[GitHub] yifeim commented on issue #11268: A binary RBM example

[GitBox <gi...@apache.org>]

yifeim commented on issue #11268: A binary RBM example
URL: https://github.com/apache/incubator-mxnet/pull/11268#issuecomment-400214222
 
 
   Very cool implementation! Helped me refresh on the RBM literature. Both test scripts make sense to me.
   
   A few comments:
   * It would be awesome to include some basic mathematical descriptions in the README. I got my intuitions from http://deeplearning.net/tutorial/rbm.html . However, you have a clearer backward function that is not explicit in the tutorial (even though I believe in your intuitions.)
   
   * An alternative idea on the sampler function. I would not compute prob_0 after getting prob_1; instead, to sample from Bernoulli distributions:
   `mx.nd.random.uniform(shape=shape_of_prob_1) < prob_1`
   
   * Would it be possible to track the training process by reporting CD-loss after each epoch? In a similar vein, an additional feature to consider is to include aux in the forward loop for persistent-CD.
   
   * Minor styling ideas: Naming `interaction_weight` instead of `interaction` can receive the extra benefit of default initialization. Also, since the spirit of the commit is a CustomOperator, it may be simpler to use gluon SymbolBlock instead of HybridBlock?

----------------------------------------------------------------
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:
users@infra.apache.org


With regards,
Apache Git Services