AMICA via MNE-Python

This example shows how to fit AMICA through the amica.fit_ica() wrapper.

The wrapper returns a standard mne.preprocessing.ICA object, so the usual MNE methods work as expected:

• plot_components

• plot_sources

• get_sources

• apply

The example uses the MNE sample dataset. To run it on your own data, replace load_raw() with your own loading code.

Imports

from __future__ import annotations

import mne

from amica import fit_ica

Load data

Here we use the MNE sample dataset. You can replace this function with your own data loader, for example:

raw = mne.io.read_raw_fif("your_data_raw.fif", preload=True)
raw = mne.io.read_raw_eeglab("your_data.set", preload=True)
raw = mne.io.read_raw_brainvision("your_data.vhdr", preload=True)

def load_raw() -> mne.io.BaseRaw:
    """Load and minimally preprocess a Raw recording."""
    sample = mne.datasets.sample.data_path()
    raw = mne.io.read_raw_fif(
        sample / "MEG" / "sample" / "sample_audvis_raw.fif",
        preload=True,
    )

    raw.pick("eeg")
    raw.filter(1.0, 40.0)

    return raw


raw = load_raw()

print(
    f"Raw data: {len(raw.ch_names)} EEG channels, "
    f"{raw.n_times} samples @ {raw.info['sfreq']:.0f} Hz"
)

Fit AMICA

fit_ica fits AMICA and returns a standard MNE ICA object.

ica = fit_ica(
    raw,
    n_components=15,
    max_iter=500,
    random_state=42,
)

print(f"Fitted ICA with {ica.n_components_} components")

Inspect sources

Once fitted, the result behaves like any other MNE ICA object.

sources = ica.get_sources(raw)

print(f"Sources shape: {sources.get_data().shape}")

Next steps

In an interactive session, you can use standard MNE visualization and cleaning methods:

ica.plot_components()
ica.plot_sources(raw)
ica.exclude = [0, 1]
ica.apply(raw)