• The pip version of NIR is not automatically updated and is currently version 1.0.1 from Dec 25. Hence, it does not contain AvgPool2d yet.

• To try this feature you must build NIR from source:
  - git clone https://github.com/neuromorphs/NIR.git
  - cd NIR
  - pip install .

Yep, this would be really practical to have (pls review) :)

There are a couple of conflicts in snntorch/functional/loss.py.

• Line 100: arguments are different; population coding is missing.
• Line 114: weights are now an instance variable; I'm not totally sure if these are inherited properly.

Conflicts in snntorch/functional/loss.py were resolved by merging the changes from master branch to Adding-Conv2D_and_Pooling branch because the original changes for export/import_nir__Adding-Conv2D_and_AvgPool2d were not not intended to affect the loss function logic.

The tests haven't passed as there is now a dependency on sinabs.layers.
I don't think we should add dependencies on external libraries to enable NIR exports to function correctly.
How easy would it be to remove that requirement?

The sinabs.layers dependencies for sumpooling layers were easily removed, as this feature was used for a custom model conversion to NIR and indeed shouldn't be added to the snntorch NIR functions.

Thank you! I ran the tests and it oddly worked with 3.9/3.10, but it didn't work on Python 3.8.
One of the tests throws module 'nir' has no attribute 'AvgPool2d'.

I'm not totally sure why this wouldn't be compatible with 3.8...

@stevenabreu7 : have you observed anything like this among the other libraries?

I would rather not eliminate compatibility with 3.8 over this, but a non-ideal option is to modify the test if we're not sure why this error persists.

I checked the NIR repo and in the pyproject.toml file it indicates in line 35 requires-python = ">=3.9" and I think that's why the tests in snntorch for python 3.8 fails, so I guess if it is the case, then the snntorch compatibility with 3.8 should be eliminated.

thanks for checking that. Yep, let's remove Python 3.8 from the test matrix. I tried to commit to your fork but I don't have the necessary permissions. All that needs to be changed are the following files:

• .github/workflows/build-tag.yml
• CONTRIBUTING.rst
• setup.py

If you update those, the tests should pass and I'll merge the PR.

I removed python 3.8 dependencies from the appointed files.