idref="f0005"><b>FIG.</b> 5</figref>, the indexlet <b>552</b> can include pointers or references to respective columns <b>558, 560,</b> and <b>562</b> as set forth above with respect to bidirectional table indexes. Each of the indexlets <b>552, 554,</b> and <b>556</b> are logically associated with bidirectional global attribute lists <b>564</b> and <b>566</b> that index a particular attribute to the blocks it is present in. Accordingly, an entry in the bidirectional global attribute list <b>564</b> and <b>566</b> for a given attribute can comprise a reference to an indexlet corresponding to a block having the respective attribute. In an aspect, the reference can include a hash reference. In an aspect, as shown in <figref idref="f0006"><b>FIG. 6</b></figref><b>,</b> an implicit relationship exists between indexlets in different tables through a common field present in both tables and an attribute-to-attribute (A2A) index <b>672.</b>

0045<figref idref="f0007"><b>FIG. 7</b></figref> illustrates an example application of one or more BTIs. User input <b>704</b> can be received that impacts a selection of one or more attribute states <b>706.</b> Attribute states <b>706</b> can correspond to selection by a user of one or more attributes (e.g., values) found in Column 1 of Table X. In an aspect, the one or more attributes of Table X can comprise a hash of each respective attribute. One or more BTI's <b>708</b> can be accessed to determine one or more rows in Table X that comprise the attributes selected by the user. Row states <b>710</b> can correspond to selection of one or more rows found in Table X that comprise the one or more selected attributes. An inverted index <b>712</b> of Column 2 can be accessed to identify which rows of Table 1 comprise associated attributes. Attribute states <b>714</b> for Column 2 can be updated to reflect the associated attributes of Column 2. One or more BTI's <b>718</b> can be further accessed to determine other associated attributes in other columns as needed. Attribute states <b>714</b> can be applied to other tables via one or more BAIs. <figref idref="f0008"><b>FIG. 8</b></figref> illustrates an example of relationships identified by one or more BAIs.

0046<figref idref="f0009"><b>FIG. 9</b></figref> illustrates an example application of BTI's and BAI's to determine inferred states both inter-table and intra-table using Table 1 and Table 2 of <figref idref="f0002"><b>FIG. 2</b></figref><b>.</b> A BTI <b>920</b> can be generated for the "Client" attribute of Table 2. In an aspect, the BTI <b>920</b> can comprise an inverted index <b>921.</b> In other aspect, the inverted index <b>921</b> can be considered a separate structure. The BTI <b>920</b> can comprise a row for each unique attribute in the "Client" column of Table 2. Each unique attribute can be assigned a corresponding position <b>922</b> in the BTI <b>920.</b> In an aspect, the BTI <b>920</b> can comprise a hash for each unique attribute. The BTI <b>920</b> can comprise a column <b>923</b> for each row of Table 2. For each attribute, a "1" can indicate the presence of the attribute in the row and a "0" can indicate an absence of the attribute from the row. "0" and "1" are merely examples of values used to indicate presence or absence. Thus, the BTI <b>920</b> reflects that the attribute "Nisse" is found in rows 1 and 6 of Table 2, the attribute "Gullan" is found in row 2 of Table 2, the attribute "Kalle" is found in rows 3 and 4 of Table 2, and the attribute "Pekka" is found in row 5 of Table 2.

0047The inverted index <b>921</b> can be generated such that each position in the inverted index <b>921</b> corresponds to a row of Table 2 (e.g., first position corresponds to row 1, second position corresponds to row 2, etc.. A value can be entered into each position that reflects the corresponding position <b>922</b> for each attribute. Thus, in the inverted index <b>921,</b> position 1 comprises the value "1" which is the corresponding position <b>922</b> value for the attribute "Nisse", position 2 comprises the value "2" which is the corresponding position <b>922</b> value for the attribute "Gullan", position 3 comprises the value "3" which is the corresponding position <b>922</b> value for the attribute "Kalle", position 4 comprises the value "3" which is the corresponding position <b>922</b> value for the attribute "Kalle", position 5 comprises the value "4" which is the corresponding position <b>922</b> value for the attribute "Pekka", and position 6 comprises the value "1" which is the corresponding position <b>922</b> value for the attribute "Nisse".

0048A BTI <b>924</b> can be generated for the "Product" attribute of Table 2. In an aspect, the BTI <b>924</b> can comprise an inverted index <b>925.</b> In other aspect, the inverted index <b>925</b> can be considered a separate structure. The BTI <b>924</b> can comprise a row for each unique attribute in the "Product" column of Table 2. Each unique attribute can be assigned a corresponding position <b>926</b> in the BTI <b>924.</b> In an aspect, the BTI <b>924</b> can comprise a hash for each unique attribute. The BTI <b>924</b> can comprise a column <b>927</b> for each row of Table 2. For each attribute, a "1" can indicate the presence of the attribute in the row and a "0" can indicate an absence of the attribute from the row. "0" and "1" are merely examples of values used to indicate presence or absence. Thus, the BTI <b>924</b> reflects that the attribute "Toothpaste" is found in row 1 of Table 2, the attribute "Soap" is found in rows 2, 3, and 5 of Table 2, and the attribute "Shampoo" is found in rows 4 and 6 of Table 2.

0049By way of example, the inverted index <b>925</b> can be generated such that each position in the inverted index <b>925</b> corresponds to a row of Table 2 (e.g., first position corresponds to row 1, second position corresponds to row 2, etc...). A value can be entered into each position that reflects the corresponding position <b>926</b> for each attribute. Thus, in the inverted index <b>925,</b> position 1 comprises the value "1" which is the corresponding position <b>926</b> value for the attribute "Toothpaste", position 2 comprises the value "2" which is the corresponding position <b>926</b> value for the attribute "Soap", position 3 comprises the value "2" which is the corresponding position <b>926</b> value for the attribute "Soap", position 4 comprises the value "3" which is the corresponding position <b>926</b> value for the attribute "Shampoo", position 5 comprises the value "2" which is the corresponding position <b>926</b> value for the attribute "Soap", and position 6 comprises the value "3" which is the corresponding position <b>926</b> value for the attribute "Shampoo".

0050By way of example, a BTI <b>928</b> can be generated for the "Product" attribute of Table 1. In an aspect, the BTI <b>928</b> can comprise an inverted index <b>929.</b> In other aspect, the inverted index <b>929</b> can be considered a separate structure. The BTI <b>928</b> can comprise a row for each unique attribute in the "Product" column of Table 1. Each unique attribute can be assigned a corresponding position <b>930</b> in the BTI <b>928.</b> In an aspect, the BTI <b>928</b> can comprise a hash for each unique attribute. The BTI <b>928</b> can comprise a column <b>931</b> for each row of Table 1. For each attribute, a "1" can indicate the presence of the attribute in the row and a "0" can indicate an absence of the attribute from the row. "0" and "1" are merely examples of values used to indicate presence or absence. Thus, the BTI <b>928</b> reflects that the attribute "Soap" is found in row 1 of Table 1, the attribute "Soft Soap" is found in row 2 of Table 1, and the attribute "Toothpaste" is found in rows 3 and 4 of Table 1.

0051By way of example, the inverted index <b>929</b> can be generated such that each position in the inverted index <b>929</b> corresponds to a row of Table 1 (e.g., first position corresponds to row 1, second position corresponds to row 2, etc...). A value can be entered into each position that reflects the corresponding position <b>930</b> for each attribute. Thus, in the inverted index <b>929,</b> position 1 comprises the value "1" which is the corresponding position <b>930</b> value for the attribute "Soap", position 2 comprises the value "2" which is the corresponding position <b>930</b> value for the attribute "Soft Soap", position 3 comprises the value "3" which is the corresponding position <b>930</b> value for the attribute "Toothpaste", and position 4 comprises the value "3" which is the corresponding position <b>930</b> value for the attribute "Toothpaste".

0052By way of example, a BAI <b>932</b> can be generated as an index between the product attribute of Table 2 and Table 1. The BAI <b>932</b> can comprise a row for each unique attribute in the BTI <b>924</b> by order of corresponding position <b>926.</b> The value in each row can comprise the corresponding position <b>930</b> of the BTI <b>928.</b> Thus, position 1 of the BAI <b>932</b> corresponds to "Toothpaste" in the BTI <b>924</b> (corresponding position <b>926</b> of 1) and comprises the value "3" which is the corresponding position <b>930</b> for "Toothpaste" of the BTI <b>928.</b> Position 2 of the BAI <b>932</b> corresponds to "Soap" in the BTI <b>924</b> (corresponding position <b>926</b> of 2) and comprises the value "1" which is the corresponding position <b>930</b> for "Soap" of the BTI <b>928.</b> Position 3 of the BAI <b>932</b> corresponds to "Shampoo" in the BTI <b>924</b> (corresponding position <b>926</b> of 3) and comprises the value "-1" which indicates that the attribute "Shampoo" is not found in Table 1.

0053By way of example, a BAI <b>933</b> can be created to create an index between the product attribute of Table 1 and Table 2. The BAI <b>933</b> can comprise a row for each unique attribute in the BTI <b>928</b> by order of corresponding position <b>930.</b> The value in each row can comprise the corresponding position <b>926</b> of the BTI <b>924.</b> Thus, position 1 of the BAI <b>933</b> corresponds to "Soap" in the BTI <b>928</b> (corresponding position <b>930</b> of 1) and comprises the value "2" which is the corresponding position <b>926</b> for "Soap" of the BTI <b>924.</b> Position 2 of the BAI <b>933</b> corresponds to "Soft Soap" in the BTI <b>928</b> (corresponding position <b>930</b> of 2) and comprises the value "-1" which indicates that the attribute "Soft Soap" is not found in Table 2. Position 3 of the BAI <b>933</b> corresponds to "Toothpaste" in the BTI <b>928</b> (corresponding position <b>930</b> of 3) and comprises the value "1" which is the corresponding position <b>926</b> for "Toothpaste" of the BTI <b>924.</b>

0054<figref idref="f0009"><b>FIG. 9</b></figref> illustrates an example application of the logical inference engine <b>106</b> utilizing the BTI <b>920,</b> the BTI <b>924,</b> and the BTI <b>928.</b> A user can select the "Client" "Kalle" from within a user interface. A column for a user selection <b>934</b> of "Kalle" can be indicated in the BTI <b>920</b> comprising a value for each attribute that reflects the selection status of the attribute. Thus, the user selection <b>934</b> comprises a value of "0" for the attribute "Nisse" indicating that "Nisse" is not selected, the user selection <b>934</b> comprises a value of "0" for the attribute "Gullan" indicating that "Gullan" is not selected, the user selection <b>934</b> comprises a value of "1" for the attribute "Kalle" indicating that "Kalle" is selected, and the user selection <b>934</b> comprises a value of "0" for the attribute "Pekka" indicating that "Pekka" is not selected.

0055The BTI <b>920</b> can be consulted to determine that the attribute "Kalle" has a value of "1" in the column <b>923</b> corresponding to rows 3 and 4. In an aspect, the inverted index <b>921</b> can be consulted to determine that the user selection <b>934</b> relates to the position <b>922</b> value of "3" which is found in the inverted index <b>921</b> at positions 3 and 4, implicating rows 3 and 4 of Table 1. Following path <b>935,</b> a row state <b>936</b> can be generated to reflect the user selection <b>934</b> as applied to the rows of Table 2. The row state <b>936</b> can comprise a position that corresponds to each row and a value in each position reflecting whether a row was selected. Thus, position 1 of the row state <b>936</b> comprises the value "0" indicating that row 1 does not contain "Kalle", position 2 of the row state <b>936</b> comprises the value "0" indicating that row 2 does not contain "Kalle", position 3 of the row state <b>936</b> comprises the value "1" indicating that row 3 does contain "Kalle", position 4 of the row state <b>936</b> comprises the value "1" indicating that row 4 does contain "Kalle", position 5 of the row state <b>936</b> comprises the value "0" indicating that row 5 does not contain "Kalle", and position 6 of the row state <b>936</b> comprises the value "0" indicating that row 6 does not contain "Kalle".

0056Following path <b>937,</b> the row state <b>936</b> can be compared with the inverted index <b>925</b> to determine the corresponding position <b>926</b> contained in the inverted index <b>925</b> at positions 3 and 4. The inverted index <b>925</b> comprises the corresponding position <b>926</b> value of "2" in position 3 and the corresponding position <b>926</b> value of "3" in position 4. Following path <b>938,</b> the corresponding position <b>926</b> values of "2" and "3" can be determined to correspond to "Soap" and "Shampoo" respectively in the BTI <b>924.</b> Thus, the logical inference engine <b>106</b> can determine that both "Soap" and "Shampoo" in Table 2 are associated with "Kalle" in Table 2. The association can be reflected in an inferred state <b>939</b> in the BTI <b>924.</b> The inferred state <b>939</b> can comprise a column with a row for each attribute in the BTI <b>924.</b> The column can comprise a value indicated the selection state for each attribute. The inferred state <b>939</b> comprises a "0" for "Toothpaste" indicating that "Toothpaste" is not associated with "Kalle", the inferred state <b>939</b> comprises a "1" for "Soap" indicating that "Soap" is associated with "Kalle", and inferred state <b>939</b> comprises a "1" for "Shampoo" indicating that "Shampoo" is associated with "Kalle".

0057Following path <b>940,</b> the inferred state <b>939</b> can be compared to the BAI <b>932</b> to determine one or more associations between the selection of "Kalle" in Table 2 and one or more attributes in Table 1. As the inferred state <b>939</b> comprises a value of "1" in both position 2 and position 3, the BAI <b>932</b> can be assessed to determine the values contained in position 2 and position 3 of the BAI <b>932</b> (following path <b>941</b>). Position 2 of the BAI <b>932</b> comprises the value "1" which identifies the corresponding position <b>930</b> of "Soap" and position 3 of the BAI <b>932</b> comprises the value "-1" which indicates that Table 1 does not contain "Shampoo". Thus, the logical inference engine <b>106</b> can determine that "Soap" in Table 1 is associated with "Kalle" in Table 2. The association can be reflected in an inferred state <b>942</b> in the BTI <b>928.</b> The inferred state <b>942</b> can comprise a column with a row for each attribute in the BTI <b>928.</b> The column can comprise a value indicated the selection state for each attribute. The inferred state <b>942</b> comprises a "1" for "Soap" indicating that "Soap" is associated with "Kalle", the inferred state <b>942</b> comprises a "0" for "Soft Soap" indicating that "Soft Soap" is not associated with "Kalle", and the inferred state <b>942</b> comprises a "0" for "Toothpaste" indicating that "Toothpaste" is not associated with "Kalle". Based on the current state of BTIs and BAIs, if the data sources <b>102</b> indicate that an update or delta change has occurred to the underlying data, the BTIs and BAIs can be updated with corresponding changes to maintain consistency.

