Automatic determination of hyperparameters

1. A computer-implemented method comprising:

receiving a request to determine a search space for at least one hyperparameter of a machine learning algorithm;

determining, according to the request, optimal hyperparameter values from the search space for the at least one hyperparameter of the machine learning algorithm based on an evaluation of hyperparameters from the machine learning algorithm on different datasets, wherein the determining comprises:

defining the search space for the at least one hyperparameter, excluding any outliers;

defining a hyper-rectangle by minimizing an upper and lower bound for the defined search space that at least includes hyperparameters from the machine learning algorithm on different datasets;

determining an upper range value and a lowest range value from the hyper-rectangle;

outputting the upper and lowest range values; and

performing one of sampling the upper and lowest range values, or Bayesian optimizing the upper and lowest range values, to determine the optimal hyperparameter values; and

tuning the machine learning algorithm using the optimal hyperparameter values for the at least one hyperparameter of the machine learning algorithm to generate a machine learning model.

2. The computer-implemented method of claim 1 , wherein excluding any outliers comprises excluding any outliers of the hyper-rectangle prior to determining the upper range value and the lowest range value.

3. The computer-implemented method of claim 1 , wherein the hyper-rectangle includes previously derived default range values and user provided hyperparameter values.

4. A computer-implemented method comprising:

receiving a request to determine a search space for at least one hyperparameter of a machine learning algorithm;

determining, according to the request, optimal hyperparameter values from the search space for the at least one hyperparameter of the machine learning algorithm based on an evaluation of hyperparameters from the machine learning algorithm on different datasets, wherein the determining comprises:

defining the search space for the at least one hyperparameter, excluding any outliers;

defining a hyper-ellipsoid that includes hyperparameters from the machine learning algorithm on different datasets; and

performing one of sampling the upper and lowest range values, or Bayesian optimizing the upper and lowest range values, to determine the optimal hyperparameter values; and

tuning the machine learning algorithm using the optimal hyperparameter values for the at least one hyperparameter of the machine learning algorithm to generate a machine learning model.

5. The computer-implemented method of claim 4 , wherein the search space includes previously derived default range hyperparameter values and a user provided search space.

6. The computer-implemented method of claim 4 , further comprising displaying the search space of hyperparameter values.

7. The computer-implemented method of claim 4 , wherein the request includes null values for minimum and maximum values of the at least one hyperparameter, the null values indicating that the minimum and maximum values of the at least one hyperparameter are to be determined from the search space.

8. The computer-implemented method of claim 4 , wherein tuning the machine learning algorithm using the optimal hyperparameter values for the at least one hyperparameter of the machine learning algorithm to generate a machine learning model comprises executing a plurality of training jobs using a training dataset to determine a machine learning model to store.

9. The computer-implemented method of claim 8 , wherein the machine learning model to store is a model that performs the best as measured by a defined objective metric.

10. A system comprising:

a training engine platform to tune a machine learning algorithm implemented by a first one or more electronic devices; and

an optimal hyperparameter selector service implemented by a second one or more electronic devices, the optimal hyperparameter selector service including instructions that upon execution by one or more processors cause the optimal hyperparameter selector service to:

receive a request to determine a search space for at least one hyperparameter of the machine learning algorithm;

determine, according to the request, optimal hyperparameter values from the search space for the at least one hyperparameter of the machine learning algorithm based on an evaluation of hyperparameters from the machine learning algorithm on different datasets, wherein the determining comprises:

defining the search space for the at least one hyperparameter, excluding any outliers;

defining a hyper-rectangle by minimizing an upper and lower bound for the defined search space that at least includes hyperparameters from the machine learning algorithm on different datasets;

determining an upper range value and a lowest range value from the hyper-rectangle;

outputting the upper and lowest range value; and

performing one of sampling the upper and lowest range values, or Bayesian optimizing the upper and lowest range values, to determine the optimal hyperparameter values; and

cause the training engine platform to tune the machine learning algorithm using the optimal hyperparameter values for the at least one hyperparameter of the machine learning algorithm to generate a machine learning model.

11. The system of claim 10 , wherein excluding any outliers comprises excluding any outliers of the hyper-rectangle prior to determining the upper range value and the lowest range value.

12. The system of claim 10 , wherein the search space includes previously derived default range hyperparameter values and a user provided search space.

13. The system of claim 10 , wherein the training engine platform is to use the optimal hyperparameter values for the at least one hyperparameter of the machine learning algorithm to generate a machine learning model by an execution of a plurality of training jobs using a training dataset to determine a machine learning model to store.

14. The system of claim 10 , further comprising a display to display the search space of hyperparameter values.

15. The computer-implemented method of claim 1 , wherein the request includes null values for minimum and maximum values of the at least one hyperparameter, the null values indicating that the minimum and maximum values of the at least one hyperparameter are to be determined from the search space.

16. The computer-implemented method of claim 1 , wherein tuning the machine learning algorithm using the optimal hyperparameter values for the at least one hyperparameter of the machine learning algorithm to generate a machine learning model comprises executing a plurality of training jobs using a training dataset to determine a machine learning model to store.

17. A system comprising:

a training engine platform to tune a machine learning algorithm implemented by a first one or more electronic devices; and

an optimal hyperparameter selector service implemented by a second one or more electronic devices, the optimal hyperparameter selector service including instructions that upon execution by one or more processors cause the optimal hyperparameter selector service to:

receive a request to determine a search space for at least one hyperparameter of the machine learning algorithm;

determine, according to the request, optimal hyperparameter values from the search space for the at least one hyperparameter of the machine learning algorithm based on an evaluation of hyperparameters from the machine learning algorithm on different datasets, wherein the determining comprises:

defining the search space for the at least one hyperparameter, excluding any outliers;

defining a hyper-ellipsoid that includes hyperparameters from the machine learning algorithm on different datasets; and

performing one of sampling the upper and lowest range values, or Bayesian optimizing the upper and lowest range values, to determine the optimal hyperparameter values; and

cause the training engine platform to tune the machine learning algorithm using the optimal hyperparameter values for the at least one hyperparameter of the machine learning algorithm to generate a machine learning model.

18. The system of claim 17 , wherein the request includes null values for minimum and maximum values of the at least one hyperparameter, the null values indicating that the minimum and maximum values of the at least one hyperparameter are to be determined from the search space.