Description

In the sklearn.linear_model.LassoLarsIC implementation, it seems that log should not be taken on the mean squared error on line 1482, see here
https://github.com/scikit-learn/scikit-learn/blob/412996f/sklearn/linear_model/least_angle.py#L1482.

If we take the AIC definition and work on likelihood function, the log should be taken on the likelihood function. However, for linear models, AIC simplifies to a Cp type penalty and the log should not be applied. See equation 2.15 in this paper for a reference.