Skip certain points of the hyperparameter space

Ray Libraries (Data, Train, Tune, Serve) Ray Tune
1

I’m using ray tune to find hyperparameters for a model using cross-validation. One of the hyperparameters modifies how much training data the model receives, as I noticed that sometimes using less training data yielded greater results.

The batch_size should depend on this parameter, because it can’t be higher than the number of elements in the training set. How can I skip running experiments for the hyperparamenters where the batch_size is higher than the number of elements in train?

I don’t want them to even appear in the table of results, or count towards the experiment count. Is that possible?

2

I know i can return {‘done’: True} to STOP the experiment, but is there any way to avoid counting it towards the number of experiments? I want it to stop after 100 valid experiments.