AMICA & Picard

Why do I need this step?

• AMICA: EEGLAB lists it in the ICA window as AMICA (slowest; best). It can learn more than one ICA model for the same data.
• Picard: a faster way to reach an Infomax or FastICA solution. Its Infomax mode is plain Infomax, not the extended Infomax that runica uses on our path. EEGLAB lists it as Infomax picard.m and FastICA picard.m (fastest).

Both are extensions. The ICA window shows an algorithm only when its function exists on your computer, so AMICA and Picard appear in the list only after you install them.

This lesson is optional. Everything else in the roadmap works with runica.

How do I do it in EEGLAB?

Start with the main dataset D1 main loaded, as in the Starting point.

First: install an extension

1. Open File › then Manage EEGLAB extensions.
2. Search for picard, or for AMICA, select it, and install it. Restart EEGLAB if it asks you to.

Second: Picard from the ICA window

1. Make an ICA copy filtered at 1 Hz, as in the Running ICA lesson.
2. Open Tools › then Decompose data by ICA. In ICA algorithm to use (click to select), choose Infomax picard.m. The Commandline options (See help messages) field shows 'maxiter', 500, 'mode', 'standard'.
3. Add the rank to the options yourself: 'maxiter', 500, 'mode', 'standard', 'pca', 31 in D1. With Picard, pop_runica does not check the rank for you.
4. Press OK, then compare the maps from Plot › then Component maps › then In 2-D with those from runica.

Third (optional): AMICA

1. After installing AMICA, open Tools › then Run AMICA on the ICA copy.
2. Make sure the In File: field is empty. If it shows eeglab_data.fdt, clear it, or AMICA reads the original data from disk instead of the ICA copy. Set # PCA Dims.: to the expected rank, 31 in D1, and choose an output folder you can find.
3. AMICA takes much longer than runica. Leave it running and compare the maps when it finishes.

Whichever algorithm you use, copy the weights to the main dataset the same way as in the 1-Hz filtering lesson.

How do I write it as code?

% The same ICA copy, decomposed by runica and by Picard (Infomax mode)
EEG_ica = pop_eegfiltnew(EEG, 'locutoff', 1);
expectedRank = EEG_ica.nbchan - numel(badChans) - 1;

tic;
EEG_runica = pop_runica(EEG_ica, 'icatype', 'runica', 'extended', 1);
tRunica = toc;

% pop_runica does not reduce the rank for Picard, so pass 'pca' yourself
tic;
EEG_picard = pop_runica(EEG_ica, 'icatype', 'picard', 'maxiter', 500, ...
    'mode', 'standard', 'pca', expectedRank);
tPicard = toc;

fprintf('runica: %.0f s, %d components\n', tRunica, size(EEG_runica.icaweights, 1));
fprintf('Picard: %.0f s, %d components\n', tPicard, size(EEG_picard.icaweights, 1));

% First 10 scalp maps of each, to compare by eye
pop_topoplot(EEG_runica, 0, 1:10);
pop_topoplot(EEG_picard, 0, 1:10);

% AMICA, after installing the AMICA extension; it also needs the rank
% EEG_ica.datfile = '';   % AMICA must use the filtered data in memory, not eeglab_data.fdt on disk
% EEG_amica = pop_runamica(EEG_ica, 'pcakeep', expectedRank, 'outdir', fullfile(pwd, 'amicaout'));

The script times runica and Picard on the same copy and prints the number of components of each: 31 in D1 for both. The AMICA line is a comment, because it runs only after you install the extension.

Which value should I choose?

How do I check it worked?

1. The algorithm appears in ICA algorithm to use (click to select) after you install it.
2. The number of components is the expected rank: 31 in D1 for each algorithm.
3. The maps are similar: the eye component appears in both decompositions, possibly at another position.
4. You noted the time of each algorithm on your computer, from the script.

What mistakes do people make?

• Looking for the algorithm before installing it. The list in Tools › then Decompose data by ICA hides any algorithm whose function is not installed.
• Forgetting the rank with Picard or AMICA. The automatic rank check in pop_runica belongs to runica. With the other two, set the number of components yourself.
• Expecting identical components. Different algorithms, and even two runs of runica, give components in a different order and sign. Compare them by shape.
• AMICA reading the file on disk. If the dataset is linked to a .fdt file, AMICA reads that file, not the filtered data in memory. Empty In File:, or set EEG.datfile = '' in a script.
• Mixing algorithms within one study. Use the same algorithm and settings for every participant, and write them in Methods.

How do I write it in my paper?

A template: edit the data, settings and versions to match what you ran.

Independent component analysis was computed with Picard (Infomax mode, maximum 500 iterations; picard plugin [version], EEGLAB 2026.1; [31] components after PCA reduction to the data rank) on a copy of the data high-pass filtered at 1 Hz, and the resulting weights were applied to the 0.1 Hz high-pass filtered data.

Going further

• EEGLAB plugin: https://eeglab.org/plugins/amica/ (AMICA)
• Picard: https://github.com/pierreablin/picard (Picard)
• Makoto's preprocessing pipeline: https://eeglab.ucsd.edu/wiki/Makoto%27s_preprocessing_pipeline (§13.7 Unofficial instruction to install and use AMICA; §18.8 What does AMICA do?)