This PR introduces the hyperparameter search. The current version closely follows the random parameter optimization as implemented in hyperopt package.
This PR is multiprocessing version (non-distributed).
The key optimization logic is in the HyperRunner class and an additional ModelBuilder.hyper_build_model method. The latter should be modified and possibly included in the ModelBuilder.build_model method using conf instead of space dictionary.
HyperRunner class introduces:
keras_fmin_fnctget_space
and
frnn_minimize
The distributed version and API changes to improve user experience will be added in the following PRs
This PR introduces the hyperparameter search. The current version closely follows the random parameter optimization as implemented in
hyperopt
package.This PR is multiprocessing version (non-distributed).
The key optimization logic is in the
HyperRunner
class and an additionalModelBuilder.hyper_build_model
method. The latter should be modified and possibly included in theModelBuilder.build_model
method usingconf
instead ofspace
dictionary.HyperRunner
class introduces:keras_fmin_fnct
get_space
andfrnn_minimize
The distributed version and API changes to improve user experience will be added in the following PRs