0058In aspects implementing indexlets, the logical inference engine <b>106</b> can apply query language by first performing intra-table inferencing on respective tables. Intra-table inferencing comprises transferring the imposed state of one field to other fields within the same table. In an aspect, shown in <figref idref="f0007"><b>FIG. 7</b></figref><b>,</b> intra-table inferencing can comprise computing the union of the index of the active attributes in a user input <b>704.</b> The intersection of the result of the union operation and record states (e.g., row states <b>710</b>) is then determined. This result is then intersected with the attribute states <b>714</b> of other columns using the inverted index <b>712.</b> If other selection vectors from a previously provided user input vector <b>704</b> has zero active entries, a conflict can be detected. In an aspect, the logical inference engine <b>106</b> can resolve the detected conflict. In an aspect, resolving a conflict can include deleting or otherwise eliminating one or more incompatible selections. In another aspect, resolving a conflict can include reverting the data model <b>301</b> or a portion of the data model <b>301,</b> e.g. a table, record, or attribute, to a previous state.

0059In an aspect, after performing intra-table inferencing, the logical inference engine <b>106</b> can perform inter-table inferencing based on the intra-table inferencing output of a plurality of tables, as is depicted in <figref idref="f0006"><b>FIG. 6</b></figref><b>.</b> In an aspect, intra-table inferencing can include transferring a common field attribute of one table <b>669</b> to a child in its branch. In an aspect, this can be performed by running the attribute states <b>670</b> output from intra-table inferencing through an attribute-to-attribute (A2A) index <b>672</b> referencing the attribute states <b>674</b> in a second table <b>676.</b> In an aspect, the A2A index <b>672</b> can be partitioned into one or more indexlets as described herein with respect to other data tables. In another aspect, transferring a common field attribute of one table <b>669</b> to a child in its branch by running the attribute states <b>670</b> output from intra-table inferencing through a function or logic performing similar functionality as the A2A index <b>672.</b> For example, a function, service, or other logic can accept as input a pair of symbols and return an indication of whether or not they are related, e.g. TRUE or FALSE. In another aspect, attribute-to-attribute relations can be indicated by user input.

0060As described previously with regard to <figref idref="f0001"><b>FIG. 1</b></figref><b>,</b> the chart engine <b>108</b> can receive dimensions, expressions, and sorting parameters and can compute a hypercube data structure containing aggregations along the dimensions. Implementation of the indexlet data structure, via the logical inference engine <b>106</b> as described above, enables the chart engine <b>108</b> to use indexlets to build hypercube domains. For example, using bidirectional indexing, walking the database for traversals can be performed as shown in <figref idref="f0011"><b>FIG. 11</b></figref><b>.</b> A recursion algorithm starts with a consistently computed state. A "forest" covering the data model can be constructed and the "roots" of the "trees" can be either the largest table in the tree or one table containing the first needed field. Starting from that point, the database can be (naively) traversed as follows. Initially, virtual record fields present in the current table can be checked and saved in a list of local fields. An active attribute <i>a_i</i> can be taken from the input field (e.g., iteratively). Using the A2R index, the record set {<i>r_k</i>}<i>_k</i> associated to that symbol can be identified. The chart engine <b>108</b> can assign for every <i>r_k,</i> the virtual record entries present in the local field list. If the virtual record is totally assigned, the virtual record can be processed. The chart engine <b>108</b> can go to next symbol. Using R2S on the output field, the chart engine <b>108</b> can find corresponding attribute <i>b_j.</i> Using the A2A index, the chart engine <b>108</b> can translate <i>b_j</i> to the next-in-branch table <maths id="math0001"><math display="inline"><mover accent="true"><msub><mi>b</mi><mi>J</mi></msub><mo>˜</mo></mover></math><img file="EP3678032B1_D0001.tif" /></maths>. The chart engine <b>108</b> can then perform this same process with <maths id="math0002"><math display="inline"><mover accent="true"><msub><mi>b</mi><mi>J</mi></msub><mo>˜</mo></mover></math><img file="EP3678032B1_D0002.tif" /></maths> as input. While the preceding operations are on the global symbol maps, the impact on respective indexlets is included in the scheduling planning. Parallelisation of hypercube traversal is also possible for common aggregation linear operators, as described below with regard to <figref idref="f0012"><b>FIG. 12</b></figref><b>.</b>

0061A system architecture <b>1200</b> for scalable index generation and chart generation is described with regard to <figref idref="f0012"><b>FIG. 12</b></figref><b>.</b> The system architecture <b>1200</b> is configured to receive one or more global queries, divide the global query into local components, and associate the local components with respective indexlets to generate partial results according to a global plan that maintains an association between partial results that are related to the same query. The system architecture <b>1200</b> can be configured to receive and parse queries (e.g., a batch of queries). The parsed queries can be analyzed to obtain common subexpressions and data model subpaths for efficient processing. Based on the analysis, the system architecture <b>1200</b> can determine a logical plan. The logical plan can define how to traverse the data model to execute the queries and obtain all needed information to compute the queries. The system architecture <b>1200</b> can determine a physical plan. The physical plan can define the various worker entities (e.g., indexlet worker, hypercube workers, and the like) along with locality data indicative of where workers and/or underlying data are physically stored (e.g., which server, which facility, etc...). The physical plan can be determined based on the logical plan.

0062The system architecture <b>1200</b> maintains a plan for executing the global query across a logically and physically distributed workflow.

0063As described previously, with regard to <figref idref="f0001"><b>FIG. 1</b></figref><b>,</b> the logical inference engine <b>106</b> can generate one or more indexes for the data model. The one or more indexes can be loaded into memory in lieu of the data model <b>301</b> shown in <figref idref="f0003"><b>FIG. 3</b></figref><b>.</b> In some instances, the amount of data contained with the data sources <b>102</b> can become so large that relying on the logical inference engine <b>106</b> to generate indexes for a large data source can be time and computational prohibitive. Moreover, some data sources are subject to frequent, and often large, additions and/or updates to the data sources (e.g., appends). Frequent appends results in a need to generate new indexlets and, in some instances, recreate existing indexlets. Concurrently with receipt of appends, there is a need for users to query and/or otherwise analyze both the existing data and the appends. Existing systems are not capable of providing a satisfactory level of service to users desirous of analyzing large data sets with frequent appends for at least the reason that indexlet generation can be computationally and time intensive for the logical inference engine <b>106.</b> Accordingly, the system architecture <b>1200</b> can comprise an indexing coordinator <b>1201</b> configured to implement an indexation process (e.g., symbol indexation) and coordinate indexlet generation across multiple indexlet indexers <b>1202.</b> An indexlet indexer <b>1202</b> can comprise a computing device and/or can comprise a virtual machine operating on a computing device. As an example, the indexlet indexers <b>1202</b> can comprise a distributed cluster of computing devices and/or a cluster of virtual machines operating on one or more computing devices.

0064The indexlet indexers <b>1202</b> can access the data source(s) <b>102</b> to generate one or more indexlets. The data source(s) <b>102</b> can comprise one or more databases stored on one or more computing devices. The indexlets thus generated can serve as a foundation for providing bi-directional indexing information for the both inferencing and/or hypercube domain calculation techniques. As new data is added (e.g., appends) to the data sources <b>102,</b> one or more associated indexing tasks can be assigned to the indexing coordinator <b>1201.</b> The indexing coordinator <b>1201</b> can cause one or more indexlet indexers <b>1202</b> to generate indexlets for the newly appended data. The indexing coordinator <b>1201</b> can distribute indexing tasks amongst one or more indexlet indexers <b>1202.</b> The indexing coordinator <b>1201</b> can determine if any indexlet indexers <b>1202</b> are available to generate indexlets for the newly appended data. If no indexlet indexer <b>1202</b> is available, the indexing coordinator <b>1201</b> can spool up an additional indexlet indexer <b>1202</b> (e.g., bring an additional computing device online that was not previously performing an indexing function). If any indexlet indexers <b>1202</b> are available, the indexing coordinator <b>1201</b> can assign an indexlet generation task to the available indexlet indexer <b>1202.</b> Through the indexing coordinator <b>1201,</b> the indexlet indexers <b>1202,</b> perform an indexation process that can be divided into two components: an indexer service and a global symbol service. The indexing coordinator <b>1201</b> handles indexation requests as well as distributing tasks of creating the intermediate indexlet structures amongst the indexlet indexers <b>1202.</b> The indexing coordinator <b>1201</b> implements the global symbol service by splitting global symbol maps across the indexlet indexers <b>1202.</b> Even in good hash map implementations there is always overhead in memory consumption due to the management of the internal data structure. As a result, the ability to split global symbol maps across the indexlet indexers <b>1202</b> helps to share the load as well as supporting both horizontal and vertical scaling when dealing with large data sets and appends.

0065Given an input data source <b>102</b> in an interpretable format, e.g., CSV, the indexlet indexers <b>1202</b> can partition the data source <b>102</b> into disjoint, same-sized blocks of rows. In some aspects, the indexation process will not partition the last row (e.g., the size of the last block might be smaller than the size of the other blocks). These "slices" of the data can be then processed independently to generate intermediate indexlet structures. Intermediate indexlet structures can be processed sequentially to generate a global symbol map. In addition to bi-directional information (symbol to row and row to symbol), a mapping between the symbols can reside locally in the indexlet and in the global symbol map. This mapping enables a simple yet fast and efficient transformation between symbols in an indexlet and in global symbol maps and vice versa through select and rank operations on bit vectors.

0066There are two main challenges to the indexation process: parallelization of the creation of intermediate indexlet structures and the creation and handling of large global symbol maps that contain potentially billions of symbols.

0067In order to achieve the maximum parallelization of the creation of intermediate indexlet structures, the indexer service can utilize a distributed computing environment. <figref idref="f0013"><b>FIG. 13</b></figref> illustrates an indexation service <b>1300.</b> The indexing coordinator <b>1201</b> can comprise information regarding the capability of indexlet indexers <b>1202a, 1202b,</b> and <b>1202c</b> registered during their initialization. On receiving an indexation request, the indexing coordinator <b>1201</b> distributes tasks to indexlet indexers <b>1202a, 1202b,</b> and <b>1202c</b> based on the registered capability. In this setup, more indexlet indexers can be dynamically added and registered with the indexing coordinator <b>1201</b> to reduce the required creation time of the intermediate indexlet structures <b>1302a, 1302b, 1302c, 1302d, 1302e,</b> and <b>1302f.</b> Moreover, if an indexlet indexer dies (e.g., malfunctions or otherwise degrades in operational capability) during the process, a new indexlet indexer can be instantiated and registered to the indexing coordinator <b>1201</b> to take over the corresponding tasks.

0068When dealing with large data sets, global symbol maps can comprise billions of symbols. Naturally, an in-memory hash map can provide better performance on both look up and insert operations in comparison to file-based hash map implementations. Unfortunately, it is not practical to have an unlimited amount of physical memory available. Although virtual memory can help to elevate the limitation of physical memory, the performance of look up and insert operations degrades dramatically.

0069A global symbol service is provided in which global symbol maps are split across machines to share the load as well as the stress on memory requirements while achieving the desired performance.

0070<figref idref="f0014"><b>FIG. 14</b></figref> illustrates a global symbol service <b>1400.</b> During an initialization process, indexlet indexers <b>1202a, 1202b,</b> and <b>1202c</b> register their capabilities, e.g., amount of memory, processing power, bandwidth, etc, with the indexing coordinator <b>1201.</b> On receiving an indexation request, the indexing coordinator <b>1201</b> can request initialization of global symbol maps <b>1402a, 1402b,</b> and <b>1402c</b> on indexlet indexers <b>1202a, 1202b,</b> and <b>1202c</b> based on the registered capabilities. As a result, the global symbol maps <b>1402a, 1402b,</b> and <b>1402c</b> are initialized and proper capability reserved accordingly.

0071The indexer service and the global symbol service can generate intermediate indexlet structures and process the intermediate indexlet structures sequentially to generate the global symbol maps together with bi-directional indexing information. This constraint on processing order permits fast and efficient mappings between symbols that reside locally in an indexlet and the global symbol maps. The global symbol service allows parallelism to improve indexation performance.

0072For example, a state, S, can be introduced into the global symbol maps <b>1402a, 1402b,</b> and <b>1402c</b> on the indexlet indexers <b>1202a, 1202b,</b> and <b>1202c</b> as follows <maths id="math0003"><math display="block"><mi mathvariant="normal">S</mi><mo>=</mo><mfenced open="{" close="}"><mi>standing_by</mi><mo>,</mo><mspace width="1ex" /><mi>serving</mi><mo>,</mo><mspace width="1ex" /><mi>closed</mi></mfenced></math><img file="EP3678032B1_D0003.tif" /></maths> where "standing_by" indicates that the global symbol map on the indexlet indexer <b>1202</b> is not in use, "serving" indicates that the global symbol map on the indexlet indexer <b>1202</b> can be used for both look up and insert operations, "closed" indicates that the global symbol map on the indexlet indexer <b>1202</b> is full, and, thus, only supports a look up operation.

