Dimension grouping and reduction for model generation, testing, and documentation

• US 10,417,523 B2
• Filed: 11/07/2017
• Issued: 09/17/2019
• Est. Priority Date: 11/07/2016
• Status: Active Grant

• Current Assignee: SymphonyAI Sensa LLC (Fortive Corp.)
• Original Assignee: Ayasdi Al LLC
• Inventors: Singh, Gurjeet, Horton, Noah, Eakin, Bryce
• Primary Examiner(s): Smith, Paulinho E
• Application Number:
  US15/806,265
• Publication Number:
  US 20180285685A1
• Time in Patent Office: 679 Days
• Field of Search: None
• CPC Class Codes: G06F 16/254 Extract, transform and load [ETL] procedures, e.g. ETL data flows in data warehouses
• CPC Class Codes: G06F 16/9024 Graphs; Linked lists G06F16/9027 takes precedence
• CPC Class Codes: G06F 16/9038 Presentation of query results
• CPC Class Codes: G06F 17/153 Multidimensional correlation or convolution
• CPC Class Codes: G06F 18/2113 by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation
• CPC Class Codes: G06N 7/00 Computing arrangements based on specific mathematical models
• CPC Class Codes: G06Q 10/06375 Prediction of business process outcome or impact based on a proposed change

Abstract

An example method includes receiving analysis data and output indicator, mapping data points from a transposition of the analysis data to a reference space, generating a cover of the reference space, clustering the data points mapped to the reference space using the cover and a metric function to determine each node of a plurality of nodes, for each node, identifying data points that are members to identify similar features, grouping features as being similar to each other based on node(s), for each feature, determining correlation with at least some data associated with the output indicator and generate a correlation score, displaying at least groupings of similar features and displaying the correlation scores, receiving a selection of features, generating a set of models based on selection, determining fit of each generated model to output data and generate a model score, and generating a model recommendation report.

Claims

1. A non-transitory computer readable medium including executable instructions, the instructions being executable by a processor to perform a method, the method comprising:

receiving analysis data and output indicator, the output indicator indicating a subset of data of the analysis data, the analysis data including multiple dimensions associated with data points;

receiving a lens function identifier, a metric function identifier, and a resolution function identifier;

mapping data points from a transposition of the analysis data, to a reference space utilizing a lens function identified by the lens function identifier, the transposition of the analysis data transforming the analysis data such that the features are data points, the mapping of data points being performed by applying the lens functions across dimensions for each data point of the transposition of the analysis data;

generating a cover of the reference space using a resolution function identified by the resolution identifier;

clustering the data points mapped to the reference space using the cover and a metric function identified by the metric function identifier to determine each node of a plurality of nodes of a graph, each node including at least one data point;

for each node, identifying data points that are members of that node to identify similar features;

grouping features that are members of the same node as being similar to each other;

for each feature, determining correlation with at least some of the subset of data of the analysis data and generate a correlation score;

displaying at least a subset of groups that include features that are similar to each other and display the correlation score for each displayed feature;

receiving a selection of a subset of features from the at least the subset of groups;

generating a set of models, each model including at least one of the selection of the subset of features;

determining fit of each generated model to the subset of data of the analysis data and generate a model score; and

generating a report recommending the model with the highest score.

2. The non-transitory computer readable medium of claim 1, the method further comprising receiving a maximum number of features and wherein generating a set of models comprising generating a model for every possible combination of the selection of the subset of features.

3. The non-transitory computer readable medium of claim 1, the method further comprising receiving model parameters and wherein every model that is generated is based on the model parameters.

4. The non-transitory computer readable medium of claim 1, the method further comprising receiving scenario data and applying a selected model from the set of models to the scenario data to generate scenario results.

5. The non-transitory computer readable medium of claim 1, the method further comprising documenting the features of the analysis data, the subset of the groups that include features that are similar to each other, the correlation score for each feature, the selection of the subset of the features, the model with the highest score, and the highest score.

6. The non-transitory computer readable medium of claim 1, the documenting further comprising indicating the features that were not selected and correlation scores of each of the features that were not selected.

7. The non-transitory computer readable medium of claim 1, the method further comprising generating a group score for each of the subset of groups, the group score being based on the correlation score of each of the features that are members of the group, and ordering each of the subset of the groups based on the group score.

