hugorichard / fastsrm Goto Github PK

https://github.com/hugorichard/FastSRM/blob/master/fastsrm/check_inputs.py#L424
This contradicts the logic of SRM, that does not require a fixed spatial layout across individuals.
@hugorichard could we discuss about that ?

I tried to use add_subjects() to generate subject-level spatial maps corresponding to temporal components on a FastSRM model that was defined but not fit. This is the error traceback I get:

TypeError Traceback (most recent call last)
in
----> 1 fast_srm_ica.add_subjects(sub_movie, shared_resp_ica)

~/FastSRM/fastsrm/fastsrm.py in add_subjects(self, imgs, shared_response)
1752 else:
1753 path = os.path.join(
-> 1754 self.temp_dir, "basis_%i" % (len(self.basis_list) + i))
1755 np.save(path, basis_i)
1756 basis.append(path + ".npy")

TypeError: object of type 'NoneType' has no len()

add_subjects() works well if I use it on a model that has been fit previously.

When run on
`
X = [np.random.rand(10, 12) for _ in range(2)]
Xnew = [np.random.rand(10, 12)]
srm = FastSRM(n_components=12, n_iter=10, verbose=True,)
srm.fit(X)
S = srm.transform(X)
print(S.shape)
srm.add_subjects(Xnew, S)

`
The fit does not complain and an error is thrown on add_subjects.
This is missleadind. An error should be thrown at fit time when n_components > min(n_features, n_samples).

This is because check_imgs is applied in transform() (line 1400) and raises error, but in fact these checks are not needed here (at least for time dimension).

When adding fastSRM as an fmralign dependency, we see a build failure. Local installation confirms this is a deprecation from pip:

DEPRECATION: fastsrm 0.0.4 has a non-standard dependency specifier scipy>=0.18.0numpy>=1.12scikit-learn>=0.23joblib>=1.1.0matplotlib>=2.0.0pytest>=6.2.5. pip 23.3 will enforce this behaviour change. A possible replacement is to upgrade to a newer version of fastsrm or contact the author to suggest that they release a version with a conforming dependency specifiers. Discussion can be found at pypa/pip#12063

We can correct this by adding delimiters in fastSRM's install_requires.

I'll open a PR for this, but please let me know how you'd like to handle longer term maintenance, @hugorichard ! i.e., should we vendor the relevant code within fmralign ? Any thoughts are appreciated !