0073The creation of the global symbol map can start with inserting symbols into a serving hash map on the corresponding indexlet indexer <b>1202.</b> When the optimal capacity of the hash map is reached, the corresponding indexlet indexer <b>1202</b> informs the indexing coordinator <b>1201</b> and changes its state to closed. The indexing coordinator <b>1201</b> can then request another indexlet indexer <b>1202</b> to handle the upcoming tasks, e.g., changing the state of hash map from "standing_by" to "serving." On subsequent processes, look up operations can be carried out in a bulk and in a parallelized manner on a closed hash map to maximize the performance. The remaining new symbols can then be inserted into the serving hash map on the corresponding indexlet indexer <b>1202.</b> If a indexlet indexer <b>1202</b> in "standing_by" state dies during the process, it can be replaced by instantiating another indexlet indexer <b>1202</b> that registers itself to the indexing coordinator <b>1201.</b> If a indexlet indexer <b>1202</b> in "closed" or "serving" state dies, it can be replaced by either another indexlet indexer <b>1202</b> in "standing_by" state or a newly instantiated indexlet indexer <b>1202.</b> In this case, the indexing coordinator <b>1201</b> can cause the range of corresponding data to be indexed again to reconstruct the corresponding hash map.

0074In an aspect, a Bloom filter <b>1404a, 1404b,</b> and <b>1404c</b> can be used to further optimize look up performance. A Bloom filter is a probabilistic data structure that can indicate whether an element either definitely is not in the set or may be in the set. In other words, false positive matches are possible, but false negatives are not. The base data structure of a Bloom filter is a bit vector. On a very large hash map that contains several billion symbols, the performance of the look up operation can degrade dramatically as the size increases. The Bloom filter is a compact data structure that can represent a set with an arbitrarily large number of elements. The Bloom filter enables fast querying of the existence of an element in a set. Depending on the registered resource information, the false positive rate can be specified to achieve both the compactness of the Bloom filter and the minimum access to the hash map. A Bloom filter can improve the performance of look up operation on closed hash map by 3 to 5 times. The constructed Bloom filter <b>1404a, 1404b,</b> and <b>1404c</b> can be used to minimize the amount of communication required in the inferencing as well as hypercube domain construction process. Particularly, by performing look up operations in the Bloom filters <b>1404a, 1404b,</b> and <b>1404c</b> first, the number of hash maps that possibly contain the desired information will be minimized, and, thus, reduce the number of requests that need to be transferred through the network.

0075The indexer service and the global symbol service allows both local as well as cloud-based (e.g., distributed) deployment of symbol indexation. With cloud-based deployment, more resources can be added to improve the indexation process. The indexation process is bounded by the amount of resources and the available bandwidth.

0076Returning to <figref idref="f0012"><b>FIG. 12</b></figref><b>,</b> indexlets thus generated by the indexlet indexers <b>1202</b> can be stored in an indexlet repository <b>1203.</b> An indexlet generated by an indexlet indexer can be stored locally on the machine where the indexlet indexer is resident. The indexlets can be stored as files. The indexlet repository <b>1203</b> can comprise a database of indexlets and/or an aggregation of indexlets, accessible by, for example, file name. Some or all of the indexlet repository <b>1203</b> can be distributed across multiple computing devices. Accordingly, some indexlets may be stored in a physical device separate from physical devices storing other indexlets. A persistence manager can be configured to track and/or maintain a record of where each indexlet is physically stored. The persistence manager can be configured to provide workers with sufficient location information to enable the workers to find indexlets needed for worker processing, locally and/or globally. In some aspects, the indexlet files can be co-located with the data source <b>102</b> from which the indexlet was derived. In some aspects, the indexlet files can be co-located with the datalet that corresponds to the indexlet. In an aspect, an indexlet and corresponding datalet can be stored together in a database.

0077Once stored in the indexlet repository <b>1203,</b> the indexlets can be used to process one or more queries. A query director <b>1204</b> can be configured to receive one or more queries. In an aspect, multiple query directors <b>1204</b> can be configured to receive the one or more queries. As described previously with regard to <figref idref="f0001"><b>FIG. 1</b></figref><b>,</b> a query can take the form of a user selection in an underlying data set, causing the logical inference engine <b>106</b> to evaluate a number of filters on the data set. However, the query director <b>1204</b> can also receive one or more batches of queries (e.g., multi-queries). The query director <b>1204</b> can optimize the execution of the one or more batches of queries. The query director <b>1204</b> can determine common operations/subexpressions across the batch of queries and coordinate global execution of the batch queries. Multi-query optimization may identify common subexpressions within a batch of queries that may be optimized together and generates physical plans that share the common computation.

0078In an aspect, the query director <b>1204</b> may enable work sharing at run-time. For example, using an operator-centric approach, where each operator, e.g., hash join, sort, or sequential scan, may act as a mini query engine that services several concurrent queries. This design may enable an operator to examine holistically the requests of concurrent queries and identify on-the-fly any operator-specific computation that may be shared. A hash join operator, for instance, may share the build phase of a relation that participates in different hash-joins in several queries. The methods and system disclosed may also rely on work sharing in order to scale to many more concurrent queries. In an aspect, the query director <b>1204</b> may use a shared scan operator to share disk I/O among multiple scan operations executing concurrently on multiple processors. The in-memory data and the state of other operators may not be shared. Concurrent queries can be schedules so as to coordinate tuple accesses in the processor's cache. This approach is useful with main-memory databases. In an aspect, the query director <b>1204</b> may use work sharing in a streaming database systems. By sharing work (or a state) among continuous-query operators, the query director <b>1204</b> may maintain a low per-tuple processing cost and thus may handle a large number of continuous queries over fast streams.

0079Under control of the query director <b>1204</b> (or multiple query directors <b>1204</b>), the query director <b>1204,</b> the one or more indexlet workers <b>1205,</b> and the one or more hypercube workers <b>1206</b> can be configured to coordinate hybercube generation across multiple indexlet workers <b>1205</b> and hypercube workers <b>1206.</b> In effect, one or more indexlet workers <b>1205</b> and one or more hypercube workers <b>1206</b> can be configured as the chart engine <b>108</b> described previously.

0080An indexlet worker <b>1205</b> can comprise a computing device and/or can comprise a virtual machine operating on a computing device. As an example, the indexlet workers <b>1205</b> can comprise a distributed cluster of computing devices and/or a cluster of virtual machines operating on one or more computing devices. A hypercube worker <b>1206</b> can comprise a computing device and/or can comprise a virtual machine operating on a computing device. As an example, the hypercube workers <b>1206</b> can comprise a distributed cluster of computing devices and/or a cluster of virtual machines operating on one or more computing devices.

0081The indexlet workers <b>1205</b> can access the indexlet repository <b>1203</b> to generate inferences based on their local indexlets, fetch data from associated files or traverse hypercubes (e.g., generate virtual records).

0082The hypercube workers <b>1206</b> can receive indexlets and/or virtual records from the indexlet workers <b>1205</b> and use the received indexlets and/or virtual records to produce calculations (which can be stored in a distributed fashion). For example, (OrderDate, sum(Sales)) needs virtual records of the type (OrderDate, Sales). The indexlet workers <b>1205</b> can generate the virtual records (OrderDate, Sales), transmit to the virtual records to the hypercube workers <b>1206</b> which can then perform the aggregation of the sum(Sales).

0083As new data is added (e.g., appends) to the data sources <b>102,</b> additional indexlets will be generated and deposited into the indexlet repository <b>1203,</b> enabling regeneration of inferences and recalculation of hypercubes to incorporate newly added data. The indexlet workers <b>1205</b> and the hypercube workers <b>1206</b> can utilize a cache <b>1207</b> to store local results and/or branch results locally on the machine where a respective indexlet worker <b>1205</b> and hypercube worker <b>1206</b> are resident.

0084The query director <b>1204</b> can determine if any indexlet workers <b>1205</b> or any hypercube workers <b>1206</b> are available to generate hybercubes for the newly appended data. If no indexlet workers <b>1205</b> or hypercube workers <b>1206</b> are available, the query director <b>1204</b> can spool up an indexlet worker <b>1205</b> and/or a hypercube worker <b>1206</b> (e.g., bring an additional computing device online that was not previously performing an indexing function). If any indexlet workers <b>1205</b> or hypercube workers <b>1206</b> are available, the query director <b>1204</b> can assign an indexlet generation task to the available indexlet workers <b>1205</b> or hypercube workers <b>1206.</b>

0085As described previously, parallelisation of hypercube traversal is also possible for common aggregation linear operators. The hypercube domain can be paged and the traversal can be stopped at any point by saving the state defined in the virtual record, and the tables being traversed at that point.

0086Expressions are strings representing which operations to be done on fields from the database. So, consider the field set <i>S</i> = {<i>S₀, S₁, S₂,</i> ..., <i>S_n</i>} of the database. This is the union of all possible attributes per column, and those associated are included within the same set under a unique symbol. The part of an expression that is the subset of <i>S̃</i> ⊂ <i>S</i> can be used for computing the values. Furthermore, the property of globality of the expression, e.g., the expression requiring all records to be available, can be taken into account when evaluating the expression.

0087Provided herein is a hypercube taxonomy. Two attributes <i>x</i>_i ∈ <i>S_i</i>, <i>x_j</i> ∈ <i>S_j</i> can be defined as associated if the inference engine <b>106</b> can create a connected path between the attributes. Such can be denoted as <i>x_i</i> ~ <i>x_j.</i> Consider the space of virtual records on <i>Ŝ</i> ⊂ <i>S</i> as <maths id="math0004"><math display="block"><msubsup><mi>R</mi><mi>υ</mi><mover accent="true"><mi>S</mi><mo>^</mo></mover></msubsup><mo>=</mo><mfenced open="{" close="}"><mi>r</mi><mo>=</mo><mfenced><msub><mi>x</mi><mn>1</mn></msub><mo>,</mo><msub><mi>x</mi><mn>2</mn></msub><mi>,...</mi><mo>,</mo><msub><mi>x</mi><mi>k</mi></msub></mfenced><mrow><mo>|</mo><mrow><msub><mi>x</mi><mi>i</mi></msub><mo>∈</mo><msub><mi>S</mi><mi>i</mi></msub><mo>∧</mo><msub><mi>x</mi><mi>i</mi></msub><mo>∼</mo><msub><mi>x</mi><mi>j</mi></msub><mo>∀</mo><mi>i</mi><mo>,</mo><mi>j</mi></mrow></mrow></mfenced></math><img file="EP3678032B1_D0004.tif" /></maths>

0088Subsequently, the dimension set can be defined as <maths id="math0005"><math display="block"><mi>Dim</mi><mo>=</mo><msubsup><mi>R</mi><mi>υ</mi><mover accent="true"><mi>S</mi><mo>^</mo></mover></msubsup></math><img file="EP3678032B1_D0005.tif" /></maths> where <i>Ŝ</i> is the set of dimensions defined to compute the hypercube. This set is naturally provided with an enumeration of all its elements where <i>i</i> is mapped to a unique element, denoted by Dim<i>ⁱ</i>. Furthermore, provided some expression and its corresponding computation set <i>s̃</i>, the hypercube domain can be defined as <maths id="math0006"><math display="block"><mi>X</mi><mo>=</mo><msubsup><mi>R</mi><mi>υ</mi><mfenced><mover accent="true"><mi>S</mi><mo>^</mo></mover><mover accent="true"><mi>S</mi><mo>˜</mo></mover></mfenced></msubsup></math><img file="EP3678032B1_D0006.tif" /></maths>

0089X has a natural partition <maths id="math0007"><math display="block"><mrow><mo>=</mo><mstyle displaystyle="true"><munder><mo>∪</mo><mi>i</mi></munder><mrow><msub><mi>i</mi></msub><mo>,</mo><mtext></mtext><msub><mi>i</mi></msub><mo>=</mo><mrow><mo>{</mo><mrow><mi>r</mi><mo>∈</mo><mi>X</mi><mrow><mo>|</mo><mrow><msub><mi>x</mi><mi>j</mi></msub><mo>=</mo><msup><mrow><mtext>Dim</mtext></mrow><mi>i</mi></msup><mrow><mo>(</mo><mi>j</mi><mo>)</mo></mrow><mo>,</mo><mi>j</mi><mo>=</mo><mn>0,...,</mn><mi>k</mi></mrow></mrow></mrow><mo>}</mo></mrow></mrow></mstyle></mrow></math><img file="EP3678032B1_D0007.tif" /></maths>

0090The hypercube problem can then be formulated as follows. Consider Dim and X together with <i>I</i> : <img file="EP3678032B1_D0008.tif" /> → <img file="EP3678032B1_D0009.tif" />, <i>F :</i><img file="EP3678032B1_D0010.tif" /> → <img file="EP3678032B1_D0011.tif" /> with <i>p</i> = <i>#</i>(<i>X</i>)<i>.</i> the problem becomes computing <maths id="math0008"><math display="block"><mfenced><mi>F</mi><mo>∘</mo><mi>I</mi></mfenced><mfenced /><mo>=</mo><mstyle displaystyle="true"><munder><mo>∑</mo><mi>i</mi></munder><mrow><mfenced><msub><mi>ƒ</mi><mi>i</mi></msub><mo>∘</mo><mi>I</mi></mfenced><mfenced /><msub><mi mathvariant="normal">e</mi><mi>i</mi></msub></mrow></mstyle></math><img file="EP3678032B1_D0012.tif" /></maths><ul id="ul0001" list-style="none" compact="compact"><li>with <i>e_i</i> the canonical vector with just 1 in <i>i</i>-th position, from the data model as quickly as possible. The hypercube problems can now be classified according to the characteristics of the expressions and the record domains. Some cases are inherently dependent of each other while others are parallelisable. Classifications include, for example,</li><li>a component independent operator: <maths id="math0009"><math display="block"><mfenced><mi>F</mi><mo>∘</mo><mi>I</mi></mfenced><mfenced><mi>X</mi></mfenced><mo>=</mo><mstyle displaystyle="true"><munder><mo>∑</mo><mi>i</mi></munder><mrow><mfenced><mi>ƒ</mi><mo>∘</mo><mi>I</mi></mfenced><mfenced><msub><mi>X</mi><mi>i</mi></msub></mfenced><msub><mi mathvariant="normal">e</mi><mi>i</mi></msub></mrow></mstyle><mo>;</mo></math><img file="EP3678032B1_D0013.tif" /></maths><ul id="ul0002" list-style="none" compact="compact"><li>a summable operator: <maths id="math0010"><math display="block"><mfenced><mi>F</mi><mo>∘</mo><mi>I</mi></mfenced><mfenced><mi>X</mi></mfenced><mo>=</mo><mstyle displaystyle="true"><munder><mo>∑</mo><mi>i</mi></munder><mstyle displaystyle="true"><munder><mo>∑</mo><mrow><mi>x</mi><mo>∈</mo><msub><mi>X</mi><mi>i</mi></msub></mrow></munder><mrow><mfenced><mi>f</mi><mo>∘</mo><mi>I</mi></mfenced><mfenced><mfenced open="{" close="}"><mi>x</mi></mfenced></mfenced><msub><mi mathvariant="normal">e</mi><mi>i</mi></msub><mo>;</mo></mrow></mstyle></mstyle></math><img file="EP3678032B1_D0014.tif" /></maths> and</li><li>a recursive operator: <maths id="math0011"><math display="block"><mfenced><msub><mi>ƒ</mi><mrow><mi>i</mi><mo>+</mo><mn>1</mn></mrow></msub><mo>∘</mo><mi>I</mi></mfenced><mfenced><mi>X</mi></mfenced><mo>=</mo><mi>g</mi><mfenced><msub><mi>X</mi><mrow><mi>i</mi><mo>+</mo><mn>1</mn></mrow></msub><mstyle displaystyle="true"><munderover><mo>∑</mo><mrow><mi>j</mi><mo>=</mo><mn>0</mn></mrow><mi>i</mi></munderover><mrow><mfenced><msub><mi>ƒ</mi><mi>j</mi></msub><mo>∘</mo><mi>I</mi></mfenced><mfenced><msub><mi>X</mi><mi>j</mi></msub></mfenced></mrow></mstyle></mfenced><mo>.</mo></math><img file="EP3678032B1_D0015.tif" /></maths></li></ul></li></ul>

0091This algebra works with strings, if string concatenation is regarded as sum.

