Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

Set ICA's max_iter to a higher default

See original GitHub issue

Currently, ICA.max_iter is set to 200 by default. This is one of the few MNE defaults I always change when working with data from our lab, and I’ve seen quite a few support requests indicating that ICA didn’t converge within those 200 iterations too, the most recent one here: https://mne.discourse.group/t/cannot-find-ecg-artifacts-during-ica/2778/11?u=richard

I’d like to suggest to increase the max_iter default to 1000 to ensure that non-convergence will really be a rather rare exception for most users. I don’t think there’s many valid use cases for working with non-converged ICA decompositions anyway, or are there? Every time I’ve encountered the warning about non-converged ICA, the decomposition looked … flaky.

WDYT?

cc @cbrnr

Issue Analytics

State:
Created 3 years ago
Comments:9 (9 by maintainers)

Top GitHub Comments

1reaction

hoechenbergercommented, Mar 3, 2021

–> My proposal:

Change the signature of ICA(..., max_iter=200) to ... max_iter=None
fill in the default based on the used algorithm:
- Infomax & picard -> 500
- fastica -> 1000

Needs a deprecation cycle.

Would also be suitable as a task for the sprint if we agree this is a change we’d like to have.

0reactions

hoechenbergercommented, Mar 15, 2021

Perfect!

Top Results From Across the Web

sklearn.decomposition.FastICA

If 'arbitrary-variance' (default), a whitening with variance arbitrary is used. If 'unit-variance', the whitening matrix is rescaled to ensure that each ...

psi1/INPSCF.FORT11 at master · psi4/psi1 · GitHub

C JUST TO SET UP DYNAMICAL ALLOCATION. C. IMPLICIT REAL*8 (A-H,O-Z) ... IF(ICAS.EQ.1)NROOT=2. END IF. IICO=0. IF(IWF.EQ.11)THEN ... DEFAULT 2. C IORDER.

icatb-discuss Mailing List for Group ICA fMRI Toolbox (Page 6)

I insert patients+controls preprocessed images in GIFT. I selected 'remove mean per timepoint' and 'no scaling' in Setupdefaults Tab. I leave default options....

Deep Matrix Approximately Nonlinear Decomposition ... - arXiv

Deep Neural Networks (DNNs) have already become a crucial computational approach to revealing the spatial patterns in the human brain; however, there are....

Package 'fda.usc' - CiteSeerX

Description. This package carries out exploratory and descriptive analysis of functional data exploring its most important features: such as depth ...