xuliuac commented on issue #8025: How to use softlabel in MXNET? URL: https://github.com/apache/incubator-mxnet/issues/8025#issuecomment-332204025 @unrealwill Yes, as you mentioned above,metrics has nothing to do with training. It seems like the mx.symbol.SoftmaxOutput() is support softlabel already and I just use a inappropriate eval metric at beginning. After check the definition ,I think I got your point about using Kullback-Leibler divergence,my task is about video content classification,some video may have two label,such like 'dance' and 'sing', prediction is equally right for both label,so I think in this case cross entropy is good enough for the training.Now I have run the training task for a few hours, it do show the training accuracy increased much faster than use onehot label. I hope the final test accuracy is better than before. (??????)?? ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: [email protected]
With regards, Apache Git Services