0092The objective can be to minimize I/O by grouping results by indexlet, and distributing the work accordingly. The workload can be distributed considering the partition <i>X_i</i>. Without loss of generality, the component independent operator approach is disclosed and along with related load distribution.

0093In an aspect, the chart engine <b>108</b> can utilize indexlets for parallelization (e.g., distributing computations). The chart engine <b>108</b> can determine one or more aggregation functions (S) and one or more internal functions (e.g., scalar functions) using indexlets. In an example where an aggregation function is to be determined with respect to a single table (e.g., when fields in S belong to the same table) is disclosed. For the partition {<i>X_i</i>}, the disclosed methods can determine a subset <i>X̂</i> ⊂ {<i>X_i</i>} that belongs to an indexlet Idx<i>_k</i>. A query can be made with the respective partition in parallel for the set of all indexlets {Idx<i>_k</i>}<i>_k</i>. Results can be aggregated at a top level when all queries are returned. The primary computational overhead comes from data transmission, in this case of all partitions, as the query is executed for every indexlet and results returned.

0094As shown in <figref idref="f0015"><b>FIG. 15</b></figref><b>,</b> the chart engine <b>108</b> can perform the following example method in the case of <i>S̃</i> being contained within a single table. The chart engine <b>108</b> can determine Dim. The chart engine <b>108</b> can annotate which indexlets are needed per Dim<i>ⁱ</i>. The chart engine <b>108</b> can regroup Dim<i>ⁱ</i> per indexlet Idx<i>_k</i>. The chart engine <b>108</b> assigns to every indexlet a subset Dim<i>_k</i> ⊂ Dim. The chart engine <b>108</b> can take <i>f̂</i> as <i>f</i> if the function is summable and domain independent, or its approximation <i>f̃</i> if not. The chart engine <b>108</b> can loop in <i>k</i> : query (in parallel) every indexlet Idx<i>_k</i> with <i>f̂</i> ∘ <i>I</i> partitioned by Dim<i>_k</i>. For example, by computing: <maths id="math0012"><math display="block"><msub><mi>F</mi><mi>k</mi></msub><mo>=</mo><mstyle displaystyle="true"><munder><mo>∑</mo><mi>i</mi></munder><mrow><mstyle displaystyle="true"><munder><mo>∑</mo><mrow><mi>x</mi><mo>∈</mo><msub><mi>X</mi><mi>i</mi></msub></mrow></munder><mrow><mfenced><mi>ƒ</mi><mo>∘</mo><mi>I</mi></mfenced><mfenced><mfenced open="{" close="}"><mi>x</mi></mfenced></mfenced><msub><mi>e</mi><mrow><mi>i</mi><mo>,</mo></mrow></msub></mrow></mstyle><mspace width="1ex" /><msub><mi>X</mi><mi>i</mi></msub><mo>∈</mo><msub><mi>Dim</mi><mi>k</mi></msub></mrow></mstyle></math><img file="EP3678032B1_D0016.tif" /></maths>

0095The chart engine <b>108</b> can aggregate at top level as a last pass. For example, by computing: <maths id="math0013"><math display="block"><mfenced><mi>F</mi><mo>∘</mo><mi>I</mi></mfenced><mfenced><mi>X</mi></mfenced><mo>=</mo><mstyle displaystyle="true"><munder><mo>∑</mo><mi>k</mi></munder><msub><mi>F</mi><mi>k</mi></msub></mstyle></math><img file="EP3678032B1_D0017.tif" /></maths>

0096Thus, the chart engine <b>108</b> can introduce a clear parallel pattern by every subcomputation on indexlets.

0097As shown in <figref idref="f0016"><b>FIG. 16</b></figref><b>,</b> the chart engine <b>108</b> can perform the following example method in the case of <i>S̃</i> being contained within multiple tables. In the multiple table case, the computational overhead is found in constructing virtual records in parallel. The chart engine <b>108</b> can determine Dim. The chart engine <b>108</b> can annotate which indexlets in all tables are needed per Dim<i>^i,j</i>. The chart engine <b>108</b> can regroup Dim<i>ⁱ</i> per indexlet <maths id="math0014"><math display="inline"><msubsup><mi>Idx</mi><mi>k</mi><mi>j</mi></msubsup></math><img file="EP3678032B1_D0018.tif" /></maths> and table T_j<i>.</i> The chart engine <b>108</b> thus assigns a subset Dim<i>_k</i> ⊂ Dim to every indexlet in every table. The chart engine <b>108</b> can identify the table with the largest number of indexlets as a distributed master table. The chart engine <b>108</b> can pass the expression to compute as tasks to all the indexlets of the master table, and all other tables involved prepare the domain pieces imposed by the partitions <maths id="math0015"><math display="inline"><msubsup><mi>Dim</mi><mi>k</mi><mi>j</mi></msubsup></math><img file="EP3678032B1_D0019.tif" /></maths>. The chart engine <b>108</b> can utilize the master table indexlets to query the other indexlets in the closest neighbors indexlets. If the closest neighbors indexlets need further pieces to complete their domains, they ask downstream to their respective dependencies. Every indexlet of the master table then completes the computation of F<i>_k</i>. The chart engine <b>108</b> can aggregate at top level as a last pass. For example, by computing: <maths id="math0016"><math display="block"><mfenced><mi>F</mi><mo>∘</mo><mi>I</mi></mfenced><mfenced><mi>X</mi></mfenced><mo>=</mo><mstyle displaystyle="true"><munder><mo>∑</mo><mi>k</mi></munder><msub><mi mathvariant="normal">F</mi><mi>k</mi></msub></mstyle></math><img file="EP3678032B1_D0020.tif" /></maths>

0098The chart engine <b>108</b> thus distributes a hypercube domain across nodes by using the largest table to compute the domain and maximize I/O.

0099In the event of a data update the underlying data source/data set, a new indexlet can be generated as described above and the new indexlet incorporated into the methods for evaluating an expression (single table or multiple data). A data update can include a modification to existing data and/or appending new data to existing data (e.g., adding data to an existing table or adding a new table of data). Thus, the methods for evaluating an expression shown in <figref idref="f0015"><b>FIG. 15</b></figref> and <figref idref="f0016"><b>FIG. 16</b></figref> enables the system to use the new indexlet (and any existing indexlets as needed) to evaluate an expression, thus avoiding the need to recreate the hypercube/multidimensional cube.

0100In an aspect, the chart engine <b>108</b> can distribute hypercube domain computation. For example, by finding the partitions of each indexlet per dimensions prescribed. The basic case is given when considering two adjacent tables are associated. The chart engine <b>108</b> can utilize an indexlet centric technique instead of a symbol centric technique. The symbol centric traversal is done intra-indexlet. The indexlet centric technique is depicted in <figref idref="f0017"><b>FIG. 17</b></figref><b>.</b>

0101The mapping shown in <figref idref="f0017"><b>FIG. 17</b></figref> is from indexlet to indexlet in order to establish partitions from symbols to other symbols. This partition can be pre-computed and stored when constructing the A2A map. During the construction of the A2A map, the construction of the I2I map can be performed. The map can be constructed as <i>121 :</i> N → N where <i>k</i> is associated to <i>j</i> if the indexlet <maths id="math0017"><math display="inline"><msubsup><mi>Idx</mi><mi>a</mi><mi>k</mi></msubsup></math><img file="EP3678032B1_D0021.tif" /></maths> of table <i>a</i> is associated to indexlet <maths id="math0018"><math display="inline"><msubsup><mi>Idx</mi><mi>b</mi><mi>j</mi></msubsup></math><img file="EP3678032B1_D0022.tif" /></maths> of table <i>b.</i> Such a coarse mapping can be used then to label the records of indexlet <maths id="math0019"><math display="inline"><msubsup><mi>Idx</mi><mi>a</mi><mi>k</mi></msubsup></math><img file="EP3678032B1_D0023.tif" /></maths> by the records of <maths id="math0020"><math display="inline"><msubsup><mi>Idx</mi><mi>b</mi><mi>j</mi></msubsup></math><img file="EP3678032B1_D0024.tif" /></maths> or the symbols of the common dimension of A2A. This can be computed and stored at indexing time for A2A.

0102Labeling and/or coloring can be distributed per indexlet, collapsing in the resulting indexlets, and then retransmitted to the next table using the I2I map. By way of example, the chart engine <b>108</b> can label all active symbols of the dimensions involved and can select the largest table (as measured by number of indexlets) with a column in <i>Ŝ</i> as a root table. The chart engine <b>108</b> can transport the labels using the I2I map to the root table from all tables as shown in <figref idref="f0017"><b>FIG. 17</b></figref><b>.</b> The result is that the different combinations of labels are then conforming every Dim<i>_k</i>. In the case of multiple tables, the chart engine <b>108</b> can further propagate the domains to spoke tables, considering the root table as a hub.

0103Returning to <figref idref="f0012"><b>FIG. 12</b></figref><b>,</b> computed hypercube data structures containing aggregations along the dimensions of the one or more queries provided to the query director <b>1204</b> can be provided to a client <b>1208</b> for use by the associative data indexing engine <b>100,</b> as previously described. As described with regard to <figref idref="f0001"><b>FIG. 1</b></figref><b>,</b> results of any calculations can be passed to the rendering engine <b>116.</b> A user client <b>1209</b> can be configured as a client application installed on a user computing device to access the associative data indexing engine <b>100.</b> The IDX worker <b>1205a</b> and the hypercube worker <b>1206a</b> that are connected to the client <b>1208</b> can be under the control of the query director <b>1204</b> through the client <b>1208</b> as a proxy, as described previously. However, the IDX worker <b>1205a</b> and the hypercube worker <b>1206a</b> that are connected to the client <b>1208</b> can be located on a computing device separate from the other IDX workers <b>1205</b> and the hypercube workers <b>1206</b> and can be further configured to operate on data local to the computing device in which they are resident. The indexlet repository <b>1203a</b> can be configured as the indexlet repository <b>1203</b> described previously. The indexlet repository <b>1203a</b> represents indexlets stored locally on the computing device.

0104<figref idref="f0018"><b>FIG. 18</b></figref> illustrates an example system and process for traversal of a data model. A data model <b>301</b> can comprise one or more tables. The one or more tables of the data model <b>301</b> can be stored in one or more databases on one or more separate computing devices. The data model <b>301</b> is shown with a table T1 <b>1802,</b> a table T2 <b>1803,</b> and a table T3 <b>1804.</b> The table <b>T1 1802</b> can comprise values for dimensions F1, F2, and F3. The table T2 <b>1803</b> can comprise values for dimensions F2, F4, and F5. The table T3 <b>1804</b> can comprise values for dimensions F5 and F6. One or more of the table T1 <b>1802,</b> the table T2 <b>1803,</b> and/or the table T3 <b>1804</b> can share one or more dimensions. The table T1 <b>1802</b> and the table T2 <b>1803</b> are shown as sharing the dimension F2. The table T1 <b>1802</b> and the table T3 <b>1804</b> are shown as sharing the dimension F5.

0105One or more workers can be employed to traverse the data model <b>301.</b> The one or more workers can comprise one or more indexlet workers and/or hypercube workers as described previously. For example, while shown as a single worker, a worker W1 <b>1808</b> can represent one or more indexlet workers and one or more hypercube workers working together. The one or more workers can be resident on a single computing device or resident on many computing devices (e.g., distributed). A worker W1 <b>1808,</b> a worker W2 <b>1809,</b> and a worker W2 <b>1810.</b> The worker W1 <b>1808,</b> the worker W2 <b>1809,</b> and the worker W2 <b>1810</b> can be configured to generate one or more virtual records by traversing the data model <b>301.</b><figref idref="f0018"><b>FIG. 18</b></figref> shows a virtual record <b>1811,</b> a virtual record <b>1812,</b> and a virtual record <b>1813.</b> Traversal of the data model <b>301</b> can be done in parallel by the worker W1 <b>1808,</b> the worker W2 <b>1809,</b> and the worker W2 <b>1810.</b> The worker W1 <b>1808</b> can generate the virtual record <b>1811</b> by a traversal <b>1805</b> of the table T2 <b>1803.</b> The worker W2 <b>1809</b> can generate the virtual record <b>1812</b> by a traversal <b>1806</b> of the table T1 <b>1802.</b> The worker W3 <b>1810</b> can generate the virtual record <b>1813</b> by a traversal <b>1807</b> of the table T3 <b>1804.</b>

0106As illustrated in <figref idref="f0018"><b>FIG. 18</b></figref><b>,</b> an expression <b>1801</b> (by way of example, the expression <b>1801</b> is shown as a hypercube expression, but any expression can be evaluated) can be determined over dimensions F1 and F6, evaluating expressions Expri(F4, F3) and Expr₂ (F2). Expression 1 (Expn) can depend only on dimensions F4 and F3. Expression 2 (Expr₂) can depend only on dimension F2. To evaluate Expression 1 and Expression 2, one or more virtual records that contain one or more of dimension F1, F2, F3, F4, and F6 can be generated to allow the hypercube to be determined and the expressions to be evaluated.

