Ordinary Least Squares Compute Time

Scikit-learn’s linear_model.LinearRegression() documentation lists:

From the implementation point of view, this is just plain Ordinary Least Squares (scipy.linalg.lstsq) wrapped as a predictor object.

Why might scikit-learn’s implementation seem much faster than calling linalg.lstsq() on its own?

The above chart created using numpy.ones()

It appears scikit will cache results to precalculate features that haven’t changed. If you use random data instead with numpy.random.random_integers():

The results are more comparable.