objective="multi:softprob" returns a 2d array instead of a 1d array during prediction. This caused issues with predict as it expected a 1d array. This PR fixes this, as well as the handling of BATCH sharding mode during prediction (though I think it still fails on training).
The only change to the public API would be that the callback receives a pandas DataFrame instead of a Series, but I don't think it will cause many issues.
Yard1
commented
3 years ago
objective="multi:softprob"returns a 2d array instead of a 1d array during prediction. This caused issues withpredictas it expected a 1d array. This PR fixes this, as well as the handling of BATCH sharding mode during prediction (though I think it still fails on training).The only change to the public API would be that the callback receives a pandas DataFrame instead of a Series, but I don't think it will cause many issues.
A test was added.