8. The non-transitory computer readable medium of claim 1, wherein the metric is one of a set of metrics, and, the method further comprises for each for each metric of a set of metrics:

for each point in the analysis data, determining a point in the data set closest to that particular data point using that particular metric and change a metric score if that particular data point and the point in the data set closest to that particular data point share a same or similar shared characteristic;

comparing metric scores associated with different metrics of the set of metrics;

selecting one or more metrics from the set of metrics based at least in part on the metric score to generate a subset of metrics;

for each metric of the subset of metrics, evaluating at least one metric-lens combination by calculating a metric-lens score based on entropy of shared characteristics across subspaces of a reference map generated by the metric-lens combination;

selecting one or more metric-lens combinations based at least in part on the metric-lens score to generate a subset of metric-lens combinations;

generating topological representations using the received data set, each topological representation being generated using at least one metric-lens combination of the subset of metric-lens combinations, each topological representation including a plurality of nodes, each of the nodes having one or more data points from the data set as members, at least two nodes of the plurality of nodes being connected by an edge if the at least two nodes share at least one data point from the data set as members;

scoring each group within each topological representation based, at least in part, on entropy, to generate a group score for each group; and

scoring each topological representation based on the group scores of each group of that particular topological representation to generate a graph score for each topological representation, wherein the subset of groups that include features that are similar to each other is selected from nodes of the topological representation with the highest score.

9. The non-transitory computer readable medium of claim 1, further comprising building a first partition of subsets of the analysis data, each subset of the first partition containing elements being exclusive of other subsets of the first partition;

computing a first subset score for each subset of the first partition using a scoring function based on the subset of the data of the analysis data;

generating a next partition including all of the elements of the first partition, the next partition including at least one subset that includes the elements of two or more subsets of the first partition, each particular subset of the next partition being related to one or more subsets of a previously generated partition if that particular subset shares membership of at least one element with the one or more subsets of the previously generated partition;

computing a second subset score for each subset of the next partition using the scoring function;

defining a max score for each particular subset of the next partition using a max score function, each max score being based on maximal subset scores of that particular subset of the next partition and at least the subsets of the first partition related to that particular subset;

selecting output subsets from all subsets of the next partition and the previously generated partitions including the first partition, the output subsets together including all elements of the first partition, selection of each of the output subsets being made, at least in part, using a maximum score of previously computed subset scores, the maximum score being a largest score of all subset scores of the next partition and previously generated partitions including the first partition; and

wherein the subset of groups that include features that are similar to each other is selected from nodes of an output partition containing the output subsets, the output subsets of the output partition being associated with the received analysis data, each subset of the output partition containing elements being exclusive of other subsets of the output partition.

10. A method comprising:

receiving analysis data and output indicator, the output indicator indicating a subset of data of the analysis data, the analysis data including multiple dimensions associated with data points;

receiving a lens function identifier, a metric function identifier, and a resolution function identifier;

mapping data points from a transposition of the analysis data, to a reference space utilizing a lens function identified by the lens function identifier, the transposition of the analysis data transforming the analysis data such that the features are data points, the mapping of data points being performed by applying the lens functions across dimensions for each data point of the transposition of the analysis data;

generating a cover of the reference space using a resolution function identified by the resolution identifier;

clustering the data points mapped to the reference space using the cover and a metric function identified by the metric function identifier to determine each node of a plurality of nodes of a graph, each node including at least one data point;

for each node, identifying data points that are members of that node to identify similar features;

grouping features that are members of the same node as being similar to each other;

for each feature, determining correlation with at least some of the subset of data of the analysis data and generate a correlation score;

displaying at least a subset of groups that include features that are similar to each other and display the correlation score for each displayed feature;

receiving a selection of a subset of features from the at least the subset of groups;

generating a set of models, each model including at least one of the selection of the subset of features;

determining fit of each generated model to the subset of data of the analysis data and generate a model score; and

generating a report recommending the model with the highest score.

11. The method of claim 10, the method further comprising receiving a maximum number of features and wherein generating a set of models comprising generating a model for every possible combination of the selection of the subset of features.

12. The method of claim 10, the method further comprising receiving model parameters and wherein every model that is generated is based on the model parameters.

13. The method of claim 10, the method further comprising receiving scenario data and applying a selected model from the set of models to the scenario data to generate scenario results.

