global name 'multiply' is not defined when using LBSGD

[xmfbit]

When I use LBSGD, the error occurs that reminds me

   File "/home/me/anaconda/lib/python2.7/site-packages/mxnet/optimizer.py", line 758, in _l2norm
    norm = multiply(v, v).asnumpy().sum()
NameError: global name 'multiply' is not defined

I add this in the optimizer.py manually and have solved the problem. So is this a bug existing in MXNet?

from .ndarray import multiply

And I want to know whether the LBSGD optimizer is done or not in MXNet. Because I see this in https://github.com/apache/incubator-mxnet/blob/master/python/mxnet/optimizer.py#L734:

    def _get_lbmult(self, nup):
        """Returns lr scaling factor for large batch according to warmup schedule
        (to be implemented)

[kalyc]

Thanks for submitting this issue @xmfbit
@sandeep-krishnamurthy could you add label "Question" to this issue?

[ThomasDelteil]

@xmfbit I confirm there is indeed a bug with the 'lars' warmup strategy. I am not sure the optimizer has been properly tested, as that would have been picked up instantly
@ashokei could you follow up on that? Thanks!

~/anaconda3/lib/python3.6/site-packages/mxnet/optimizer.py in _l2norm(self, v)
    767     def _l2norm(self, v):
    768         "inner product implementation"
--> 769         norm = multiply(v, v).asnumpy().sum()
    770         return norm
    771 

NameError: name 'multiply' is not defined

@indhub can you remove [Question] and add [Bug] ? Thanks!

[vandanavk]

the fix for this has been merged #12005. @xmfbit Please verify at your end too. Thanks

[vandanavk]

@sandeep-krishnamurthy Please close this issue as the PR has been merged.

@xmfbit Please feel free to reopen this issue if you see it again