Feature: Conditional Hyperparameters

[vincent-antaki]

Sometimes the usage of a hyperparameters defined for a step within the pipeline depends on another hyperparameter. We should have a way to know if which hyperparameters are used by a trial and which arent.

e.g. We a ChooseOneStepOf step used to choose a model type, It has a 'choice' hyperparameter. Of its two childrens, ModelA and ModelB, only one of the two model hyperparameters are going to be used (assuming model choice is fixed for every trial). At the moment, we don't make the distinction and, to an AutoML loop, both model hyperparameters are used.

Having this distinction could be very useful both for hyperparameters research strategies and post processing hyperparameters plots.

I'm not entirely sure how that should be implemented. Maybe we should have special classes called "ConditionalHyperparameterSample", "ConditionalHyperparameterSpace" and then we can filter out the unused hyperparams based on the condition when we do a get_hyperparameters and set_hyperparameters operation.

[vincent-antaki]

Just looked at the opened issues and #254 seems to be the exact same suggestion.

[stale[bot]]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs in the next 180 days. Thank you for your contributions.