lasso_view calling form_covariance seems problematic #32
Description
I was comparing the results of highdim and lasso_view with conditioning on subgradients line by line and all matched until producing the covariance between target and score, with the choice of "full_target" . That covariance produced by lasso_view calling form_covariance doesn't match with that from highdim, so I suspect the parametric cov formula from form_covariance might be wrong...