Reducing computing power for generating test scenario files for decision models
Test scenarios can be generated for decision models on a computer according to some examples. In one example, a system can determine groupings of user input tests for a decision model based on outputs produced by inputting the user input tests into the decision model. Each grouping can include a portion of the user input tests. For each grouping, the system can determine an impact score for each input for each user input test. The system can determine sub-groupings within each grouping based on the impact scores for the inputs. The system can then identify a particular user input test within each sub-grouping. The system can then generate a test scenario file for the decision model based on the particular user input test for use in testing the decision model.
1. A system comprising:
a processor; and
a memory including instructions that are executable by the processor for causing the processor to:
determine a plurality of groupings of a plurality of user input tests used in creating a decision model based on a plurality of outputs produced by inputting the plurality of user input tests into the decision model, each grouping of the plurality of groupings comprising a portion of the plurality of user input tests; and
for each grouping of the plurality of groupings:
determine an impact score for each input of a plurality of inputs for each user input test within the grouping;
determine a sub-grouping of the portion of the plurality of user input tests based on a similarity between inputs with high impact scores;
randomly identify a particular user input test within the sub-grouping; and
generate a test scenario file for the decision model based on the particular user input test for use in validating the decision model.
2. The system of claim 1 , wherein the instructions are further executable by the processor for causing the processor to determine the impact score for each input by:
perturbating the plurality of inputs for each user input test;
inputting the plurality of inputs into the decision model to produce a plurality of perturbated outputs; and
determining the impact score for each input of the plurality of inputs for each user input test based on an amount of variation between the plurality of perturbated outputs.
3. The system of claim 1 , wherein each user input test of the plurality of user input tests comprises an input and an output for validating a rule within the decision model, and wherein the test scenario file comprises a plurality of inputs and an output for validating the decision model.
4. The system of claim 1 , wherein the instructions are further executable by the processor for causing the processor to determine the impact score for each input of the plurality of inputs for each user input test by applying a model-agnostic technique to each input of the plurality of inputs.
5. The system of claim 1 , wherein the instructions are further executable by the processor for causing the processor to determine the plurality of groupings of the plurality of user input tests by:
identifying the portion of the plurality of user input tests with a same output; and
grouping together the portion of the plurality of user input tests with the same output.
6. The system of claim 1 , wherein the instructions are further executable by the processor for causing the processor to:
execute a build of the decision model; and
in response to executing the build of the decision model, executing the test scenario file to validate the decision model by:
inputting the test scenario file into the decision model to produce an output; and
validating that the output matches an expected output included in the test scenario file.
7. The system of claim 1 , wherein the instructions are further executable by the processor for causing the processor to generate the test scenario file using a plurality of user input tests from the plurality of groupings.
8. A method comprising:
determining, by a processor, a plurality of groupings of a plurality of user input tests used in creating a decision model based on a plurality of outputs produced by inputting the plurality of user input tests into the decision model, each grouping of the plurality of groupings comprising a portion of the plurality of user input tests; and
for each grouping of the plurality of groupings:
determining, by the processor, an impact score for each input of a plurality of inputs for each user input test within the grouping;
determining, by the processor, a sub-grouping of the portion of the plurality of user input tests based on a similarity between inputs with high impact scores;
randomly identifying, by the processor, a particular user input test within the sub-grouping; and
generating, by the processor, a test scenario file for the decision model based on the particular user input test for use in validating the decision model.
9. The method of claim 8 , wherein determining the impact score for each input further comprises:
perturbating the plurality of inputs for each user input test;
inputting the plurality of inputs into the decision model to produce a plurality of perturbated outputs; and
determining the impact score for each input of the plurality of inputs for each user input test based on an amount of variation between the plurality of perturbated outputs.
10. The method of claim 8 , wherein each user input test of the plurality of user input tests comprises an input and an output for validating a rule within the decision model, and wherein the test scenario file comprises a plurality of inputs and an output for validating the decision model.
11. The method of claim 8 , wherein determining the impact score for each input of the plurality of inputs for each user input test further comprises applying a model-agnostic technique to each input of the plurality of inputs.
12. The method of claim 8 , wherein determining the plurality of groupings of the plurality of user input tests further comprises:
identifying the portion of the plurality of user input tests with a same output; and
grouping together the portion of the plurality of user input tests with the same output.
13. The method of claim 8 , further comprising:
execute a build of the decision model; and
in response to executing the build of the decision model, executing the test scenario file to validate the decision model by:
inputting the test scenario file into the decision model to produce an output
validating that the output matches an expected output included in the test scenario file.
14. The method of claim 8 , further comprising generating the test scenario file using a plurality of user input tests from the plurality of groupings.
15. A non-transitory computer-readable medium comprising program code that is executable by a processor for causing the processor to:
determine a plurality of groupings of a plurality of user input tests used in creating a decision model based on a plurality of outputs produced by inputting the plurality of user input tests into the decision model, each grouping of the plurality of groupings comprising a portion of the plurality of user input tests; and
for each grouping of the plurality of groupings:
determine an impact score for each input of a plurality of inputs for each user input test within the grouping;
determine a sub-grouping of the portion of the plurality of user input tests based on a similarity between inputs with high impact scores;
randomly identify a particular user input test within the sub-grouping; and
generate a test scenario file for the decision model based on the particular user input test for use in validating the decision model.
16. The non-transitory computer-readable medium of claim 15 , wherein the program code is further executable by the processor for causing the processor to determine the impact score for each input by:
perturbating the plurality of inputs for each user input test;
inputting the plurality of inputs into the decision model to produce a plurality of perturbated outputs; and
determining the impact score for each input of the plurality of inputs for each user input test based on an amount of variation between the plurality of perturbated outputs.
17. The non-transitory computer-readable medium of claim 15 , wherein each user input test of the plurality of user input tests comprises an input and an output for validating a rule within the decision model, and wherein the test scenario file comprises a plurality of inputs and an output for validating the decision model.
18. The non-transitory computer-readable medium of claim 15 , wherein the program code is further executable by the processor for causing the processor to determine the impact score for each input of the plurality of inputs for each user input test by applying a model-agnostic technique to each input of the plurality of inputs.
19. The non-transitory computer-readable medium of claim 15 , wherein the program code is further executable by the processor for causing the processor to determine the plurality of groupings of the plurality of user input tests by:
identifying the portion of the plurality of user input tests with a same output; and
grouping together the portion of the plurality of user input tests with the same output.
20. The non-transitory computer-readable medium of claim 15 , wherein the program code is further executable by the processor for causing the processor to:
execute a build of the decision model; and
in response to executing the build of the decision model, executing the test scenario file to validate an output of the decision model by:
inputting the test scenario file into the decision model to produce an output; and
validating that the output matches an expected output included in the test scenario file.