0107The expression <b>1801</b> can be provided to the worker W1 <b>1808,</b> the worker W2 <b>1809,</b> and/or the worker W2 <b>1810.</b> The expression <b>1801</b> can be provided to a worker manager <b>1814.</b> The worker manager <b>1814</b> can determine how the worker W1 <b>1808,</b> the worker W2 <b>1809,</b> and/or the worker W2 <b>1810</b> will traverse the data model <b>301.</b> For example, the worker manager <b>1814</b> can assign each worker a number of tables. As shown in <figref idref="f0018"><b>FIG. 18</b></figref><b>,</b> the worker W1 <b>1808</b> is traversing <b>1805</b> the table T2 <b>1803</b> to retrieve values for the dimensions needed to generate a virtual record <b>1811</b> (e.g., values for dimensions F2 and F4). Notably, the worker W1 <b>1808</b> will not retrieve values for the dimension F5, as those values are not necessary to generate the virtual record. The retrieved values can be assembled into the virtual record <b>1811.</b> The worker W2 <b>1809</b> is shown traversing <b>1806</b> the table T1 <b>1802</b> to retrieve values for the dimensions needed to generate a virtual record <b>1812</b> (e.g., values for dimensions F1, F2, and F3). Notably, the worker W2 <b>1809</b> will retrieve all values for all dimensions in the table T1 <b>1802,</b> as those values are necessary to generate the virtual record <b>1812.</b> The retrieved values can be assembled into the virtual record <b>1812.</b> The worker W3 <b>1810</b> is shown traversing <b>1807</b> the table T3 <b>1804</b> to retrieve values for the dimensions needed to generate a virtual record <b>1813</b> (e.g., values for dimension F6). Notably, the worker W3 <b>1810</b> will not retrieve values for the dimension F5, as those values are not necessary to generate the virtual record <b>1813.</b> The retrieved values can be assembled into a virtual record <b>1813.</b>

0108In an aspect, the values of the data model <b>301</b> can be divided by the worker manager <b>1814</b> into one or more partitions. For example, the worker manager <b>1814</b> can divide the data model <b>301</b> into partitions by quantity of data and/or number of values. The worker manager <b>1814</b> can partition within a dimension. Each of the worker W1 <b>1808,</b> the worker W2 <b>1809,</b> and/or the worker W2 <b>1810</b> can be assigned to a partition. For example, for a three worker scenario, the data model <b>301</b> can be partitioned into three partitions (e.g., into thirds). Each of the worker W1 <b>1808,</b> the worker W2 <b>1809,</b> and the worker W2 <b>1810</b> can then be assigned a partition that makes up one third of the data model <b>301.</b> Accordingly, each worker will retrieve values for all dimensions needed to evaluate the expression <b>1801.</b> The worker W1 <b>1808</b> would traverse any of the tables T1 <b>1802,</b> the table T2 <b>1803,</b> and the table T3 <b>1804</b> resident in the partition assigned to the worker W1 <b>1808</b> and retrieve values for the dimensions F1, F2, F3, F4, and F6 to generate the virtual record <b>1811.</b> Notably, the worker W1 <b>1808</b> would not retrieve values for dimension F5 as those values are necessary to generate the virtual record <b>1811.</b> The worker W2 <b>1809</b> would traverse any of the tables T1 <b>1802,</b> the table T2 <b>1803,</b> and the table T3 <b>1804</b> resident in the partition assigned to the worker W2 <b>1809</b> and retrieve values for the dimensions F1, F2, F3, F4, and F6 to generate the virtual record <b>1812.</b> Notably, the worker W2 <b>1809</b> would not retrieve values for dimension F5 as those values are necessary to generate the virtual record <b>1812.</b> The worker W3 <b>1810</b> would traverse any of the tables T1 <b>1802,</b> the table T2 <b>1803,</b> and the table T3 <b>1804</b> resident in the partition assigned to the worker W3 <b>1810</b> and retrieve values for the dimensions F1, F2, F3, F4, and F6 to generate the virtual record <b>1813.</b> Notably, the worker W3 <b>1810</b> would not retrieve values for dimension F5 as those values are necessary to generate the virtual record <b>1813.</b>

0109The worker W1 <b>1808,</b> the worker W2 <b>1809,</b> and the worker W2 <b>1810</b> thus can traverse the data model <b>301</b> with no overlap, and generate the virtual records <b>1811, 1812,</b> and <b>1813,</b> respectively. The virtual records <b>1811, 1812,</b> and <b>1813</b> can be sent to one or more aggregators. The one or more aggregators can be resident on a single computing device or resident on many computing devices (e.g., distributed). <figref idref="f0018"><b>FIG. 18</b></figref> shows an aggregator <b>1815</b> and an aggregator <b>1816.</b> In an aspect, each virtual record <b>1811, 1812,</b> and <b>1813</b> can be hashed by a respective aggregator and the hashes (e.g., hash value) can be associated with the respective virtual records. Each virtual record <b>1811, 1812,</b> and <b>1813</b> can be sent to the aggregator <b>1815</b> and the aggregator <b>1816.</b> The aggregator <b>1815</b> and the aggregator <b>1816</b> can each receive the virtual records <b>1811, 1812,</b> and <b>1813,</b> generate the hashes, aggregate the values, evaluate the expressions, and store a result <b>1819</b> and a result <b>1820</b> in one or more databases (e.g., a database <b>1817</b> and/or a database <b>1818</b>). The aggregator <b>1815</b> and the aggregator <b>1816</b> can generate a hash table. The values of the dimensions in the virtual record can be hashed (e.g., <figref idref="f0004"><b>FIG. 4</b></figref><b>,</b> "Dim for hash"). That hash can be used to point to a container of the hash table that keeps track of the aggregations. If the aggregation is additive, e.g., sum, count, the performance of the aggregation can be distributed, performed in parallel, and the results combined as needed. In another aspect, an accumulator can be configured to perform the aggregation by considering only newly arrived data and subsequently combing those results with previously determined results. But if the aggregation is, for example, an average, the aggregator <b>1815</b> and the aggregator <b>1816</b> can sum and keep track of the number of elements that have been aggregated. And then at the end the operation can be executed fully (e.g., divide sum by number of elements pushed). The one or more databases can be stored on one or more separate computing devices. The aggregator <b>1815</b> and the aggregator <b>1816</b> can store the virtual records <b>1811, 1812,</b> and <b>1813</b> and/or the results according to a hashing strategy.

0110To increase parallelism, all virtual records and/or all the hashes of the virtual records can be sent to all the aggregators. The one or more databases can reside on one or more separate computing devices. A simple hash, such as a modulus of an identifier of a virtual record can be used. For example, if all identifiers of virtual records are numbers all odd numbered identifiers can be assigned to one computing device and all even numbered identifiers can be assigned to another computing device. Assuming a balanced number of odd and even numbered identifiers, and a balanced data size per identifier, the virtual records would be balanced between the two computing devices. Other hashing algorithms can be used for more complex hashing. For example, the MD5 algorithm.

0111In another aspect, a hash-ring, or consistent hashing, strategy can be implemented whereby the virtual records are distributed using a hashing algorithm to determine the locations of the virtual records. Using the hash of an identifier of a virtual record as a key, or of the virtual record itself as a key, can enable later determinations of where the virtual records should reside. Consistent hashing is a distributed hashing scheme that operates independently of the number of servers or objects in a distributed hash table by assigning them a position on an abstract circle, or hash ring. This allows servers and objects to scale without affecting the overall system. In a hash ring, a file name (e.g., virtual record identifier) can generate a hash output corresponding to a position around a circle. The closest clockwise node (e.g., computing device) to that location can contain the file itself (e.g., the virtual record), or a link to another location containing or leading to the file. When a link has become obsolete and no longer points to the file, it may be necessary to search for that file (e.g., virtual record).

0112The one or more aggregators can report the results of evaluating the expression <b>1801.</b> One or more rows of the resultant hypercube can be reported. The one or more aggregators can be queried. As shown in <figref idref="f0018"><b>FIG. 18</b></figref><b>,</b> the aggregator <b>1815</b> and the aggregator <b>1816</b> can be queried. The aggregator <b>1815</b> and the aggregator <b>1816</b> can be queried for one or more results and/or virtual records. The aggregator <b>1815</b> and the aggregator <b>1816</b> can be queried for a combined virtual record that contains all values for the dimensions used in the expression <b>1801.</b> The results can be sorted. In an aspect, the one or more aggregators can stop evaluating the hypercube arbitrarily at some point if a windowed fetch of results is satisfied. In an aspect, the one or more aggregators can completely evaluate the hypercube and fetch a window of results.

0113The methods provided can be implemented by means of a computer program as illustrated in a flowchart of a method <b>1900</b> in <figref idref="f0019"><b>FIG. 19</b></figref><b>.</b> In a step <b>1902,</b> the program can read some or all data records in the database, for instance using a SELECT statement which selects all the tables of the database, e.g. Tables 1-5 of <figref idref="f0002"><b>FIG. 2</b></figref><b>.</b> In an aspect, the database can be read into primary memory of a computer.

0114To increase evaluation speed, each unique value of each data variable in said database can be assigned a different binary code and the data records can be stored in binary-coded form. This can be performed, for example, when the program first reads the data records from the database. For each input table, the following steps can be carried out. The column names, e.g. the variables, of the table can be read (e.g., successively). Every time a new data variable appears, a data structure can be instantiated for the new data variable. An internal table structure can be instantiated to contain some or all the data records in binary form, whereupon the data records can be read (e.g., successively) and binary-coded. For each data value, the data structure of the corresponding data variable can be checked to establish if the value has previously been assigned a binary code. If so, that binary code can be inserted in the proper place in the above-mentioned table structure. If not, the data value can be added to the data structure and assigned a new binary code, for example the next binary code in ascending order, before being inserted in the table structure. In other words, for each data variable, a unique binary code can be assigned to each unique data value.

0115After having read some or all data records in the database, the method <b>1900</b> can analyze the database in a step <b>1904</b> to identify all connections between the data tables. A connection between two data tables means that these data tables have one variable in common. In an aspect, step <b>1904</b> can comprise generation of one or more bidirectional table indexes and one or more bidirectional associative indexes. In an aspect, generation of one or more bidirectional table indexes and one or more bidirectional associative indexes can comprise a separate step. In another aspect, generation of one or more bidirectional table indexes and one or more bidirectional associative indexes can be on demand. After the analysis, all data tables are virtually connected. In <figref idref="f0002"><b>FIG. 2</b></figref><b>,</b> such virtual connections are illustrated by double ended arrows. The virtually connected data tables can form at least one so-called "snowflake structure," a branching data structure in which there is one and only one connecting path between any two data tables in the database. Thus, a snowflake structure does not contain any loops. If loops do occur among the virtually connected data tables, e.g. if two tables have more than one variable in common, a snowflake structure can in some cases still be formed by means of special algorithms known in the art for resolving such loops.

0116After this initial analysis, the user can explore the database. In doing so, the user defines in step <b>1906</b> a mathematical function, which could be a combination of mathematical expressions. Assume that the user wants to extract the total sales per year and client from the database in <figref idref="f0002"><b>FIG. 2</b></figref><b>.</b> The user defines a corresponding mathematical function "SUM (x*y)", and selects the calculation variables to be included in this function: "Price" and "Number." The user also selects the classification variables: "Client" and" Year."

0117The method <b>1900</b> then identifies in step <b>1908</b> all relevant data tables, e.g. all data tables containing any one of the selected calculation and classification variables, such data tables being denoted boundary tables, as well as intermediate data tables in the connecting path(s) between these boundary tables in the snowflake structure, such data tables being denoted connecting tables. There are no connecting tables in the present example. In an aspect, one or more bidirectional table indexes and one or more bidirectional associative indexes can be accessed as part of step <b>1908.</b>

0118In the present example, all occurrences of every value, e.g. frequency data, of the selected calculation variables can be included for evaluation of the mathematical function. In <figref idref="f0002"><b>FIG. 2</b></figref><b>,</b> the selected variables ("Price," "Number") can require such frequency data. Now, a subset (B) can be defined that includes all boundary tables (Tables 1-2) containing such calculation variables and any connecting tables between such boundary tables in the snowflake structure. It should be noted that the frequency requirement of a particular variable is determined by the mathematical expression in which it is included. Determination of an average or a median calls for frequency information. In general, the same is true for determination of a sum, whereas determination of a maximum or a minimum does not require frequency data of the calculation variables. It can also be noted that classification variables in general do not require frequency data.

0119Then, a starting table can be selected in step <b>1910,</b> for example, among the data tables within subset (B). In an aspect, the starting table can be the data table with the largest number of data records in this subset. In <figref idref="f0002"><b>FIG. 2</b></figref><b>,</b> Table 2 can be selected as the starting table. Thus, the starting table contains selected variables ("Client," "Number"), and connecting variables ("Date," "Product"). These connecting variables link the starting table (Table 2) to the boundary tables (Tables 1 and 3).

