Comment by sapiogram

It's only 25% speedup in actual Python code, in domains like data science this won't matter much, because your cpu cycles are mostly spent inside numpy.