[Relay][TensorFlow] input_0d_mismatch check seems not necessary

1

This special handling seems inherent from NNVM tensorflow frontend, but Relay can use 0-dim tensor.

I try to remove it locally and unit test works fine, should we remove it?

@srkreddy1238 Do you think it still necessary to keep that special handling?

2

yes, I think we can remove from relay in favour of 0-dim support.

cc @alexeyr Please confirm this ?

3

If it’s not necessary, I’m happy. As I commented in the PR, I would be even happier if there were tests for expand_dims :slight_smile:.