Application to Extreme Multi-label Classification? about edgeml HOT 1 CLOSED

Hi,

Thanks for the interest. @AIgen can comment on the C++ ProtoNN for Extreme Classification.
Coming to tensorflow, I think it is as simple as changing the final loss from multi-class to a multi-label loss and train it normally as done right now. You can use all the 3 (ProtoNN, Bonsai and FastGRNN) for extreme classification, but only ProtoNN has been well tested. FastGRNN runs on raw-text and has a 1-vs-All classifier at the end.

Please refer to the extreme classification repository for the datasets, It should be straight forward to change the codes accordingly. But, as the datasets are sparse, tf ProtoNN code might not be efficient on GPU.

Let us know if there is any other issue.

from edgeml.