If input pandas, output should be pandas (i.e. output should have pandas names for columns put back in) about h2o4gpu HOT 6 CLOSED

It's very fundamental to sklearn input/output. If we input pandas and output numpy, would be worse than API inconsistency.

from h2o4gpu.

Can you explain this a little more?

from h2o4gpu.

Ya, like we discussed ... if input was pandas, we should save labels (names) in self and then when return results (like errors or other output that user wants) we should put back the labels and return pandas. That is, numpy in numpy out, but pandas in pandas out.

from h2o4gpu.

👍 to this but is it really necessary for 0.0.4?

from h2o4gpu.

The solution X (should be coef_ for sklearn) should have column names. This is especially important because add a column when doing fit_intercept=True. Predictions should be named as such if pandas. And like we talked a while back, errors should have column names.

from h2o4gpu.

scikit-learn/scikit-learn#5523

Ok, looks like long-term issue that we aren't going to solve. But pandas mostly goes in, and numpy always comes out right now.

from h2o4gpu.