Constrained sample selection for training models

1. A system for selecting training set samples, the system comprising:

processing circuitry to:

divide an initial set of samples into a plurality of sets including at least a training universal set and a validation universal set, each sample comprising a plurality of properties and a class from a plurality of classes;

validate each of the plurality of sets meet a condition;

for each selected class in the plurality of classes:

create a plurality of selected class training sets and an all other classes training set from the training universal set based upon the class of each sample;

train a plurality of models, each model trained on one of the selected class training sets and the all other classes training set; and

determine a size of a final selected class training set based upon the plurality of models; and

provide a final training set that includes samples from each of the plurality of classes that includes for each class a number of samples based upon the size of the final selected class training set for each class.

2. The system of claim 1 , wherein the plurality of sets comprise equal representation of each of the plurality of classes.

3. The system of claim 1 , wherein to validate each of the plurality of sets meet the condition the processing circuitry:

determines a mean of a property for each of the plurality of sets; and

determines the mean of each of the plurality of sets is within a predetermined range, a set is accepted only when the mean is within the predetermined range.

4. The system of claim 3 , wherein to validate each of the plurality of sets meet a condition the processing circuitry:

determines a standard deviation of a property for each of the plurality of sets; and

determines the standard deviation of each of the plurality of sets is within a predetermined range, a set is accepted only when the standard deviation is within the predetermined range.

5. The system of claim 4 , wherein to validate each of the plurality of sets meet a condition the processing circuitry determines the samples in each of the plurality of sets each comprise samples of each type of a particular property from a plurality of genders.

6. The system of claim 1 , wherein to divide an initial set of samples into the plurality of sets the processing circuitry:

determines samples from a subject has a property value that is the same over all samples from the subject, the samples from the subject are only in one set of the plurality of sets.

7. The system of claim 1 , wherein to create a plurality of selected class training sets the processing circuitry:

randomly selects samples from the training universal set; and

validates the selected class training set meets the condition.

8. The system of claim 7 , wherein to validate the selected class training set meets the condition the processing circuitry:

calculates a mean of a property for each of the plurality of sets; and

determines the mean of each of the plurality of sets is within a predetermined range, a set s accepted only when the mean is within the predetermined range.

9. The system of claim 8 , wherein to validate the selected class training set meets the condition the processing circuitry:

calculates a standard deviation of property for each of the plurality of sets; and

determines the standard deviation of each of the plurality of sets is within a predetermined range, a set is accepted only when the standard deviation is within the predetermined range.

10. The system of claim 7 , wherein each successive selected class training set comprises a previous selected class training set, and wherein each successive selected class training set is larger than the previous selected class training set.

11. The system of claim 1 , wherein to determine the size of the final selected class training set based upon the trained models the processing circuitry:

determines a sensitivity value of each of the trained models; and

determines a selected model with a smallest, number of samples in the selected class train set where other models with larger select class training sets have a sensitivity value that is not greater than some threshold, wherein the size of the final selected class training set is the size of the selected model.

12. The system of claim 11 , wherein to determine the size of the final selected class training set the processing circuitry:

creates a second plurality of selected class training sets, the size of each of the second plurality of selected class training sets determined from the size of the selected model;

trains a second model on each of the second selected class training sets and the all other classes training set;

determines a second sensitivity value of each of the second trained models; and

determines a second selected model with a second smallest number of samples in the selected class training set where other models with larger select class training sets have a sensitivity value that is not greater than some second threshold, wherein the size of the final selected class training set is the size of the second selected model.

13. A method for selecting training set samples, the method comprising:

dividing an initial set of samples into a plurality of sets including at least a training universal set and a validation universal set, the initial set of samples comprising a plurality of samples, each sample comprising a plurality of properties and a class from a plurality of classes;

validating each of the plurality of sets meet a condition;

for each selected class in the plurality of classes:

creating a plurality of selected class training sets and an all other classes training set from the training universal set based upon the class of each sample, each of the selected class universal training set comprises samples from the selected class, and the all other classes training set comprises samples from classes that are not the selected class;

training a model on each of the selected class training sets and the all other classes training set; and

determining a size of a final selected class training set based upon the trained models; and

providing a final training set that includes samples from each of the plurality of classes that includes for each class a number of samples based upon the size of the final selected class training set for each class.

14. The method of claim 13 , wherein the plurality of sets comprise equal representation of each of the plurality of classes.

15. The method of claim 13 , wherein the validating each of the plurality of sets meet the condition comprises:

determining a mean of a property for each of the plurality of sets; and

determining the mean of each of the plurality of sets is within a predetermined range, a set is accepted only when the mean is within the predetermined range.

16. The method of claim 15 , wherein to validate each of the plurality of sets meet the condition the processing circuitry:

determining a standard deviation of a property for each of the plurality of sets; and

determining the standard deviation of each of the plurality of sets is within a predetermined range, a set is accepted only when the standard deviation is within the predetermined range.

17. The method of claim 16 , wherein the validating each of the plurality of sets meet the condition comprises determining the samples in each of the plurality of sets each comprise samples of each type of a particular property from a plurality of genders.

18. At least one non-transitory computer-readable medium, including instructions, which when executed by a machine, cause the machine to perform operations:

dividing an initial set of samples into a plurality of sets including at least a training universal set and a validation universal set, the initial set of samples comprising a plurality of samples, each sample comprising a plurality of properties and a class from a plurality of classes;

validating each of the plurality of sets meet a condition;

for each selected class in the plurality of classes:

creating a plurality of selected class training sets and an all other classes training set from the training universal set based upon the class of each sample, each of the selected class universal training set comprises samples from the selected class, and the all other classes training set comprises samples from classes that are not the selected class;

training a model on each of the selected class training sets and the all other classes training set; and

determining a size of a final selected class training set based upon the trained models; and

providing a final training set that includes samples from each of the plurality of classes that includes for each class a number of samples based upon the size of the final selected class training set for each class.

19. The at least one non-transitory computer-readable medium of claim 18 , wherein the plurality of sets comprise equal representation of each of the plurality of classes.

20. The at least one non-transitory computer-readable medium of claim 18 , wherein the validating each of the plurality of sets meet the condition comprises:

determining a mean of a property for each of the plurality of sets; and

determining the mean of each of the plurality of sets is within a predetermined range, a set is accepted only when the mean is within the predetermined range.

21. The at least one non-transitory computer-readable medium of claim 20 , wherein to validate each of the plurality of sets meet the condition the processing circuitry:

determining a standard deviation of a property for each of the plurality of sets; and

determining the standard deviation of each of the plurality of sets is within a predetermined range, a set is accepted only when the standard deviation is within the predetermined range.

22. The at least one non-transitory computer-readable medium of claim 21 , wherein the validating each of the plurality of sets meet the condition comprises determining the samples in each of the plurality of sets each comprise samples of each type of a particular property from a plurality of genders.

23. The at least one non-transitory computer-readable medium of claim 21 , wherein the dividing an initial set of samples into plurality of sets comprises determining samples from a subject has a property value that is the same over all samples from the subject, the samples from the subject are only in one set of the plurality of sets.

24. The at least one non-transitory computer-readable medium of claim 21 , wherein the creating a plurality of selected class training sets comprises:

randomly selecting samples from the training universal set; and

validating the selected class training set meets the condition.

25. The at least one non-transitory computer-readable medium of claim 24 , wherein the validating the selected class training set meets the condition comprises:

calculating a mean of a property for each of the plurality of sets and determining the mean of each of the plurality of sets is within a predetermined range, a set is accepted only when the mean is within the predetermined range.