LGTM. Let's wait for appveyor to complete.

Unfortunately it can still randomly fail even with this fix:

https://ci.appveyor.com/project/amueller/scikit-learn/build/1.0.1025/job/jliou8eykp4k911m

So this is not fixing the original issue but it's a good practice to fix the precision level we expect from the input data and not rely on possibly platform specific defaults. Merging.

If one would like to fix those over all sklearn, it's easy to find them.

git grep dtype=np.float,
sklearn/cluster/hierarchical.py:    inertia = np.empty(len(coord_row), dtype=np.float, order='C')
sklearn/cluster/hierarchical.py:        ini = np.empty(n_additions, dtype=np.float, order='C')
sklearn/linear_model/ridge.py:        X, y = check_X_y(X, y, ['csr', 'csc', 'coo'], dtype=np.float,
sklearn/preprocessing/data.py:                 dtype=np.float, sparse=True, handle_unknown='error'):

git grep "dtype=np.float[^36]" rather! And here's a PR for that - #5375

They are many more than I thought :)