[GENERAL SUPPORT]: dependent_parameters and DerivedParameterConfig are not documented

[chanansh]

Question

In some search problems, the search space is conditioned on a value of another parameter. For example, if the value of param1 is 'A' then param2 is relevant, if param1 is 'B' then param3 is relevant (think of param1 representing different algorithms, for example, see scikit search space definition). It seems like Ax.dev supports this when I searched the API but there is no documentation or example I was able to find.

Please provide any relevant code snippet if applicable.

# example from sklearn
param_grid = [
    {'optimizer': ['sgd'], 'momentum': [0.1, 0.9], 'lr': [0.01]},
    {'optimizer': ['adam'], 'lr': [0.01]} # momentum is absent here
]

Code of Conduct

• I agree to follow this Ax's Code of Conduct