nipype.algorithms.modelgen.SpecifySPMModel can't handle relative filepaths when being inside a node

[JohannesWiesner]

Summary

This issue is potentially related to #3301. I would like to use nipype.algorithms.modelgen.SpecifySPMModel to set up my design matrix. I noticed that the class itself is able to handle relative filepaths for its argument functional_runs, but not when being inside a node. Is this an expected behavior?

How to replicate the behavior

import nipype.interfaces.matlab as nim
from nilearn.datasets import fetch_spm_auditory
from nilearn.image import concat_imgs
from nipype.algorithms import modelgen
from nipype.interfaces.base import Bunch
from nipype import Node
from nipype.algorithms.modelgen import SpecifySPMModel
import os

mlab = nim.MatlabCommand()
nim.MatlabCommand.set_default_matlab_cmd('/opt/matlab/R2022a/bin/matlab')

# get data (this will automatically create a separate data directory)
subject_data = fetch_spm_auditory(data_dir='./data')

# create .nii file and save to data dir. We pretend this would be two different runs
fmri_img = concat_imgs(subject_data.func)
fmri_img.to_filename('./data/functional2.nii')
fmri_img.to_filename('./data/functional3.nii')

# taken from docs: https://nipype.readthedocs.io/en/latest/api/generated/nipype.algorithms.modelgen.html#specifyspmmodel
s = modelgen.SpecifySPMModel()
s.inputs.input_units = 'secs'
s.inputs.output_units = 'scans'
s.inputs.high_pass_filter_cutoff = 128.
s.inputs.functional_runs = ['./data/functional2.nii', './data/functional3.nii']
s.inputs.time_repetition = 6
s.inputs.concatenate_runs = True
evs_run2 = Bunch(conditions=['cond1'], onsets=[[2, 50, 100, 180]], durations=[[1]])
evs_run3 = Bunch(conditions=['cond1'], onsets=[[30, 40, 100, 150]], durations=[[1]])
s.inputs.subject_info = [evs_run2, evs_run3]
s.run()

# but when we put the same code into a node it will fail
model_specifier = Node(SpecifySPMModel(input_units='secs',
                                       output_units='scans',
                                       high_pass_filter_cutoff=128,
                                       functional_runs = ['./data/functional2.nii', './data/functional3.nii'],
                                       time_repetition = 6,
                                       concatenate_runs = True,
                                       subject_info=[evs_run2, evs_run3]),
                       name='model_specifier')

model_specifier.run()

Actual behavior

model_specifier.run() results in TraitError: Each element of the 'functional_runs' trait of a SpecifySPMModelInputSpec instance must be a list of items which are a pathlike object or string representing an existing file or a pathlike object or string representing an existing file, but a value of '/tmp/tmprbnvs25y/model_specifier/functional2.nii' <class 'str'> was specified.

Expected behavior

As a user I would expect that model_specifier.run() should run successfully just like s.run() , because the only difference between the first and the second case is that the latter is packed inside a Node, everything else is the same. One can fix this by Node(SpecifySPMModel(functional_runs=...) with absolute paths like such:

functional_runs = [os.path.join(os.getcwd(),'./data/functional2.nii'),
                   os.path.join(os.getcwd(),'./data/functional3.nii')]

But as a user I would not expect to do this,

Script/Workflow details

Please put URL to code or code here (if not too long).

Platform details:

{'commit_hash': '%h',
 'commit_source': 'archive substitution',
 'networkx_version': '3.0',
 'nibabel_version': '5.0.1',
 'nipype_version': '1.8.5',
 'numpy_version': '1.23.5',
 'pkg_path': '/home/johannes.wiesner/.conda/envs/csp_wiesner_johannes/lib/python3.9/site-packages/nipype',
 'scipy_version': '1.10.1',
 'sys_executable': '/home/johannes.wiesner/.conda/envs/csp_wiesner_johannes/bin/python',
 'sys_platform': 'linux',
 'sys_version': '3.9.16 | packaged by conda-forge | (main, Feb  1 2023, '
                '21:39:03) \n'
                '[GCC 11.3.0]',
 'traits_version': '6.4.1'}

Execution environment

Choose one

• My python environment outside container

[effigies]

Yes. One of the functions of Nodes is isolation so that your CWD doesn't influence the behavior of tools. You will want to use absolute paths here.

[JohannesWiesner]

@effigies Alright thanks! I wonder if one could adapt the Error message for this case as tmp/tmprbnvs25y/model_specifier/functional2.nii is a symbolic link pointing to an existing file so I couldn't really understanding what's going on.

[JohannesWiesner]

Perhaps one could implement a method within the Node() class that checks if the provided paths are absolute or relative and throw an error for the latter case. One could use os.path.isabs() for example.

[effigies]

Relative paths are frequently used for output filenames. I don't think we could make it an error without breaking that use case.

You could dig through the traceback and find where we make the inputs absolute, and put that in a try/catch block and provide a better error message than the more opaque TraitError.

I should note though that Nipype 1 is in maintenance mode and it will probably die by the time Python 3.11 hits end-of-life. In my opinion, a better use of time would be to work to improve Pydra and write Tasks (which replaces Interfaces and Nodes) to cover your use cases.