I am not sure why I cannot get close results for an SGD variant of ridge OLS regression, while the same model, absent L2 regularization, can get exact matches with skl
skl