MAINT update GitHub workflows

[lostanlen]

A proposed fix for Muawiz's

Looking at making the jenkins and github workflows use pytorch 1.10, 1.11, instead of 1.7, 1.8, would be baller.

Except i think it would be more secure to set the widest version span possible. So i propose:

python-version = [3.6, 3.10]
pytorch-version = [1.7, 1.9, 1.11]

Note that i have remove Python 3.5 (hence the PR name), which has reached EOL and is throwing a DeprecationWarning in the test suite.

I have also removed Torch 1.5, which was released in April 2020.

[lostanlen]

GitHub has parsed my python-version: 3.10 as 3.1. It's going great.

[codecov-commenter]

Codecov Report

Merging #892 (0f9b6b6) into dev (e89ec79) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##              dev     #892   +/-   ##
=======================================
  Coverage   87.94%   87.94%           
=======================================
  Files          64       64           
  Lines        2257     2257           
=======================================
  Hits         1985     1985           
  Misses        272      272           

Flag	Coverage Δ
jenkins_main	87.94% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update e89ec79...0f9b6b6. Read the comment docs.

[lostanlen]

@MuawizChaudhary OK, it's working. Note the new testing matrix:

                python-version: ['3.7', '3.9]
                pytorch-version: ['1.7', '1.9', '1.11']

Any other python-torch pair we want to test?

[lostanlen]

I vote for saying: "Kymatio may work on Python as old as 3.5 and as recent as 3.10, but we make no promises outside of the 3.7--3.9 interval"

[janden]

I'm not sure about this. We're making an assumption here that bugs/features interpolate linearly between versions, which isn't necessarily the case.

I suggest the following matrix:

                python-version: ['3.7', '3.8', '3.9', '3.10']
                pytorch-version: ['1.10', '1.11']

This way we're testing all current Pythons and the most recent PyTorches.

[lostanlen]

I'm not sure if Torch is Python 3.10-ready, but i can try.
For the rest, no problem. On it

[lostanlen]

I have removed torchaudio from our worfklow. We never had a use for it anyway (not even in the tests or examples). I'm hoping that this removal will speed up CI a bit and generally make our lives easier.

[lostanlen]

@MuawizChaudhary changes approved?! but Pip CI does not pass

[MuawizChaudhary]

my approvals are an implied "approval upon successful completion of tests" :)

[lostanlen]

@MuawizChaudhary but i don't know how i can get the tests to pass

 Installing collected packages: iniconfig, zipp, typing-extensions, tomli, pyparsing, py, coverage, attrs, packaging, importlib-metadata, pluggy, pytest, pytest-cov
Successfully installed attrs-21.4.0 coverage-6.4.1 importlib-metadata-4.11.4 iniconfig-1.1.1 packaging-21.3 pluggy-1.0.0 py-1.11.0 pyparsing-3.0.9 pytest-7.1.2 pytest-cov-3.0.0 tomli-2.0.1 typing-extensions-4.2.0 zipp-3.8.0
Collecting https://download.pytorch.org/whl/torch_stable.html
  Downloading https://download.pytorch.org/whl/torch_stable.html (322 kB)
     ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 322.0/322.0 kB 20.9 MB/s eta 0:00:00
  ERROR: Cannot unpack file /tmp/pip-unpack-_h707x8x/torch_stable.html (downloaded from /tmp/pip-req-build-4a_udco5, content-type: text/html); cannot detect archive format
ERROR: Cannot determine archive format of /tmp/pip-req-build-4a_udco5
Error: Process completed with exit code 1.

So clearly this is not ready for merge

[lostanlen]

thanks @janden

it seems like only Torch 1.11 is compatible with Python 3.10 ?

ERROR: Could not find a version that satisfies the requirement torch==1.10.1+cpu (from versions: 1.11.0, 1.11.0+cpu, 1.11.0+cu102, 1.11.0+cu113, 1.11.0+cu115, 1.11.0+rocm4.3.1, 1.11.0+rocm4.5.2)
ERROR: No matching distribution found for torch==1.10.1+cpu
Error: Process completed with exit code 1.

[janden]

thanks @janden

it seems like only Torch 1.11 is compatible with Python 3.10 ?

ERROR: Could not find a version that satisfies the requirement torch==1.10.1+cpu (from versions: 1.11.0, 1.11.0+cpu, 1.11.0+cu102, 1.11.0+cu113, 1.11.0+cu115, 1.11.0+rocm4.3.1, 1.11.0+rocm4.5.2)
ERROR: No matching distribution found for torch==1.10.1+cpu
Error: Process completed with exit code 1.

Makes sense. Probably should just do an include then for that pair so we'd have

                python-version: ['3.7', '3.8', '3.9']
                pytorch-version: ['1.10', '1.11']
                include:
                    - python-version: '3.10'
                      pytorch-version: '1.11'

[lostanlen]

alright, it's finally working. sorry that it took me two days to figure this out. i'm not great at CI
since @MuawizChaudhary has already approved it ... 😆
... @janden merge at your convenience

then we can return to work on the JAX backend

[janden]

Small nitpick (but might as well while we're here): looks like the latest Torch 1.10 is 1.10.2, which does seem to have py310 compatibility. Could you try this quickly? We should also remove the install cases for Torch 1.7 and 1.9 unless we want to include these.

[lostanlen]

I'm surprised i wasn't offered Torch 1.10.2 in this list then?

ERROR: Could not find a version that satisfies the requirement torch==1.10.1+cpu (from versions: 1.11.0, 1.11.0+cpu, 1.11.0+cu102, 1.11.0+cu113, 1.11.0+cu115, 1.11.0+rocm4.3.1, 1.11.0+rocm4.5.2)

Trying with this now:

  if [ ${{ matrix.pytorch-version }} == '1.10' ]; then
                      pip install torch==1.10.2+cpu \
                                  torchvision==0.11.2+cpu \
                                  -f https://download.pytorch.org/whl/torch_stable.html

(note the 1.10.2)

and removing 1.7 and 1.9

[janden]

Judging from the errors, looks like a conflicting dependency for torchvision. Maybe we just let that one loose (i.e., not pinned) and see if it picks a good one.

[janden]

Hmm so according to https://download.pytorch.org/whl/torch_stable.html, torchvision==0.11.3 is available on py310, but not torchvision==0.11.3+cpu. Looks like the oldest torchvision that runs on py310 and CPU is torchvision==0.12.0+cpu.

[janden]

Looks like the issue is the same for torch==1.10.2 (CPU version is not available on py310, only GPU). So let's forget about this. Go back to what you had before.

[janden]

Sorry for all the noise!

[lostanlen]

@janden on y va ?

[janden]

allons-y!