0120Thereafter, a conversion structure can be built in step <b>1912.</b> This conversion structure can be used for translating each value of each connecting variable ("Date," "Product") in the starting table (Table 2) into a value of a corresponding selected variable ("Year," "Price") in the boundary tables (Table 3 and 1, respectively). A table of the conversion structure can be built by successively reading data records of Table 3 and creating a link between each unique value of the connecting variable ("Date") and a corresponding value of the selected variable ("Year"). It can be noted that there is no link from value 4 ("Date: 1999-01-12"), since this value is not included in the boundary table. Similarly, a further table of the conversion structure can be built by successively reading data records of Table 1 and creating a link between each unique value of the connecting variable ("Product") and a corresponding value of the selected variable ("Price"). In this example, value 2 ("Product: Toothpaste") is linked to two values of the selected variable ("Price: 6.5"), since this connection occurs twice in the boundary table. Thus, frequency data can be included in the conversion structure. Also note that there is no link from value 3 ("Product: Shampoo").

0121When the conversion structure has been built, a virtual data record can be created. Such a virtual data record accommodates all selected variables ("Client," "Year," "Price," "Number") in the database. In building the virtual data record, a data record is read in step <b>1914</b> from the starting table (Table 2). Then, the value of each selected variable ("Client", "Number") in the current data record of the starting table can be incorporated in the virtual data record in a step <b>1916.</b> Also, by using the conversion structure each value of each connecting variable ("Date", "Product") in the current data record of the starting table can be converted into a value of a corresponding selected variable ("Year", "Price"), this value also being incorporated in the virtual data record.

0122In step <b>1918</b> the virtual data record can be used to build an intermediate data structure. Each data record of the intermediate data structure can accommodate each selected classification variable (dimension) and an aggregation field for each mathematical expression implied by the mathematical function. The intermediate data structure can be built based on the values of the selected variables in the virtual data record. Thus, each mathematical expression can be evaluated based on one or more values of one or more relevant calculation variables in the virtual data record, and the result can be aggregated in the appropriate aggregation field based on the combination of current values of the classification variables ("Client," "Year").

0123The above procedure can be repeated for one or more additional (e.g., all) data records of the starting table. In a step <b>1920</b> it can be checked whether the end of the starting table has been reached. If not, the process can be repeated from step <b>1914</b> and further data records can be read from the starting table. Thus, an intermediate data structure can be built by successively reading data records of the starting table, by incorporating the current values of the selected variables in a virtual data record, and by evaluating each mathematical expression based on the content of the virtual data record. If the current combination of values of classification variables in the virtual data record is new, a new data record can be created in the intermediate data structure to hold the result of the evaluation. Otherwise, the appropriate data record is rapidly found, and the result of the evaluation is aggregated in the aggregation field.

0124Thus, data records can be added to the intermediate data structure as the starting table is traversed. The intermediate data structure can be a data table associated with an efficient index system, such as an AVL or a hash structure. The aggregation field can be implemented as a summation register, in which the result of the evaluated mathematical expression is accumulated.

0125In some aspects, e.g. when evaluating a median, the aggregation field can be implemented to hold all individual results for a unique combination of values of the specified classification variables. It should be noted that only one virtual data record is needed in the procedure of building the intermediate data structure from the starting table. Thus, the content of the virtual data record can be updated for each data record of the starting table. This can minimize the memory requirement in executing the computer program.

0126After traversing the starting table, the intermediate data structure can contain a plurality of data records. If the intermediate data structure accommodates more than two classification variables, the intermediate data structure can, for each eliminated classification variable, contain the evaluated results aggregated over all values of this classification variable for each unique combination of values of remaining classification variables.

0127When the intermediate data structure has been built, a final data structure, e.g., a multidimensional cube, as shown in non-binary notation in Table 6 of <figref idref="f0020"><b>FIG. 20</b></figref><b>,</b> can be created in a step <b>1922</b> by evaluating the mathematical function ("SUM (x*y)") based on the results of the mathematical expression ("x*y") contained in the intermediate data structure. In doing so, the results in the aggregation fields for each unique combination of values of the classification variables can be combined. In the example, the creation of the final data structure is straightforward, due to the trivial nature of the present mathematical function. The content of the final data structure can be presented to the user, for example in a two-dimensional table, in step <b>1924,</b> as shown in Table 7 of <figref idref="f0020"><b>FIG. 20</b></figref><b>.</b> Alternatively, if the final data structure contains many dimensions, the data can be presented in a pivot table, in which the user can interactively move up and down in dimensions, as is well known in the art.

0128In an aspect, step <b>1922</b> can involve any of the processes described herein as part of a process for creating the hypercube/multidimensional cube. For example, output from the logical inference engine <b>106</b> utilizing one or more BTIs and or one or more A2A indexes can be used in creation of the hypercube/multidimensional cube. When a user makes a selection, the logical inference engine <b>106</b> calculates a data subset of which one or more BTIs and/or A2A indexes can be generated and provided to the chart engine <b>108</b> for use in generating a hypercube/multidimensional cube and/or evaluating one or more expressions against a hypercube/multidimensional cube via one or more BTIs and/or A2A indexes as described herein.

0129At step <b>1926,</b> input from the user can be received. For example, input form the user can be a selection and/or de-selection of the presented results.

0130Optionally, input from the user at step <b>1926</b> can comprise a request for external processing. In an aspect, the user can be presented with an option to select one or more external engines to use for the external processing. Optionally, at step <b>1928,</b> data underlying the user selection can be configured (e.g., formatted) for use by an external engine. Optionally, at step <b>1930,</b> the data can be transmitted to the external engine for processing and the processed data can be received. The received data can undergo one or more checks to confirm that the received data is in a form that can be appended to the data model. For example, one or more of an integrity check, a format check, a cardinality check, combinations thereof, and the like. Optionally, at step <b>1932,</b> processed data can be received from the external engine and can be appended to the data model as described herein. In an aspect, the received data can have a lifespan that controls how long the received data persists with the data model. For example, the received data can be incorporated into the data model in a manner that enables a user to retrieve the received data at another time/session. In another example, the received data can persist only for the current session, making the received data unavailable in a future session.

0131<figref idref="f0021"><b>FIG. 21</b></figref> illustrates how a selection <b>2100</b> operates on a scope <b>2101</b> of presented data to generate a data subset <b>2102.</b> The data subset <b>2102</b> can form a state space, which is based on a selection state given by the selection <b>2100.</b> In an aspect, the selection state (or "user state") can be defined by a user clicking on list boxes and graphs in a user interface of an application. An application can be designed to host a number of graphical objects (charts, tables, etc.) that evaluate one or more mathematical functions (also referred to as an "expression") on the data subset <b>2102</b> for one or more dimensions (classification variables). The result of this evaluation creates a chart result <b>2103</b> which can be a multidimensional cube which can be visualized in one or more of the graphical objects.

0132The application can permit a user to explore the scope <b>2101</b> by making different selections, by clicking on graphical objects to select variables, which causes the chart result <b>2103</b> to change. At every time instant during the exploration, there exists a current state space, which can be associated with a current selection state that is operated on the scope <b>2101</b> (which always remains the same).

0133As illustrated in <figref idref="f0021"><b>FIG. 21</b></figref><b>,</b> when a user makes a selection, the inference engine <b>2104</b> calculates a data subset. Also, an identifier ID1 for the selection together with the scope can be generated based on the filters in the selection and the scope. Subsequently, an identifier ID2 for the data subset is generated based on the data subset definition, for example a bit sequence that defines the content of the data subset. ID2 can be put into a cache using ID1 as a lookup identifier. Likewise, the data subset definition can be put in the cache using ID2 as a lookup identifier.

0134As shown in <figref idref="f0021"><b>FIG. 21</b></figref><b>,</b> a chart calculation in a calculation/chart engine <b>2105</b> takes place in a similar way. Here, there are two information sets: the data subset <b>2102</b> and relevant chart properties <b>2106.</b> The latter can be, but not restricted to, a mathematical function together with calculation variables and classification variables (dimensions). Both of these information sets can be used to calculate the chart result <b>2103,</b> and both of these information sets can be also used to generate identifier ID3 for the input to the chart calculation. ID2 can be generated already in the previous step, and ID3 can be generated as the first step in the chart calculation procedure.

0135The identifier ID3 can be formed from ID2 and the relevant chart properties. ID3 can be seen as an identifier for a specific chart generation instance, which can include all information needed to calculate a specific chart result. In addition, a chart result identifier ID4 can be created from the chart result definition, for example a bit sequence that defines the chart result <b>2103.</b> ID4 can be put in the cache using ID3 as a lookup identifier. Likewise, the chart result definition can be put in the cache using ID4 as a lookup identifier.

0136Optionally, further calculations, transforming, and/or processing can be included through an extension engine <b>2107.</b> Optionally, associated results from the inference engine <b>2104</b> and further computed by hypercube computation in said calculation/chart engine <b>2105</b> can be coupled to an external engine <b>2108</b> that can comprise one or more data processing applications (e.g., simulation applications, statistical applications, mathematical computation applications, database applications, combinations thereof, and the like). Context of a data model processed by the inference engine <b>2104</b> can comprise a tuple or tuples of values defined by dimensions and expressions computed by hypercube routines. Data can be exchanged through an interface <b>2109.</b>

0137The associated results coupled to the external engine <b>2108</b> can be intermediate. Further results that can be final hypercube results can also be received from the external engine <b>2108.</b> Further results can be fed back to be included in the Data/Scope <b>2101</b> and enrich the data model. The further results can also be rendered directly to the user in the chart result <b>2103.</b> Data received from and computed by the external engine <b>2108</b> can be used for further associative discovery.

0138The database as referred to in Tables 1-5 of <figref idref="f0002"><b>FIG. 2</b></figref> can be queried by specifying the data element types and data element values of interest and by further specifying any functions to apply to the data contained within the specified data element types of the database. The functions which can be used within a query can include, for example, expressions using statistics, sub-queries, filters, mathematical formulas, and the like, to help the user to locate and/or calculate the specific information wanted from the database. Once located and/or calculated, the results of a query can be displayed to the user with various visualization techniques and objects such as list boxes of a user interface illustrated in <figref idref="f0022"><b>FIG. 22</b></figref><b>.</b>

0139The graphical objects ( or visual representations) can be substantially any display or output type including graphs, charts, trees, multi-dimensional depictions, images (computer generated or digital captures), video/audio displays describing the data, hybrid presentations where output is segmented into multiple display areas having different data analysis in each area and so forth. A user can select one or more default visual representations; however, a subsequent visual representation can be generated on the basis of further analysis and subsequent dynamic selection of the most suitable form for the data.

0140In an aspect, a user can select a data point and a visualization component can instantaneously filter and re-aggregate other fields and corresponding visual representations based on the user's selection. In an aspect, the filtering and re-aggregation can be completed without querying a database. In an aspect, a visual representation can be presented to a user with color schemes applied meaningfully. For example, a user selection can be highlighted in green, datasets related to the selection can be highlighted in white, and unrelated data can be highlighted in gray. A meaningful application of a color scheme provides an intuitive navigation interface in the state space.

0141The result of a standard query can be a smaller subset of the data within the database, or a result set, which is comprised of the records, and more specifically, the data element types and data element values within those records, along with any calculated functions, that match the specified query. For example, as indicated in <figref idref="f0022"><b>FIG. 22</b></figref><b>,</b> the data element value "Nisse" can be specified as a query or filtering criteria as indicated by a frame in the "Client" header row. In some aspects, the selected element can be highlighted in green. By specifically selecting "Nisse," other data element values in this row are excluded as shown by gray areas. Further, "Year" "1999" and "Month" "Jan" are selected in a similar way.

0142Optionally, in this application, external processing can also be requested by ticking "External" in the user interface of <figref idref="f0022"><b>FIG. 22</b></figref><b>.</b> Data as shown in <figref idref="f0023"><b>FIG. 23</b></figref> can be exchanged with an External engine <b>2108</b> through the interface <b>2109</b> of <figref idref="f0021"><b>FIG. 21</b></figref><b>.</b> In addition to evaluating the mathematical function ("SUM (Price*Number)") based on the results of the mathematical expression ("Price*Number") contained in the intermediate data structure the mathematical function ("SUM (ExtFunc(Price*Number))") can be evaluated. Data sent out are (Nisse, 1999, Jan, {19.5, null}). In this case the external engine <b>2108</b> can process data in accordance with the formula <pre listing-type="program-listing">if (x==null) y=0.5 else y=x</pre> as shown in in <figref idref="f0023"><b>FIG. 23</b></figref><b>.</b> The result input through the interface <b>2109</b> will be (19.5, 0.5) as reflected in the graphical presentation in <figref idref="f0022"><b>FIG. 22</b></figref><b>.</b>

0143In a further aspect, external processing can also be optionally requested by ticking "External" in a box as shown in <figref idref="f0023"><b>FIG. 23</b></figref><b>.</b> Data as shown in <figref idref="f0025"><b>FIG. 25</b></figref> can be exchanged with an external engine <b>2108</b> through the Interface <b>2109</b> of <figref idref="f0021"><b>FIG. 21</b></figref><b>.</b> In addition to evaluating the mathematical function ("SUM(Price*Number)") based on the results of the mathematical expression ("Price*Number") contained in the intermediate data structure the mathematical function <maths id="math0021"><math display="block"><mi>SUM</mi><mfenced><mi>ExtFunc</mi><mfenced><mi>Price*Number</mi></mfenced></mfenced></math><img file="EP3678032B1_D0025.tif" /></maths> can be evaluated. Data sent out are (Nisse, 1999, Jan, {19.5, null}). In this case the external engine <b>2108</b> will process data in accordance with Function (1) as shown below and in <figref idref="f0025"><b>FIG. 25</b></figref><b>.</b> The result input through the Interface <b>2109</b> will be (61.5) as reflected in the graphical presentation in <figref idref="f0024"><b>FIG. 24</b></figref><b>.</b><img file="EP3678032B1_D0026.tif" />

0144A further optional embodiment is shown in <figref idref="f0026"><b>FIG. 26</b></figref> and <figref idref="f0027"><b>FIG. 27</b></figref><b>.</b> The same basic data as in previous examples apply. A user selects "Pekka," "1999," "Jan," and "External." By selecting "External," already determined and associated results are coupled to the external engine <b>2108.</b> Feedback data from the external engine <b>2108</b> based on an external computation, ExtQualification(Sum(Price*Number)), as shown in <figref idref="f0028"><b>FIG. 28</b></figref> will be the information "MVG." This information can be fed back to the logical inference engine <b>2104.</b> The information can also be fed back to the graphical objects of <figref idref="f0026"><b>FIG. 26</b></figref> and as a result a qualification table <b>2600</b> will highlight "MVG" (illustrated with a frame in <figref idref="f0026"><b>FIG. 26</b></figref>). Other values (U, G, and VG) are shown in gray areas. The result input through the Interface <b>2109</b> will be Soap with a value of 75 as reflected in the graphical presentation (bar chart) of <figref idref="f0026"><b>FIG. 26</b></figref><b>.</b><figref idref="f0027"><b>FIG. 27</b></figref> is a schematic representation of data exchanged with an external engine based on selections in <figref idref="f0026"><b>FIG. 26</b></figref><b>.</b><figref idref="f0029"><b>FIG. 29</b></figref> is a table showing results from computations based on different selections in the presentation of <figref idref="f0026"><b>FIG. 26</b></figref><b>.</b>

0145Should a user instead select "Gullan," "1999," "Jan," and "External," the feedback signal would include "VG" based on the content shown in qualification table <b>2600.</b> The computations actually performed in the external engine <b>2108</b> are not shown or indicated, since they are not relevant to the inference engine.

