[Bug] reduce_logsumexp examples do not work in 0.11.0rc1 #5291

Activity

1. drpngx commented on Oct 30, 2016

   

   drpngx

   on Oct 30, 2016

   Last edited by drpngx

   Contributor

   More actions

   Nice find! So you mean we should use 0.0 etc? Feel free to send a PR for the docs.
2. drpngx

   added

   type:docs-bugDocument issuesDocument issues

   stat:contribution welcomeStatus - Contributions welcomeStatus - Contributions welcome

   on Oct 30, 2016
3. hoangmit commented on Oct 30, 2016

   

   hoangmit

   on Oct 30, 2016

   ContributorAuthor

   More actions

   No. I mean this is a bug. Currently, tf.reduce_logsumexp(x, 0) crashes. It should return the result in the doc.
4. hoangmit

   changed the title [-]reduce_logsumexp examples do not work in 0.11.0rc1[/-] [+][Bug] reduce_logsumexp examples do not work in 0.11.0rc1[/+] on Oct 30, 2016
5. drpngx commented on Oct 31, 2016

   

   drpngx

   on Oct 31, 2016

   Contributor

   More actions

   Does it crash or return an error?
6. hoangmit commented on Oct 31, 2016

   

   hoangmit

   on Oct 31, 2016

   Last edited by hoangmit

   ContributorAuthor

   More actions

   It returns an error:

   ...lib/python2.7/site-packages/tensorflow/python/ops/math_ops.pyc in reduce_logsumexp(input_tensor, reduction_indices, keep_dims, name)
      1293         keep_dims=True)) + my_max
      1294     if not keep_dims:
   -> 1295       result = array_ops.squeeze(result, reduction_indices)
      1296     return result
      1297 
   
   ...lib/python2.7/site-packages/tensorflow/python/ops/gen_array_ops.pyc in squeeze(input, squeeze_dims, name)
      2644   """
      2645   result = _op_def_lib.apply_op("Squeeze", input=input,
   -> 2646                                 squeeze_dims=squeeze_dims, name=name)
      2647   return result
      2648 
   
   ...lib/python2.7/site-packages/tensorflow/python/framework/op_def_library.pyc in apply_op(self, op_type_name, name, **keywords)
       636         if attr_def.type.startswith("list("):
       637           if not _IsListValue(value):
   --> 638             raise TypeError("Expected list for attr " + key)
       639           if attr_def.has_minimum:
       640             if len(value) < attr_def.minimum:
   
   TypeError: Expected list for attr squeeze_dims
   
7. hoangmit

   closed this as completedon Oct 31, 2016
8. hoangmit

   reopened this on Oct 31, 2016
9. drpngx commented on Oct 31, 2016

   

   drpngx

   on Oct 31, 2016

   Contributor

   More actions

   Oh, you can't send a list directly, you have to construct a tensor, for instance, with zero_initializer.
10. hoangmit commented on Oct 31, 2016

    

    hoangmit

    on Oct 31, 2016

    ContributorAuthor

    More actions

    It is not the problem with the input. (e.g.# 'x' is [[0, 0, 0], [0, 0, 0]])
    I did construct a tensor and pass it in.

    The problem is at line 1295 in file "tensorflow/python/ops/math_ops.py". array_ops.squeeze does not take integer as the 2nd parameter.
11. drpngx commented on Oct 31, 2016

    

    drpngx

    on Oct 31, 2016

    Contributor

    More actions

    Ah, ok, thanks! Feel free to submit a PR with doc changes.
12. hoangmit

    added a commit that references this issue on Oct 31, 2016

    Update tensorflow#5291

    ...

    57a0c7f
13. vrv

    closed this as completedin #5292on Nov 5, 2016
14. vrv

    added a commit that references this issue on Nov 5, 2016

    reduce_logsumexp fix for reduction_indices. Fixes issue #5291 (#5292)

    ...

    4160019
15. dfbassi

    added a commit that references this issue on Nov 14, 2016

    Update from original Nov 14 (#1)

    ...

    cb2ef58