[GitHub] nikudyshko opened a new issue #12966: Strange behaviour/possible bug with mxnet::cpp::Symbol name argument

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

nikudyshko opened a new issue #12966: Strange behaviour/possible bug with mxnet::cpp::Symbol name argument
URL: https://github.com/apache/incubator-mxnet/issues/12966
 
 
   ## Description 
   In a nutshell: network doesn't train if I change prefix in Symbol name argument from "w" and "b" (lines 38, 39 in example [mlp_cpu.cpp](https://github.com/apache/incubator-mxnet/blob/master/cpp-package/example/mlp_cpu.cpp)) to anything with length greater than 1 char (for example: to "ws"/"bs" or "qwe"/"rty") 
   
   ## Environment info 
   Windows 10 64-bit, Visual Studio 2017 Community Edition (ver. 15.8.4) 
   MXnet built from source (ver. 1.3.0) with cmake (ver. 3.13.0) and VS2017 (ver. 15.8.4) 
   I'm using cpp package 
   
   ## Minimum reproducible example 
   [mlp_cpu.cpp](https://github.com/apache/incubator-mxnet/blob/master/cpp-package/example/mlp_cpu.cpp) 
   
   ## Steps to reproduce 
   1. Change lines 38, 39 to something like this: 
   `   weights[i] = Symbol::Variable("ws" + std::to_string(i));`
   `   biases[i] = Symbol::Variable("bs" + std::to_string(i));` 
   2. Compile and run. 
   
   ## What have you tried to solve it? 
   1. Temporary solution is simple - use only 1 char prefixes (actualy - there are 2 chars, because of string termination char) 
   
   I think, that it somehow related to .InferArgsMap function, but that's not for sure. 
   
   Probably, related to #8126 
   
   

----------------------------------------------------------------
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