0146In <figref idref="f0028"><b>FIG. 28</b></figref> a user has selected "G" as depicted by <b>2800</b> in the qualification table <b>2600.</b> As a result information fed back from the external engine <b>2108</b> to the external engine <b>2107</b> and further to the inference engine <b>2104</b> the following information will be highlighted: "Nisse," "1999," and "Jan" as shown in <figref idref="f0028"><b>FIG. 28</b></figref><b>.</b> Furthermore, the result produced will be Soap 37.5 as reflected in the graphical presentation (bar chart) of <figref idref="f0028"><b>FIG. 28</b></figref><b>.</b>

0147In an aspect, methods and data structures for filtering and/or sorting data are described. Data in the data model <b>301</b> may be sorted as described herein and/or one or more results of a hypercube may be sorted as described herein. As shown in <figref idref="f0030"><b>FIG. 30</b></figref><b>,</b> one or more data values (e.g., A, C, D, F, G, H, K, L, M, N, P, Z) may be stored in one or more databases <b>3002, 3004,</b> and/or <b>3006</b> and/or may be stored as one or more results of a hypercube operation as described herein (for example, results of the aggregator <b>1815</b> and/or the aggregator <b>1816</b> as described herein). The one or more data values may be stored in an unsorted and/or partially sorted state. One or more filters (e.g., masks, thresholds, etc...) may be determined. The one or more filters are each configured to describe the data in a non-overlapping fashion. For example, a filter may divide the one or more data values in half, another filter may further divide the one or more data values into a first quarter, another filter may further divide the one or more data values into a second quarter, another filter may further divide the one or more data values into a third quarter, another filter may further divide the one or more data values into a fourth quarter, and the like. Any number of filters may be determined (e.g., further subdivide the quarters into eighths, etc...). The one or more filters may be based on anything with a user-defined general expression returning a sortable value. The one or more filters may divide data values based on the nature of the underlying one or more data values being divided, for example, the one or more filters may be alphabetical, numerical, geographical, temporal, quantity-based, and the like. The one or more filters may divide the one or more data values based on arbitrary values. For example, a filter may divide the values based on a specific value, accordingly, any data values less than or equal to (alternatively, merely less than) the specific value will be divided from values greater than (alternatively, greater than or equal to) the specific value. The filters may be determined based on the actual data values themselves. For example, the filter may divide the one or more data values based on a midpoint (e.g., half) within the one or more data values. For example, if the data values may range from 1-10, rather than divide at 5, the filter may divide based on the actual data values present. For example, if the data values 1, 2, 3, 7, 8, 9, 10 were present, the filter may divide values based on whether the values are less than or equal to (alternatively, merely less than) 7, despite 7 not being the midpoint of the possible range of values (e.g., 5). In another aspect, the filters may be quantity-based. The filter may represent a division based on the quantity of data values present. For example, if 1,000,000 data values are present, a first filter may divide the data values into two sets of 500,000 data values, a second filter may divide out the first 250,000 data values, a third filter may divide out the second 250,000 data values, a fourth filter may divide out the third 250,000 data values, and a fifth filter may divide out the last 250,000 data values. As shown <figref idref="f0030"><b>FIG. 30</b></figref><b>, 3008</b> indicates example filter divisions for a filter f₁, a filter f₂, and a filter f₃. Filter f₁ represents a division between the first half and the second half of the data values (e.g., quantity-based). Filter f₂ represents a division between the first quarter and the second quarter of the data values. Filter f₃ represents a division between the third quarter and the fourth quarter of the data values. "Satisfies" or "does not satisfy" as used herein may be used to indicate that a data value falls within one side or another side of a division established by a filter. Additional successive filters may be implemented indicated by filter fn.

0148The one or more data values can be determined and, for each, filter, a data structure (or a portion of a data structure) can be generated. For each data value that satisfies a filter, a first symbol can be stored in the data structure (or the portion of the data structure). For each data value that does not satisfy the filter, a second symbol can be stored in the data structure (or the portion of the data structure). Any symbol may be used, for example a 1 (one) may indicate that the data value satisfies the filter whereas a 0 (zero) may indicate that the data value does not satisfy the filter.

0149As shown in data structure <b>3010,</b> the data value "C" satisfies the filter f₁ because "C" is in the first half of the data values, accordingly a 1 is entered into the data structure <b>3010</b> to indicate that "C" satisfies filter f₁. The data value "P" does not satisfy the filter f₁ because "P" is not in the first half of the data values, accordingly a 0 is entered into the data structure <b>3010</b> to indicate that "P" does not satisfy filter f₁. The data value "F" satisfies the filter f₁ because "F" is in the first half of the data values, accordingly a 1 is entered into the data structure <b>3010</b> to indicate that "F" satisfies filter f₁. The data value "G" satisfies the filter f₁ because "G" is in the first half of the data values, accordingly a 1 is entered into the data structure <b>3010</b> to indicate that "G" satisfies filter f₁.

0150As shown in data structure <b>3010,</b> the data value "H" satisfies the filter f₁ because "H" is in the first half of the data values, accordingly a 1 is entered into the data structure <b>3010</b> to indicate that "H" satisfies filter f₁. The data value "N" does not satisfy the filter f₁ because "N" is not in the first half of the data values, accordingly a 0 is entered into the data structure <b>3010</b> to indicate that "N" does not satisfy filter f₁. The data value "M" does not satisfy the filter f₁ because "M" is not in the first half of the data values, accordingly a 0 is entered into the data structure <b>3010</b> to indicate that "M" does not satisfy filter f₁. The data value "K" does not satisfy the filter f₁ because "K" is not in the first half of the data values, accordingly a 0 is entered into the data structure <b>3010</b> to indicate that "K" does not satisfy filter f₁.

0151As shown in data structure <b>3010,</b> the data value "A" satisfies the filter f₁ because "A" is in the first half of the data values, accordingly a 1 is entered into the data structure <b>3010</b> to indicate that "A" satisfies filter f₁. The data value "D" satisfies the filter f₁ because "D" is in the first half of the data values, accordingly a 1 is entered into the data structure <b>3010</b> to indicate that "D" satisfies filter f₁. The data value "L" does not satisfy the filter f₁ because "L" is not in the first half of the data values, accordingly a 0 is entered into the data structure <b>3010</b> to indicate that "L" does not satisfy filter f₁. The data value "Z" does not satisfy the filter f₁ because "Z" is not in the first half of the data values, accordingly a 0 is entered into the data structure <b>3010</b> to indicate that "Z" does not satisfy filter f₁.

0152As shown in data structure <b>3012,</b> the data value "C" satisfies the filter f₂ because "C" is in the first quarter of the data values, accordingly a 1 is entered into the data structure <b>3012</b> to indicate that "C" satisfies filter f₂. The data value "P" does not satisfy the filter f₂ because "P" is not in the first quarter of the data values, accordingly a 0 is entered into the data structure <b>3012</b> to indicate that "P" does not satisfy filter f₂. The data value "F" does not satisfy the filter f₂ because "F" is not in the first quarter of the data values, accordingly a 0 is entered into the data structure <b>3012</b> to indicate that "F" does not satisfy filter f₂. The data value "G" does not satisfy the filter f₂ because "G" is not in the first quarter of the data values, accordingly a 0 is entered into the data structure <b>3012</b> to indicate that "G" does not satisfy filter f₂.

0153As shown in data structure <b>3012,</b> the data value "H" does not satisfy the filter f₂ because "H" is not in the first quarter of the data values, accordingly a 0 is entered into the data structure <b>3012</b> to indicate that "H" does not satisfy filter f₂. The data value "N" does not satisfy the filter f₂ because "N" is not in the first quarter of the data values, accordingly a 0 is entered into the data structure <b>3012</b> to indicate that "N" does not satisfy filter f₂. The data value "M" does not satisfy the filter f₂ because "M" is not in the first quarter of the data values, accordingly a 0 is entered into the data structure <b>3012</b> to indicate that "M" does not satisfy filter f₂. The data value "K" does not satisfy the filter f₂ because "K" is not in the first quarter of the data values, accordingly a 0 is entered into the data structure <b>3012</b> to indicate that "K" does not satisfy filter f₂.

0154As shown in data structure <b>3012,</b> the data value "A" satisfies the filter f₂ because "A" is in the first quarter of the data values, accordingly a 1 is entered into the data structure <b>3012</b> to indicate that "A" satisfies filter f₂. The data value "D" satisfies the filter f₂ because "D" is in the first quarter of the data values, accordingly a 1 is entered into the data structure <b>3012</b> to indicate that "D" satisfies filter f₂. The data value "L" does not satisfy the filter f₂ because "L" is not in the first quarter of the data values, accordingly a 0 is entered into the data structure <b>3012</b> to indicate that "L" does not satisfy filter f₂. The data value "Z" does not satisfy the filter f₂ because "Z" is not in the first quarter of the data values, accordingly a 0 is entered into the data structure <b>3012</b> to indicate that "Z" does not satisfy filter f₂.

0155As shown in data structure <b>3014,</b> the data value "C" does not satisfy the filter f₃ because "C" is not in the fourth quarter of the data values, accordingly a 0 is entered into the data structure <b>3014</b> to indicate that "C" does not satisfy filter f₃. The data value "P" satisfies the filter f₃ because "P" is in the fourth quarter of the data values, accordingly a 1 is entered into the data structure <b>3014</b> to indicate that "P" satisfies filter f₃. The data value "F" does not satisfy the filter f₃ because "F" is not in the fourth quarter of the data values, accordingly a 0 is entered into the data structure <b>3014</b> to indicate that "F" does not satisfy filter f₃. The data value "G" does not satisfy the filter f₃ because "G" is not in the fourth quarter of the data values, accordingly a 0 is entered into the data structure <b>3014</b> to indicate that "G" does not satisfy filter f₃.

0156As shown in data structure <b>3014,</b> the data value "H" does not satisfy the filter f₃ because "H" is not in the fourth quarter of the data values, accordingly a 0 is entered into the data structure <b>3014</b> to indicate that "H" does not satisfy filter f₃. The data value "N" satisfies the filter f₃ because "N" is in the fourth quarter of the data values, accordingly a 1 is entered into the data structure <b>3014</b> to indicate that "N" satisfies filter f₃. The data value "M" does not satisfy the filter f₃ because "M" is not in the fourth quarter of the data values, accordingly a 0 is entered into the data structure <b>3014</b> to indicate that "M" does not satisfy filter f₃. The data value "K" does not satisfy the filter f₃ because "K" is not in the fourth quarter of the data values, accordingly a 0 is entered into the data structure <b>3014</b> to indicate that "K" does not satisfy filter f₃.

0157As shown in data structure <b>3014,</b> the data value "A" does not satisfy the filter f₃ because "A" is not in the fourth quarter of the data values, accordingly a 0 is entered into the data structure <b>3014</b> to indicate that "A" does not satisfy filter f₃. The data value "D" does not satisfy the filter f₃ because "D" is not in the fourth quarter of the data values, accordingly a 0 is entered into the data structure <b>3014</b> to indicate that "D" does not satisfy filter f₃. The data value "L" does not satisfy the filter f₃ because "L" is not in the fourth quarter of the data values, accordingly a 0 is entered into the data structure <b>3014</b> to indicate that "L" does not satisfy filter f₃. The data value "Z" satisfies the filter f₃ because "Z" is in the fourth quarter of the data values, accordingly a 1 is entered into the data structure <b>3014</b> to indicate that "Z" satisfies filter f₃.

0158Additional filters may be applied to the data values, resulting in additional data structures <b>3018</b> etc... indicating whether a data value satisfies or does not satisfy a given filter. The data structures may be combined into a single data structure (e.g., a table or an array) or may be stored separately (e.g., a table, an array, or a string). In a single data structure embodiment, the unsorted data values <b>3002, 3004,</b> and <b>3006</b> may each correspond to a respective row of the data structure, and each filter may correspond to a respective column. A cell at the intersection of a row and a column will indicate via the symbol whether or not the data value satisfies or does not satisfy the filter.

0159A query can then be run against the resulting data structure. A filter that relates to the query may be determined. For example, a query for values that are in the first half of the data (e.g., alphabetically preceding "K") will determine that filter f₁ is related to the query and the data structure can be accessed to determine the symbols in the column for filter f₁. The result of the query will determine the rows having a symbol of 1, indicating that the data values associated with those rows satisfy the filter. At <b>3018,</b> the data values associated with the rows can then be extracted from the database (e.g., data values C, F, G, H, A, and D) and sorted at <b>3020.</b> Thus, the data sorting data structure described reduces the amount of data that is ordinarily extracted in order to determine data satisfying the query and sorting the result. An ordinary query would have to extract all the data values, sort all the data values (which will ordinarily be time and computationally prohibitive), and then identify the data values in the first half of the data. The present method permits targeted extraction of data values that satisfy a filter, and then the reduced quantity of data values can be easily sorted, utilizing less time and computational resources. Advantages like these will have a greater impact on users and systems in cases of a large number of data values being processed. How large amount of data values that is required in such a processing before an impact is noticeable to such a degree that a person skilled in the art would consider it worth acting upon may vary depending on the hardware of the system processing the data values, i.e. the faster processing capacity and the more memory the larger number of data values may be processed without the person skilled in the art would consider making changes to the process. In most of the systems today the skilled person would most likely not perceive any need to make changes to the processing for a number of data values that is lesser than ten thousand, in some cases lesser than one million.

