Tests break on 1.3.0, part of the reason for the #363 CI failures, when an error traceback escapes into scipy's Cython file _csr_polynomial_exmpansion.pyx.
This is because we relied on _csr_polynomial_features, a non-public piece of the sklearn API that changed. Fortunately, PolynomialFeatures can handle sparse arrays natively, so we can probably remove a lot of the nested conditionals based upon type of sparse matrix.
It is worth noting that with ensembling removed from the feature library, the only role our PolynomialLibrary serves is to allow solely the non-interacting polynomial terms. There is probably a much lower-code version of our lib that can do that.
Tests break on 1.3.0, part of the reason for the #363 CI failures, when an error traceback escapes into scipy's Cython file _csr_polynomial_exmpansion.pyx.
This is because we relied on
_csr_polynomial_features
, a non-public piece of the sklearn API that changed. Fortunately,PolynomialFeatures
can handle sparse arrays natively, so we can probably remove a lot of the nested conditionals based upon type of sparse matrix.It is worth noting that with ensembling removed from the feature library, the only role our
PolynomialLibrary
serves is to allow solely the non-interacting polynomial terms. There is probably a much lower-code version of our lib that can do that.