Define log uniform learning_rate parameter in RandomizedSearchCV

Question

I'm doing hyperparameter tuning and I'm using scikit-optimize for Bayesian optimization and RandomizedSearchCV for a randomized search.

In sci-kit optimize I can define the learning_rate easily it like this:

space= [Real(10**-5, 10**0, "log-uniform", name='learning_rate'),
        Integer(1, 20, name='max_depth'),
        ...
       ]

How would I do it with RandomizedSearchCV (sklearn) so that the same "list" of values is used for the optimization?

params_randomSearch = {
              "learning_rate" : TODO,
              "min_samples_leaf": np.arange(1,30,1),
              ..
              }

Answer 1

As per the documentation in RandomizedSearchCV():

param_distributions : dict

Dictionary with parameters names (string) as keys and distributions or 
lists of parameters to try. Distributions must
provide a rvs method for sampling (such as those from
scipy.stats.distributions). If a list is given, it is sampled
uniformly.

Now, Real and Integer in scikit-optimize already implement the rvs() method, so you can use them directly. Just assign them in the dictionary to be used in RandomizedSearchCV.

params_randomSearch = {
      "learning_rate" : Real(10**-5, 10**0, "log-uniform", name='learning_rate'),
      "min_samples_leaf": np.arange(1,30,1),
              ..
              }