Support None as input in forward and hybrid_forward?

[yifeim]

Description

(Hybrid) forward supports None as defaults, but not as inputs. I was wondering if there are any fundamental reasons why None cannot be taken as input - if the pattern is consistent and I do not change the computation graph? Similarly, None in the output should also be allowed.

Environment info (Required)

----------Python Info----------
Version      : 3.6.5
Compiler     : GCC 7.2.0
Build        : ('default', 'Apr 29 2018 16:14:56')
Arch         : ('64bit', '')
------------Pip Info-----------
Version      : 10.0.1
Directory    : /home/ec2-user/anaconda3/envs/mxnet_p36/lib/python3.6/site-packages/pip
----------MXNet Info-----------
Version      : 1.3.0
Directory    : /home/ec2-user/anaconda3/envs/mxnet_p36/lib/python3.6/site-packages/mxnet
Commit Hash   : b3be92f4a48bce62a5a8424271871c2f81c8f7f1
----------System Info----------
Platform     : Linux-4.14.77-69.57.amzn1.x86_64-x86_64-with-glibc2.9
system       : Linux
node         : ip-172-16-72-155
release      : 4.14.77-69.57.amzn1.x86_64
version      : #1 SMP Tue Nov 6 21:32:55 UTC 2018
----------Hardware Info----------
machine      : x86_64
processor    : x86_64
----------Network Test----------
Setting timeout: 10
Timing for MXNet: https://github.com/apache/incubator-mxnet, DNS: 0.0016 sec, LOAD: 0.4079 sec.
Timing for Gluon Tutorial(en): http://gluon.mxnet.io, DNS: 0.1175 sec, LOAD: 0.1057 sec.
Timing for Gluon Tutorial(cn): https://zh.gluon.ai, DNS: 0.2767 sec, LOAD: 0.1359 sec.
Timing for FashionMNIST: https://apache-mxnet.s3-accelerate.dualstack.amazonaws.com/gluon/dataset/fashion-mnist/train-labels-idx1-ubyte.gz, DNS: 0.0274 sec, LOAD: 0.1023 sec.
Timing for PYPI: https://pypi.python.org/pypi/pip, DNS: 0.0023 sec, LOAD: 0.3404 sec.
Timing for Conda: https://repo.continuum.io/pkgs/free/, DNS: 0.0081 sec, LOAD: 0.3184 sec.

Package used (Python/R/Scala/Julia):
Python

Error Message:

AssertionError: HybridBlock input must be (nested) list of Symbol or NDArray, but got None of type <class 'NoneType'>
AssertionError: HybridBlock output must be (nested) list of Symbol or NDArray, but got None of type <class 'NoneType'>

Minimum reproducible example

import mxnet as mx
rnn = mx.gluon.rnn.GRU(10)
rnn.hybridize()
rnn.collect_params().initialize()
rnn(mx.nd.zeros((3,2,10)), ) # this is okay
rnn(mx.nd.zeros((3,2,10)), None) # this is not okay

Steps to reproduce

1. The example is self-contained.

What have you tried to solve it?

1. Manually construct an *args list, which seems error-prone.

[zachgk]

@mxnet-label-bot add [Feature Request, Python, Gluon]

[yifeim]

@zachgk Thanks! I should also mention that None output type should also be similarly allowed for completeness. Will edit the original feature request.

[zachgk]

Thank you for the suggestion @yifeim. We welcome contributions, so feel free to try making these changes yourself if you want to get involved

[yifeim]

Thanks for the offers, but I don’t think I need to fix it in a hurry, because I have my workaround. Mxnet for me still seems to have large amount of obvious hacks or bugs, understandably. Nonetheless, I like the better side of mxnet.

I appreciate that you feel this is an sensible suggestion and an obvious thing to do. Will revisit later.

[sxjscience]

@yifeim Should be solved by #16280

[sxjscience]

@yifeim Currently, only supports None as the input.