Interpretable feature discovery with grammar-based bayesian optimization
A method, a system, and a computer program product for generating an interpretable set of features. One or more search parameters and one or more constraints on one or more search parameters for searching data received from one or more data sources are defined. The data received from one or more data sources is searched using the defined search parameters and constraints. One or more first features are extracted from the searched data. The first features are associated with one or more predictive score values. The searching is repeated in response to receiving a feedback data responsive to the extracted first features. One or more second features resulting from the repeated searching are generated.
1. A computer-implemented method for generating predictive features from a dataset, the method comprising:
accessing, by one or more processors, a set of grammar rules, wherein the set of grammar rules comprise constraints on allowable structures for the predictive features to be generated;
applying, by the one or more processors, the set of grammar rules to a dataset comprising a plurality of data points to generate a plurality of candidate predictive features by transforming a subset of the plurality of data points in accordance with the constraints;
selecting, by the one or more processors, one or more predictive features from the plurality of candidate predictive features based at least in part on a predictive efficacy of each candidate predictive features in forecasting a specified outcome variable; and
outputting, by the one or more processors, the selected one or more predictive features for use in a predictive model,
wherein the set of grammar rules provides interpretability for the generated predictive features by defining the allowable structures for feature construction.
2. The computer-implemented method of claim 1 , wherein the dataset is received from one or more data sources selected from the group consisting of transactional data, time-series data, tradeline data, snapshot data, and any combination thereof.
3. The computer-implemented method of claim 1 , further comprising evaluating the plurality of candidate predictive features using one or more objective functions to determine the predictive efficacy of each candidate predictive feature.
4. The computer-implemented method of claim 1 , wherein the selecting comprises utilizing a Bayesian optimization process to assess the predictive efficacy of each candidate predictive feature.
5. The computer-implemented method of claim 1 , wherein the method further comprises iteratively refining the set of grammar rules based on feedback received regarding the predictive efficacy.
6. The computer-implemented method of claim 1 , wherein the method further comprises:
analyzing an existing score for the predictive model to determine current predictive efficacy of the predictive model in forecasting the specified outcome variable, and
utilizing an objective function that searches for predictive features representing marginal information not captured by the existing score.
7. The computer-implemented method of claim 6 , wherein the method further comprises adjusting the set of grammar rules when the marginal information not captured by the existing score is below a predefined threshold.
8. A system for generating predictive features from a dataset, comprising
at least one programmable processor; and
a non-transient machine-readable medium storing instructions that, when executed by the processor, cause the at least one programmable processor to perform operations comprising:
accessing, by one or more processors, a set of grammar rules, wherein the set of grammar rules comprise constraints on allowable structures for the predictive features to be generated;
applying, by the one or more processors, the set of grammar rules to a dataset comprising a plurality of data points to generate a plurality of candidate predictive features by transforming a subset of the plurality of data points in accordance with the constraints;
selecting, by the one or more processors, one or more predictive features from the plurality of candidate predictive features based at least in part on a predictive efficacy of each candidate predictive features in forecasting a specified outcome variable; and
outputting, by the one or more processors, the selected one or more predictive features for use in a predictive model,
wherein the set of grammar rules provides interpretability for the generated predictive features by defining the allowable structures for feature construction.
9. The system of claim 8 , wherein the dataset is received from one or more data sources selected from the group consisting of transactional data, time-series data, tradeline data, snapshot data, and any combination thereof.
10. The system of claim 8 , wherein the operations further comprises evaluating the plurality of candidate predictive features using one or more objective functions to determine the predictive efficacy of each candidate predictive feature.
11. The system of claim 8 , wherein the selecting comprises utilizing a Bayesian optimization process to assess the predictive efficacy of each candidate predictive feature.
12. The system of claim 8 , wherein the operations further comprise iteratively refining the set of grammar rules based on feedback received regarding the predictive efficacy.
13. The system of claim 8 , wherein the operations further comprise:
analyzing an existing score for the predictive model to determine current predictive efficacy of the predictive model in forecasting the specified outcome variable, and
utilizing an objective function that searches for predictive features representing marginal information not captured by the existing score.
14. The system of claim 13 , wherein the operations further comprise adjusting the set of grammar rules when the marginal information not captured by the existing score is below a predefined threshold.
15. A computer program product, for generating predictive features from a dataset, comprising a non-transient machine-readable medium storing instructions that, when executed by at least one programmable processor, cause the at least one programmable processor to perform operations comprising:
accessing, by one or more processors, a set of grammar rules, wherein the set of grammar rules comprise constraints on allowable structures for the predictive features to be generated;
applying, by the one or more processors, the set of grammar rules to a dataset comprising a plurality of data points to generate a plurality of candidate predictive features by transforming a subset of the plurality of data points in accordance with the constraints;
selecting, by the one or more processors, one or more predictive features from the plurality of candidate predictive features based at least in part on a predictive efficacy of each candidate predictive features in forecasting a specified outcome variable; and
outputting, by the one or more processors, the selected one or more predictive features for use in a predictive model,
wherein the set of grammar rules provides interpretability for the generated predictive features by defining the allowable structures for feature construction.
16. The computer program product of claim 15 , wherein the dataset is received from one or more data sources selected from the group consisting of transactional data, time-series data, tradeline data, snapshot data, and any combination thereof.
17. The computer program product of claim 15 , wherein the operations further comprises evaluating the plurality of candidate predictive features using one or more objective functions to determine the predictive efficacy of each candidate predictive feature.
18. The computer program product of claim 15 , wherein the selecting comprises utilizing a Bayesian optimization process to assess the predictive efficacy of each candidate predictive feature.
19. The computer program product of claim 15 , wherein the operations further comprise iteratively refining the set of grammar rules based on feedback received regarding the predictive efficacy.
20. The computer program product of claim 19 , wherein the operations further comprise:
analyzing an existing score for the predictive model to determine current predictive efficacy of the predictive model in forecasting the specified outcome variable, and
utilizing an objective function that searches for predictive features representing marginal information not captured by the existing score.