You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@mxnet.apache.org by GitBox <gi...@apache.org> on 2020/03/09 07:40:16 UTC

[GitHub] [incubator-mxnet] huifengguo removed a comment on issue #17728: Poor performance when we use mx.symbol.Variable(.., .., stype="row-sparse") to define the variable

huifengguo removed a comment on issue #17728: Poor performance when we use mx.symbol.Variable(..,..,stype="row-sparse") to define the variable
URL: https://github.com/apache/incubator-mxnet/issues/17728#issuecomment-596374312
 
 
   > Yes, there's no free food. You can still "use Adam and row_sparse variable" with `lazy_update=False` to get identical update, the only benefit you get is less that data is transmitted during communication
   
   OK,we just notice another setting in optimizer, namely whether we set the kvstore when initializing the optimizer.
   ```
   optim = mx.optimizer.create(optimizer, learning_rate=lr, rescale_grad=1.0/batch_size, lazy_update=False)
   mod.init_optimizer(kvstore=store, optimizer=optim)
   ```
   Then, to verity why the loss of accuracy comes from, we conduct the following experimets. There are 3 parameters we can set, so there are 8 different settings. 
   
   ![image](https://user-images.githubusercontent.com/20381178/76191822-328d9500-621b-11ea-965c-39dc932335bd.png)
   
   As the results show, the performance is bad when we choose kvstore and adopt row_sparse variable. Is there anything wrong? Waiting for your reply.
   

----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
 
For queries about this service, please contact Infrastructure at:
users@infra.apache.org


With regards,
Apache Git Services