Originally, I used

from sklearn.grid_search import GridSearchCV

to perform gridsearch on KDE, part of the code would look like this:

 grid = GridSearchCV(neighbors.KernelDensity(kernel = KDE_KERNEL),
                    {'bandwidth': bandwidth_range}, n_jobs=-1, cv=4) 

 grid.fit(bandwidth_search_sample)

Recently, the scikit-learn moved the module. It becomes

from sklearn.model_selection import GridSearchCV

The API is still the same, but after I switch to the new model_selection.GridSearchCV, the speed slowed dramatically. For one dataset that I used, it is from 18.4s to 1min 21s.

What is happening?