The RandomizedSearchCV returns masked arrays for the hyperparameters, which require some type transformations for further use, e.g. for parallel plotting in this exercise. Is there any simple option to get non-masked arrays?
For example, in this notebook, if we try parallel plotting with model_random_search.cv_results_, we would need to change the types of hyperparameters, e.g. logistic regression parameter C although it initially looked like a real number.
Is there any simple option to get non-masked arrays?
We use a masked array here to manage the combination of parameters that will be failing. Since we cannot easily know in advance the type of each parameter, we are probably using an object dtype to be the most flexible possible.
I am thinking that the easiest way to deal with this type of array would be to use np.nan as a filled value, create a DataFrame, and use the method convert_dtypes that allows to downcast the column of the dtype fo the dataframe.
