TST: Do not remove fMRI notebook from list of discovered notebooks

[jhlegarreta]

Do not remove fMRI notebook from list of discovered notebooks: this will allow running the notebook at the time scheduled by the notebooks.yml GHA workflow file.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 71.19%. Comparing base (ebfcbd8) to head (e5f6d0f).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #117   +/-   ##
=======================================
  Coverage   71.19%   71.19%           
=======================================
  Files          23       23           
  Lines        1045     1045           
  Branches      121      121           
=======================================
  Hits          744      744           
  Misses        259      259           
  Partials       42       42           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[jhlegarreta]

Looks like SynthStrip is not being found:
https://github.com/jhlegarreta/nifreeze/actions/runs/14897027210/job/41841440531#step:11:102

as it fails at:

copy(bmsk_results.outputs.out_mask, bmask_path)

It also happened when I tried to run the notebook locally. Did not investigate further.

[jhlegarreta]

@oesteban Re: f0569bc not sure where the model needs to be downloaded. Guidance is welcome.

[oesteban]

I just filed an issue reporting what you see nipreps/synthstrip#11

Now, you can configure the model to reside under a FreeSurfer home directory (which can be empty except for the model itself) https://github.com/nipreps/synthstrip/blob/dd29a6c751f6c9cc8a31983d8bfbc82d1d97c4a6/nipreps/synthstrip/wrappers/nipype.py#L38

or, you can set an input on the interface:

nifreeze/docs/notebooks/bold_realignment.ipynb

Lines 78 to 81 in 1cef030

	" bmsk_results = SynthStrip(\n",
	" in_file=str(avg_path),\n",
	" use_gpu=True,\n",
	" ).run(cwd=str(WORKDIR))\n",

Instead of:

        bmsk_results = SynthStrip(
            in_file=str(avg_path),
            use_gpu=True,
        ).run(cwd=str(WORKDIR))

you add a model argument pointing at the file you are downloading

        bmsk_results = SynthStrip(
            in_file=str(avg_path),
            use_gpu=True,
            model='/path/to/synthstripmodel.pt',
        ).run(cwd=str(WORKDIR))

[jhlegarreta]

Re #117 (comment) thanks. Let's see what the CI says:
https://github.com/jhlegarreta/nifreeze/actions/runs/14961970733/job/42025313305

[jhlegarreta]

@oesteban Seems we are still missing something:
https://github.com/jhlegarreta/nifreeze/actions/runs/14963050429/job/42028116883#step:12:124

Suggestions are welcome.

[oesteban]

Suggestions are welcome.

I guess we'll need to close nipreps/synthstrip#11 to print out where the file is being sought

[jhlegarreta]

I must be missing something very evident, but the model weights file is still not being found. The file is being downloaded properly at the testing data location:
https://github.com/jhlegarreta/nifreeze/actions/runs/14984758619/job/42096449224#step:6:632
but the nipreps interface is not being able to load it:
https://github.com/jhlegarreta/nifreeze/actions/runs/14984758619/job/42096449224#step:13:120

@oesteban Sorry, I am at a loss here about what else to check. Suggestions are welcome.

[oesteban]

Okay, now we are getting a different error. Still FileNotFound, but now the interface crashes (as it should) and the not found file is the model. Reading the error it's clear we need to pass a directory not a model to the interface.

Okay you added a specific test before entering the interface. However, the test is failing because it expects a folder.

[jhlegarreta]

This is proving time-consuming and challenging to debug 😔. My last bet would be to convert the notebook to a python script and to have it tested as such in the CI. This fsiled locally because Python didn't like asyncio being outside a function. Did not investigate further.

[jhlegarreta]

@oesteban OK, now errors make sense and I should have realized earlier: the issue is CUDA drivers:

250513-21:05:56,214 nipype.interface INFO:
	 stderr 2025-05-13T21:05:56.214307:    torch._C._cuda_init()
250513-21:05:56,214 nipype.interface INFO:
	 stderr 2025-05-13T21:05:56.214399:RuntimeError: Found no NVIDIA driver on your system. Please check that you have an NVIDIA GPU and installed a driver from http://www.nvidia.com/Download/index.aspx

https://github.com/nipreps/nifreeze/actions/runs/15006593539/job/42166634496?pr=117#step:14:1341

Thoughts:

• AFAIK, GPU access is not offered for free on GHA, so we can:
  • Tell SynthStrip to use the CPU if that will not take too long.
  • Compute the masks locally, upload them to somewhere, download to the appropriate locations, strip that code from the notebook to a python script.

[jhlegarreta]

After discussing this offline, it looks like some commit was missed when contributing the notebook and some cell is missing. I will open a separate PR to:

• Install the necessary parts in the GHA workflow file.
• Uncomment the fMRI notebook from the list of discovered notebooks.
• Make the use_gpu flag depend on whether CUDA drivers have been found.

I will go ahead and merge the new PR as soon as I see that we are stuck in the same point as here. I will open a separate PR to try to install the CUDA drivers so that the notebook can run more quickly. We will then continue here or close this PR.