I did exactly that, and it was very close to the actual sklearn output.