Reason for warning

1

Hi,

When I update my PL version to 0.9, I started getting this warning:

python3.8/site-packages/pytorch_lightning/utilities/distributed.py:37: UserWarning: Could not log computational graph since the model.example_input_array attribute is not set or input_array was not given

What is the reason for this warning, and how to disable it?

2

We added a feature that automatically logs your computational graph in tensorboard, e.g:

graphs_computation
graphs_computation1228×520 59.3 KB

This feature is only available if you provide the attribute example_input_array in your model, so it throws a warning if you do not provide it.

Perhaps the default behavior should be to not warn the user, as warning messages can be quite annoying. Let us know what you think!

3

Hi @teddy
Thank you for your response.
This can be useful when the example_input_array is an easy one, like in the case if image classification (the input is only 4-d tensor), but there are some cases where the input examples are very complex to create, and maybe the user is not interested in reading a big graph, especially that tensorboard become laggy when the graph has many component.
In my opinion, it’s better to make this feature optional, if the user is interested in it, he can activate it!

Thanks for the good job!

4

It’s already disabled on the master by default, can you try that?