0160In an aspect, illustrated in <figref idref="f0031"><b>FIG. 31</b></figref> provided is a method <b>3100, e.g. implemented in a computer,</b> comprising determining data values of a plurality of unsorted data values stored in one or more databases at <b>3102</b> and generating a data structure at <b>3104.</b> The data structure can comprise a plurality of rows, wherein each row of the plurality of rows is associated with a respective data value of the plurality of unsorted data values and a plurality of columns intersecting the plurality of rows to define a plurality of cells, wherein each column is associated with a respective filter of a plurality of filters. Each cell of the plurality of cells comprises a first symbol or a second symbol, based on whether the respective data value satisfies or does not satisfy the respective filter, respectively. The one or more databases can be derived from a result of a hypercube operation.

0161The method <b>3100</b> can further comprise receiving a query, determining that the query is related a filter of the plurality of filters, determining, in the data structure, the column associated with the filter, determining, in the data structure, the rows intersecting the column, wherein the intersection of the rows and the column comprise cells containing the first symbol, and retrieving the data values of the plurality of unsorted data values associated with the rows from the one or more databases. The method <b>3100</b> can further comprise sorting the retrieved data values.

0162In an aspect, illustrated in <figref idref="f0032"><b>FIG. 32</b></figref> provided is a method <b>3200</b> comprising determining data values of a plurality of data values stored in one or more databases that satisfy a first threshold, wherein the plurality of data values are unsorted at <b>3202.</b> The method <b>3200</b> can comprise representing, in a first data structure, data values of the plurality of data values that satisfy the first threshold with a first symbol at <b>3204.</b> The method <b>3200</b> can comprise representing, in the first data structure, data values of the plurality of data values that do not satisfy the first threshold with a second symbol at <b>3206.</b> The one or more databases can be derived from a result of a hypercube operation.

0163The method <b>3200</b> can further comprise receiving a query, determining that the query is related to the first threshold, identifying, in the first data structure, the data values of the plurality of data values represented by the first symbol, and retrieving the identified data values from the one or more databases. The method <b>3200</b> can further comprise sorting the retrieved identified data values.

0164The method <b>3200</b> can further comprise determining data values of the plurality of data values stored in the one or more databases that satisfy a second threshold, representing, in a second data structure, data values of the plurality of data values that satisfy the second threshold with the first symbol, and representing, in the second data structure, data values of the plurality of data values that do not satisfy the second threshold with the second symbol. The method <b>3200</b> can further comprise receiving a query, determining that the query is related to the second threshold, identifying, in the second data structure, the data values of the plurality of data values represented by the first symbol, and retrieving the identified data values from the one or more databases. The method <b>3200</b> can further comprise sorting the retrieved identified data values.

0165In an aspect, illustrated in <figref idref="f0033"><b>FIG. 33</b></figref> provided is a method <b>3300</b> comprising applying a first filter to a plurality of unsorted data values stored in one or more databases at <b>3302.</b> The first filter can generate a first data structure that identifies data values of the plurality of unsorted data values with a first symbol or a second symbol. The first symbol and the second symbol can be mapped to respective data values of the plurality of unsorted data values. Applying the first filter to the plurality of unsorted data values can comprise determining if each data value satisfies a first threshold, mapping, in the first data structure, to each respective data value that satisfies the first threshold, the first symbol, and mapping, in the first data structure, to each respective data value that does not satisfy the first threshold, the second symbol.

0166The method <b>3300</b> can comprise applying a second filter to the plurality of unsorted data values stored in the one or more databases at <b>3304.</b> The second filter can generate a second data structure that identifies data values of the plurality of unsorted data values with the first symbol or the second symbol. The first symbol and the second symbol can be mapped to respective data values of the plurality of unsorted data values. Applying the second filter to the plurality of unsorted data values can comprise determining if each data value satisfies a second threshold, mapping, in the second data structure, to each respective data value that satisfies the second threshold, the first symbol, and mapping, in the second data structure, to each respective data value that does not satisfy the second threshold, the second symbol.

0167The method <b>3300</b> can further comprise receiving a query, determining that the query is related to the first filter, identifying, in the first data structure, the data values of the plurality of unsorted data values represented by the first symbol, and retrieving the identified data values from the one or more databases. The method <b>3300</b> can further comprise sorting the retrieved identified data values.

0168In an aspect, a data storage and retrieval system for a computer memory, is described comprising a memory, configured according to a logical table, the logical table comprising, a plurality of logical rows, each said logical row including one or more of a first symbol or a second symbol, wherein the first symbol and the second symbol represent a result of application of a filter to a plurality of unsorted data values stored in one or more databases, each said logical row corresponding to a respective data value of the plurality of unsorted data values, and a plurality of logical columns intersecting said plurality of logical rows to define a plurality of logical cells, each said logical column including a filter identifier to identify each said logical column. The memory is configured for indexing the first symbols and the second symbols stored in said table.

0169In an exemplary aspect, the methods and systems can be implemented on a computer <b>3401</b> as illustrated in <figref idref="f0034"><b>FIG. 34</b></figref> and described below. Similarly, the methods and systems disclosed can utilize one or more computers to perform one or more functions in one or more locations. <figref idref="f0034"><b>FIG. 34</b></figref> is a block diagram illustrating an exemplary operating environment for performing the disclosed methods. This exemplary operating environment is only an example of an operating environment and is not intended to suggest any limitation as to the scope of use or functionality of operating environment architecture. Neither should the operating environment be interpreted as having any dependency or requirement relating to any one or combination of components illustrated in the exemplary operating environment.

0170The present methods and systems can be operational with numerous other general purpose or special purpose computing system environments or configurations. Examples of well-known computing systems, environments, and/or configurations that can be suitable for use with the systems and methods comprise, but are not limited to, personal computers, server computers, laptop devices, and multiprocessor systems. Additional examples comprise set top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments that comprise any of the above systems or devices, and the like.

0171The processing of the disclosed methods and systems can be performed by software components. The disclosed systems and methods can be described in the general context of computer-executable instructions, such as program modules, being executed by one or more computers or other devices. Generally, program modules comprise computer code, routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The disclosed methods can also be practiced in grid-based and distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules can be located in both local and remote computer storage media including memory storage devices.

0172Further, one skilled in the art will appreciate that the systems and methods disclosed herein can be implemented via a general-purpose computing device in the form of a computer <b>3401</b>. The components of the computer <b>3401</b> can comprise, but are not limited to, one or more processors <b>3403</b>, a system memory <b>3412</b>, and a system bus <b>3413</b> that couples various system components including the one or more processors <b>3403</b> to the system memory <b>3412</b>. The system can utilize parallel computing.

0173The system bus <b>3413</b> represents one or more of several possible types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, or local bus using any of a variety of bus architectures. The bus <b>3413</b>, and all buses specified in this description can also be implemented over a wired or wireless network connection and each of the subsystems, including the one or more processors <b>3403</b>, a mass storage device <b>3404</b>, an operating system <b>3405</b>, software <b>3406</b>, data <b>3407</b>, a network adapter <b>3408</b>, the system memory <b>3412</b>, an Input/Output Interface <b>3410</b>, a display adapter <b>3409</b>, a display device <b>3411</b>, and a human machine interface <b>3402</b>, can be contained within one or more remote computing devices <b>3414a</b>,<b>b</b>,<b>c</b> at physically separate locations, connected through buses of this form, in effect implementing a fully distributed system.

0174The computer <b>3401</b> typically comprises a variety of computer readable media. Exemplary readable media can be any available media that is accessible by the computer <b>3401</b> and comprises, for example and not meant to be limiting, both volatile and non-volatile media, removable and non-removable media. The system memory <b>3412</b> comprises computer readable media in the form of volatile memory, such as random access memory (RAM), and/or non-volatile memory, such as read only memory (ROM). The system memory <b>3412</b> typically contains data such as the data <b>3407</b> and/or program modules such as the operating system <b>3405</b> and the software <b>3406</b> that are immediately accessible to and/or are presently operated on by the one or more processors <b>3403</b>.

0175In another aspect, the computer <b>3401</b> can also comprise other removable/non-removable, volatile/non-volatile computer storage media. By way of example, <figref idref="f0034"><b>FIG. 34</b></figref> illustrates the mass storage device <b>3404</b> which can provide non-volatile storage of computer code, computer readable instructions, data structures, program modules, and other data for the computer <b>3401</b>. For example and not meant to be limiting, the mass storage device <b>3404</b> can be a hard disk, a removable magnetic disk, a removable optical disk, magnetic cassettes or other magnetic storage devices, flash memory cards, CD-ROM, digital versatile disks (DVD) or other optical storage, random access memories (RAM), read only memories (ROM), electrically erasable programmable read-only memory (EEPROM), and the like.

0176Optionally, any number of program modules can be stored on the mass storage device <b>3404</b>, including by way of example, the operating system 3405 and the software <b>3406</b>. Each of the operating system <b>3405</b> and the software <b>3406</b> (or some combination thereof) can comprise elements of the programming and the software <b>3406.</b> The data <b>3407</b> can also be stored on the mass storage device <b>3404</b>. The data <b>3407</b> can be stored in any of one or more databases known in the art. Examples of such databases comprise, DB2^®, Microsoft^® Access, Microsoft^® SQL Server, Oracle^®, mySQL, PostgreSQL, and the like. The databases can be centralized or distributed across multiple systems.

0177In an aspect, the software <b>3406</b> can comprise one or more of a script engine, a logical inference engine, a calculation engine, an extension engine, and/or a rendering engine. In an aspect, the software <b>3406</b> can comprise an external engine and/or an interface to the external engine.

0178In another aspect, the user can enter commands and information into the computer <b>3401</b> via an input device (not shown). Examples of such input devices comprise, but are not limited to, a keyboard, pointing device (e.g., a "mouse"), a microphone, a joystick, a scanner, tactile input devices such as gloves, and other body coverings, and the like These and other input devices can be connected to the one or more processors <b>3403</b> via the human machine interface <b>3402</b> that is coupled to the system bus <b>3413</b>, but can be connected by other interface and bus structures, such as a parallel port, game port, an IEEE 1394 Port (also known as a Firewire port), a serial port, or a universal serial bus (USB).

0179In yet another aspect, the display device <b>3411</b> can also be connected to the system bus <b>3413</b> via an interface, such as the display adapter <b>3409</b>. It is contemplated that the computer <b>3401</b> can have more than one display adapter <b>3409</b> and the computer <b>3401</b> can have more than one display device <b>3411</b>. For example, the display device <b>3411</b> can be a monitor, an LCD (Liquid Crystal Display), or a projector. In addition to the display device <b>3411</b>, other output peripheral devices can comprise components such as speakers (not shown) and a printer (not shown) which can be connected to the computer <b>3401</b> via the Input/Output Interface <b>3410</b>. Any step and/or result of the methods can be output in any form to an output device. Such output can be any form of visual representation, including, but not limited to, textual, graphical, animation, audio, tactile, and the like. The display device <b>3411</b> and computer <b>3401</b> can be part of one device, or separate devices.

0180The computer <b>3401</b> can operate in a networked environment using logical connections to one or more remote computing devices <b>3414a,b,c</b>. By way of example, a remote computing device can be a personal computer, portable computer, smartphone, a server, a router, a network computer, a peer device or other common network node, and so on. Logical connections between the computer <b>3401</b> and a remote computing device <b>3414a,b,c</b> can be made via a network <b>3415</b>, such as a local area network (LAN) and/or a general wide area network (WAN). Such network connections can be through the network adapter <b>3408</b>. The network adapter <b>3408</b> can be implemented in both wired and wireless environments. In an aspect, one or more of the remote computing devices <b>3414a,b,c</b> can comprise an external engine and/or an interface to the external engine.

0181For purposes of illustration, application programs and other executable program components such as the operating system <b>3405</b> are illustrated herein as discrete blocks, although it is recognized that such programs and components reside at various times in different storage components of the computing device <b>3401</b>, and are executed by the one or more processors <b>3403</b> of the computer. An implementation of the software <b>3406</b> can be stored on or transmitted across some form of computer readable media. Any of the disclosed methods can be performed by computer readable instructions embodied on computer readable media. Computer readable media can be any available media that can be accessed by a computer. By way of example and not meant to be limiting, computer readable media can comprise "computer storage media" and "communications media." "Computer storage media" comprise volatile and non-volatile, removable and non-removable media implemented in any methods or technology for storage of information such as computer readable instructions, data structures, program modules, or other data. Exemplary computer storage media comprises, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other medium which can be used to store the desired information and which can be accessed by a computer.

0182The methods and systems can employ Artificial Intelligence techniques such as machine learning and iterative learning. Examples of such techniques include, but are not limited to, expert systems, case based reasoning, Bayesian networks, behavior based AI, neural networks, fuzzy systems, evolutionary computation (e.g. genetic algorithms), swarm intelligence (e.g. ant algorithms), and hybrid intelligent systems (e.g. Expert inference rules generated through a neural network or production rules from statistical learning).

0183While the methods and systems have been described in connection with preferred embodiments and specific examples, it is not intended that the scope be limited to the particular embodiments set forth, as the embodiments herein are intended in all respects to be illustrative rather than restrictive. The scope of the invention is defined by the appended claims.

0184Unless otherwise expressly stated, it is in no way intended that any method set forth herein be construed as requiring that its steps be performed in a specific order. Accordingly, where a method claim does not actually recite an order to be followed by its steps or it is not otherwise specifically stated in the claims or descriptions that the steps are to be limited to a specific order, it is in no way intended that an order be inferred, in any respect. This holds for any possible non-express basis for interpretation, including: matters of logic with respect to arrangement of steps or operational flow; plain meaning derived from grammatical organization or punctuation; the number or type of embodiments described in the specification.