14. The method of claim 10, the method further comprising documenting the features of the analysis data, the subset of the groups that include features that are similar to each other, the correlation score for each feature, the selection of the subset of the features, the model with the highest score, and the highest score.

15. The method of claim 10, the documenting further comprising indicating the features that were not selected and correlation scores of each of the features that were not selected.

16. The method of claim 10, the method further comprising generating a group score for each of the subset of groups, the group score being based on the correlation score of each of the features that are members of the group, and ordering each of the subset of the groups based on the group score.

17. The method of claim 10, wherein the metric is one of a set of metrics, and, the method further comprises for each for each metric of a set of metrics:

for each point in the analysis data, determining a point in the data set closest to that particular data point using that particular metric and change a metric score if that particular data point and the point in the data set closest to that particular data point share a same or similar shared characteristic;

comparing metric scores associated with different metrics of the set of metrics;

selecting one or more metrics from the set of metrics based at least in part on the metric score to generate a subset of metrics;

for each metric of the subset of metrics, evaluating at least one metric-lens combination by calculating a metric-lens score based on entropy of shared characteristics across subspaces of a reference map generated by the metric-lens combination;

selecting one or more metric-lens combinations based at least in part on the metric-lens score to generate a subset of metric-lens combinations;

generating topological representations using the received data set, each topological representation being generated using at least one metric-lens combination of the subset of metric-lens combinations, each topological representation including a plurality of nodes, each of the nodes having one or more data points from the data set as members, at least two nodes of the plurality of nodes being connected by an edge if the at least two nodes share at least one data point from the data set as members;

scoring each group within each topological representation based, at least in part, on entropy, to generate a group score for each group; and

scoring each topological representation based on the group scores of each group of that particular topological representation to generate a graph score for each topological representation, wherein the subset of groups that include features that are similar to each other is selected from nodes of the topological representation with the highest score.

18. The method of claim 10, further comprising building a first partition of subsets of the analysis data, each subset of the first partition containing elements being exclusive of other subsets of the first partition;

computing a first subset score for each subset of the first partition using a scoring function based on the subset of the data of the analysis data;

generating a next partition including all of the elements of the first partition, the next partition including at least one subset that includes the elements of two or more subsets of the first partition, each particular subset of the next partition being related to one or more subsets of a previously generated partition if that particular subset shares membership of at least one element with the one or more subsets of the previously generated partition;

computing a second subset score for each subset of the next partition using the scoring function;

defining a max score for each particular subset of the next partition using a max score function, each max score being based on maximal subset scores of that particular subset of the next partition and at least the subsets of the first partition related to that particular subset;

selecting output subsets from all subsets of the next partition and the previously generated partitions including the first partition, the output subsets together including all elements of the first partition, selection of each of the output subsets being made, at least in part, using a maximum score of previously computed subset scores, the maximum score being a largest score of all subset scores of the next partition and previously generated partitions including the first partition; and

wherein the subset of groups that include features that are similar to each other is selected from nodes of an output partition containing the output subsets, the output subsets of the output partition being associated with the received analysis data, each subset of the output partition containing elements being exclusive of other subsets of the output partition.

19. A system comprising:

a processor;

a memory including instructions to configure the processor to;

receive analysis data and output indicator, the output indicator indicating a subset of data of the analysis data, the analysis data including multiple dimensions associated with data points;

receive a lens function identifier, a metric function identifier, and a resolution function identifier;

map data points from a transposition of the analysis data, to a reference space utilizing a lens function identified by the lens function identifier, the transposition of the analysis data transforming the analysis data such that the features are data points, the mapping of data points being performed by applying the lens functions across dimensions for each data point of the transposition of the analysis data;

generate a cover of the reference space using a resolution function identified by the resolution identifier;

cluster the data points mapped to the reference space using the cover and a metric function identified by the metric function identifier to determine each node of a plurality of nodes of a graph, each node including at least one data point;

for each node, identify data points that are members of that node to identify similar features;

group features that are members of the same node as being similar to each other;

for each feature, determine correlation with at least some of the subset of data of the analysis data and generate a correlation score;

display at least a subset of groups that include features that are similar to each other and display the correlation score for each displayed feature;

receive a selection of a subset of features from the at least the subset of groups;

generate a set of models, each model including at least one of the selection of the subset of features;

determine fit of each generated model to the subset of data of the analysis data and generate a model score; and

generate a report recommending the model with the highest score.