Starting with GLM v1.8.1 fit() passes dropcollinear=true by default.
Lasso uses GLM.fit() to build a nullmodel with unregularized X variables.
When this matrix had zero columns, dropcollinear would check for its rank and get a very large number causing issues.
We now do not dropcollinear when size(X,2)==0.
Because the dropping of collinear unregularized columns is a change in behavior, I am bumping up the version accordingly.
Starting with GLM v1.8.1 fit() passes dropcollinear=true by default. Lasso uses GLM.fit() to build a nullmodel with unregularized X variables. When this matrix had zero columns, dropcollinear would check for its rank and get a very large number causing issues. We now do not dropcollinear when size(X,2)==0.
Because the dropping of collinear unregularized columns is a change in behavior, I am bumping up the version accordingly.