[Frontend][MXNet] ones zeros ones_like zeros_like ops support

[imorinaga]

This PR supports 6 symbols in MXNet frontend.

@srkreddy1238 @nishi-t Could you please do the review. Thanks.

[nishi-t]

@imorinaga Could you rebase against the most recent master?

[imorinaga]

@nishi-t Thank you for comments! rebase done.

[nishi-t]

please add tests for these operators

[yzhliu]

@imorinaga Any updates?

[imorinaga]

argmax and argmin support is deleted.
Output dtype of these are fp32 in MXNet.
However, these are int32 in NNVM, Tensorflow, coreml and onnx.

[nishi-t]

Thank you for adding tests. If the point you have pointed out is a testing problem for argmin and argmax, #1908 may be able to fix it.

[imorinaga]

Thank you for comment!
What I mean is not a testing interface problem.
To support argmax Symbol, I think out_dtype interface is needed on nnvm.symbol.argmax.

[nishi-t]

@imorinaga OK. I don't have sure knowledge of the return type and couldn't confirm it from the document. But, If argmin and argmax return the indices as float32 in mxnet, we would need to discuss how to support it.
BTW, In this PR, do you intend to support only operators excluding argmin and argmax?

[imorinaga]

@nishi-t
Yes, I want to support only ones, ones_like, zeros and zeros_like in this PR. Should I change the PR title?
BTW, I accidentally closed this PR so I've reopened. Sorry for confusing.

[nishi-t]

@imorinaga Ok, Thanks. LGTM

[nishi-t]

@imorinaga yes, I'd suggest changing the title.