GridSearch and optional parameters

#1

I’m new to h20 and curious about the process of performing a gridsearch on XGBoost.

We’re curious about what happens when we pass in parameters which are only relevant to other parameters.

For example, if we’re using a grid specification like the following:

xgb_params = {  'booster': ['gbtree', 'gblinear', 'dart'],
                'sample_type': ['uniform', 'weighted'],
                'grow_policy': ['depthwise', 'lossguide']}

search_criteria = {'strategy': 'RandomDiscrete'}

xgb_grid = H2OGridSearch(model=H2OXGBoostEstimator,
                         grid_id='xgb_grid',
                         hyper_params=xgb_params,
                         search_criteria=search_criteria)

Does gridsearch automatically know that sample_type is only relevant to dart?

#2

No. You would need to make a list of hyperparam dicts.