MultiTaskLasso and GAT

[kingjr]

@jona-sassenhagen @agramfort @dengemann

On my computer, the MultiTaskLasso from sklearn seems to be unable to run parallel predictions (fit is fine though). I haven't run an investigation yet... Do you encounter the same issue?

import numpy as np
import mne
from mne.datasets import sample
from mne.decoding import GeneralizationAcrossTime
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import make_pipeline
from sklearn.linear_model import MultiTaskLasso
from sklearn.cross_validation import KFold

# preproc
data_path = sample.data_path()
raw_fname = data_path + '/MEG/sample/sample_audvis_filt-0-40_raw.fif'
events_fname = data_path + '/MEG/sample/sample_audvis_filt-0-40_raw-eve.fif'
raw = mne.io.Raw(raw_fname, preload=True)
picks = mne.pick_types(raw.info, meg=True, exclude='bads')  # Pick MEG channels
events = mne.read_events(events_fname)
event_id = {'AudL': 1, 'AudR': 2, 'VisL': 3, 'VisR': 4}
epochs = mne.Epochs(raw, events, event_id, -0.050, 0.600, proj=True,
                    picks=picks, baseline=None, preload=True,
                    reject=dict(mag=5e-12), verbose=False)

# Fit multiple conditions at once
events_side = epochs.events[:, 2] % 2
events_modality = epochs.events[:, 2] > 2
Y = np.vstack((events_side, events_modality)).T

# GAT
clf = make_pipeline(StandardScaler(), MultiTaskLasso(alpha=.01))
gat = GeneralizationAcrossTime(clf=clf, cv=KFold(len(Y)), n_jobs=-1)
gat.fit(epochs, y=Y)
gat.predict(epochs)  # seems to run a single core, and thus takes for ever

[jona-sassenhagen]

Works fine in parallel for me.

[jona-sassenhagen]

gat.predict(epochs)  # seems to run a single core, and thus takes for ever
[Parallel(n_jobs=48)]: Done   1 out of   1 | elapsed:  1.2min remaining:    0.0s
[Parallel(n_jobs=48)]: Done   1 out of   1 | elapsed:  1.2min finished

[agramfort]

no idea

[kingjr]

I updated to sklearn v.17 and it works now.

Thanks!

[kingjr]

Arg, updating it through conda made the problem reappear.. And removing or reinstalling sklearn through pip doesn't change anything anymore. I'll re-open it until I find the fix.

[kingjr]

https://github.com/mne-tools/mne-python/pull/2856 fixes it. I suspect this is the list of list repr and/or the nested lists concatenation that caused the issue. No problem with np.arrays anymore.