Defining parameters in randomized-search

[jcbirbuet]

Hi,
When we use randomized-search we not only have to define the sampling distribution, but also provide a range of values for each parameter, right? If this is so, is there any recommended criteria to establish that range of values? Is there any site where we can find the default values that scikit learn uses for each parameter by regressor type?

[glemaitre58]

@jcbirbuet

but also provide a range of values for each parameter, right?

This is not a range, but the actual value to randomly select from.

@jcbirbuet

If this is so, is there any recommended criteria to establish that range of values?

This is tricky. We would like to improve the documentation in this regard. Up to now, there are no guidelines on the value to be set and this is more some tacit data scientist knowledge there.

@jcbirbuet

Is there any site where we can find the default values that scikit learn uses for each parameter by regressor type?

The documentation of scikit-learn will provide this information: API Reference — scikit-learn 1.0.2 documentation