Feature contributors and influencers in machine learned predictive models

Patent metadata

Field	Value
Publication number	US-11250340-B2
Application number	US-201715842418-A
Country	US
Kind code	B2
Filing date	Dec 14, 2017
Priority date	Dec 14, 2017
Publication date	Feb 15, 2022
Grant date	Feb 15, 2022

Abstract

In an example, for each feature of one or more features of a target sample data, feature values for one or more pseudo-samples are generated using, localized stratified sampling. The one or more pseudo-samples are fed into the trained machine learned model to obtain their prediction values. A piecewise linear regression model is trained using the one or more pseudo-samples and their prediction values, the piecewise linear regression model having two coefficients for each feature, a first coefficient describing prediction change when a corresponding feature value is increased and a second coefficient describing prediction change when a corresponding feature value is decreased. A top positive feature influencer is identified based on a feature of the one or more features of the target sample having a greatest magnitude of positive first coefficient or greatest magnitude of negative second coefficient. A top negative feature influencer is identified based on a feature of the one or more features of the target sample having a greatest magnitude of negative first coefficient or greatest magnitude of positive second coefficient. A top feature contributor is identified based on a feature of the one or more features of the target sample having a greatest magnitude of a combination of second coefficient and feature value in the target sample data.

First claim

What is claimed is:

1. A system comprising:
a memory; and
a computer-readable medium having instructions stored thereon, which, when executed by a processor, cause the system to:
obtain target sample data, the target sample data having one or more features;
for each feature of the one or more features of the target sample data, generate feature values for one or more pseudo-samples;
feed the one or more pseudo-samples into a trained machine learned model to obtain their prediction values;
train a piecewise linear regression model using the one or more pseudo-samples and their prediction values, the piecewise linear regression model having two coefficients for each feature, a first coefficient describing prediction change when a corresponding feature value is increased and a second coefficient describing prediction change when a corresponding feature value is decreased;
identify a top positive feature influencer based on a feature of the one or more features of the target sample data having a greatest magnitude of a positive first coefficient or a greatest magnitude of a negative second coefficient; and
identify a top negative feature influencer based on a feature of the one or more features of the target sample data having a greatest magnitude of a negative first coefficient or a greatest magnitude of a positive second coefficient.

2. The system of claim 1 , wherein the generating of feature values for the one or more pseudo-samples is based on localized stratified sampling, the localized stratified sampling based on an empirical distribution of feature value and feature value in the target sample data.

3. The system of claim 2 , wherein the instructions further cause the system to identify a top feature contributor based on a feature of the one or more features of the target sample data having a greatest magnitude of a combination of second coefficient and feature value in the target sample data.

4. The system of claim 2 , wherein the generating is further based on a standard deviation.

5. The system of claim 4 , wherein the generating is further based on an interpretable range.

6. The system of claim 5 , wherein the interpretable range is set to 0.5 to cause perturbation to be one half of the standard deviation.

7. The system of claim 1 , wherein the trained machine learned model is to predict likelihood that a member of a social networking service will purchase a particular good or service.

8. A method comprising:
obtaining target sample data, the target sample data having one or more features;
for each feature of the one or more features of the target sample data, generating feature values for one or more pseudo-samples,
feeding the one or more pseudo-samples into a trained machine learned model to obtain their prediction values;
training a piecewise linear regression model using the one or more pseudo-samples and their prediction values, the piecewise linear regression model having two coefficients for each feature, a first coefficient describing prediction change when a corresponding feature value is increased and a second coefficient describing prediction change when a corresponding feature value is decreased;
identifying a top positive feature influencer based on a feature of the one or more features of the target sample data having a greatest magnitude of a positive first coefficient or a greatest magnitude of a negative second coefficient; and
identifying a top negative feature influencer based on a feature of the one or more features of the target sample data having a greatest magnitude of a negative first coefficient or a greatest magnitude of positive second coefficient.

9. The method of claim 8 , wherein the generating of feature values of the one or more pseudo-samples is based on localized stratified sampling, the localized stratified sampling based on an empirical distribution of feature value and feature value in the target sample data.

10. The method of claim 9 , further comprising identifying a top feature contributor based on a feature of the one or more features of the target sample data having a greatest magnitude of a combination of second coefficient and feature value in the target sample data.

11. The method of claim 9 , wherein the generating is further based on a standard deviation.

12. The method of claim 11 , wherein the generating is further based on an interpretable range.

13. The method of claim 12 , wherein the interpretable range is set to 0.5 to cause perturbation to be one half of the standard deviation.

14. The method of claim 8 , wherein the trained machine learned model is to predict likelihood that a member of a social networking service will purchase a particular good or service.

15. A non-transitory machine-readable storage medium comprising instructions which, when implemented by one or more machines, cause the one or more machines to perform operations comprising:
obtaining target sample data, the target sample data having one or more features;
for each feature of the one or more features of the target sample data, generating feature values for one or more pseudo-samples;
feeding the one or more pseudo-samples into a trained machine learned model to obtain their prediction values;
training a piecewise linear regression model using the one or more pseudo-samples and their prediction values, the piecewise linear regression model having two coefficients for each feature, a first coefficient describing prediction change when a corresponding feature value is increased and a second coefficient describing prediction change when a corresponding feature value is decreased;
identifying a top positive feature influencer based on a feature of the one or more features of the target sample data having a greatest magnitude of a positive first coefficient or a greatest magnitude of a negative second coefficient; and
identifying a top negative feature influencer based on a feature of the one or more features of the target sample data having a greatest magnitude of a negative first coefficient or a greatest magnitude of a positive second coefficient.

16. The non-transitory machine-readable storage medium of claim 15 , wherein the generating of feature values for the one or more pseudo-samples is based on localized stratified sampling, the localized stratified sampling based on an empirical distribution of feature value and feature value in the target sample data.

17. The non-transitory machine-readable storage medium of claim 16 , wherein the operations further comprise identifying a top feature contributor based on a feature of the one or more features of the target sample data having a greatest magnitude of a combination of second coefficient and feature value in the target sample data.

18. The non-transitory machine-readable storage medium of claim 16 , wherein the generating is further based on a standard deviation.

19. The non-transitory machine-readable storage medium of claim 18 , wherein the generating is further based on an interpretable range.

20. The non-transitory machine-readable storage medium of claim 19 , wherein the interpretable range is set to 0.5 to cause perturbation to be one half of the standard deviation.

Assignees

• Microsoft Technology Licensing Llc

Inventors

• Yang Jilei
• Di Wei
• Sehgal Nidhi
• Guo Songtao

Classifications

• G06N7/01

  Probabilistic graphical models, e.g. probabilistic networks · CPC title

• G06Q30/0201 Primary

  Market modelling; Market analysis; Collecting market data · CPC title

• G06Q10/40

  Business processes related to social networking or social networking services · CPC title

• G06F18/2115

  by evaluating different subsets according to an optimisation criterion, e.g. class separability, forward selection or backward elimination · CPC title

• G06N5/01

  Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound · CPC title