You are viewing a plain text version of this content. The canonical link for it is here.

Posted to commits@mxnet.apache.org by gi...@git.apache.org on 2017/10/01 18:29:06 UTC

[GitHub] jeremiedb opened a new issue #8120: [R] Bug introduced in initializer between 0.10.1 and 0.11.1

jeremiedb opened a new issue #8120: [R] Bug introduced in initializer between 0.10.1 and 0.11.1
URL: https://github.com/apache/incubator-mxnet/issues/8120
 
 
   Issue with MXNet R package, first observed in 0.11.1. Compiled on Ubuntu 16.04 and 14.04. 
   
   The initializers (uniform, normal, Xavier) now return all near 0s. 
   The bugs appear to come from the `mx.runif` and `mx.rnorm` in Random.R, which themselves rely on `mx.nd.internal.sample.uniform` and `mx.nd.internal.sample.normal`. 
   
   The usage of the `random` functions such as `mx.nd.random.uniform` instead of the `sample` functions seems more appropriate for the initialization task. Right now, as all weights are initialized to zeros, no model training is possible. 
   
   I'll open a pull with a quick fix for the initializers. 
 
----------------------------------------------------------------
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