juliusshufan commented on a change in pull request #9793: Enable the reporting 
of cross-entropy or nll loss value when training CNN network using the models 
defined by example/image-classification
URL: https://github.com/apache/incubator-mxnet/pull/9793#discussion_r168378166

 File path: example/image-classification/common/fit.py
 @@ -117,6 +117,8 @@ def add_fit_args(parser):
                        help='load the model on an epoch using the 
     train.add_argument('--top-k', type=int, default=0,
                        help='report the top-k accuracy. 0 means no report.')
+    train.add_argument('--loss', type=str,
+                       help='report the cross-entropy or nll-loss. ce means 
cross-entropy, nll-loss means likelihood loss')
 Review comment:
   @piiswrong @cjolivier01 Thanks for all your comments, and sorry for the 
   Let me try to explain:
   1. My purpose is to report the loss value during training, this is because 
the loss value trends is helpful to monitor the convergence trend.
   2. Per my understanding is, the cross-entropy loss and (negative) likelihood 
loss are most common used for softmax output, and therefore I choose reporting 
either cross-entropy or negative likelihood loss as long as the output layer is 
the softmax.
   The current implementation of my PR is add a string-type argument, that is 
the "--loss", and the corresponding type loss will be reported accordingly.
   Meanwhile, my understanding to Eric's comment is, it be better to set the 
the argument as a list including all the matching loss type (['ce', 'nll-loss'] 
for softmax), and give the user to decide which loss value will be used. 
   May I know which method is acceptable? I'll then update my implementation 
   Thanks for your time.

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:

With regards,
Apache Git Services

Reply via email to