[GitHub] [incubator-mxnet] sxjscience opened a new issue #17893: [Bug][Numpy] Wrong gradient of np.where

[GitBox]

sxjscience opened a new issue #17893: [Bug][Numpy] Wrong gradient of np.where
URL: https://github.com/apache/incubator-mxnet/issues/17893
 
 
   ## Description
   Example 1: Using np.where(array, array, scalar)
   
   ```python
   import mxnet as mx
   mx.npx.set_np()
   
   a = mx.np.array([1, 0, 1])
   b = mx.np.array([2, 3, 4])
   
   b.attach_grad()
   
   with mx.autograd.record():
       c = mx.np.where(a, b, -1)
       c.backward()
   print(b.grad)
   ```
   Output: [0. 1. 0.]
   
   Example 2: Using np.where(array, array, array)
   ```python
   
   import mxnet as mx
   mx.npx.set_np()
   
   a = mx.np.array([1, 0, 1])
   b = mx.np.array([2, 3, 4])
   
   b.attach_grad()
   
   with mx.autograd.record():
       c = mx.np.where(a, b, mx.np.array([-1, -1, -1]))
       c.backward()
   print(b.grad)
   ```
   Output: [1. 0. 1.]
   
   The second one is correct.
   

----------------------------------------------------------------
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:
us...@infra.apache.org


With regards,
Apache Git Services