Frequently Asked Questions#

What happened to the amica package on PyPI?#

This project was published as amica up to version 0.1.0. Since 0.2.0 it is published as jamica and installs a top-level jamica module.

pip install jamica

The old name installed a top-level amica module, which collided with amica-python (see below), so the two could not be installed together. The rename removes that conflict. The exported names are unchanged — from jamica import amica, Amica, AmicaConfig works exactly as from amica import ... did.

Is this the same project as amica-python?#

No.

amica-python is an independent implementation of the AMICA algorithm by a different author. Before the rename both projects installed a top-level amica module, so only one could be present in an environment at a time. Since 0.2.0 they no longer share a module name and both can be installed together.

Which Python versions are supported?#

jamica supports Python 3.10 and newer.

Does jamica require JAX?#

No.

jamica runs out of the box using the NumPy backend. Installing the jax extra enables hardware acceleration on supported systems.

pip install "jamica[jax]"

Does jamica require a GPU?#

No.

The NumPy backend runs on any machine. If JAX with GPU support is installed, jamica will automatically use the GPU.

Can I use jamica with MNE-Python?#

Yes.

jamica provides a high-level fit_ica function that returns a standard mne.preprocessing.ICA object, allowing you to use the full MNE visualization and artifact-rejection workflow.

What input format does jamica expect?#

The core API expects a NumPy array with shape

(n_channels, n_samples)

When using the MNE interface, simply pass an mne.io.Raw object.

Can jamica fit multiple ICA models?#

Yes.

AMICA supports fitting multiple ICA models (num_models > 1) to capture non-stationary data. See the examples and API documentation for details.

How does jamica compare with the reference AMICA implementation?#

jamica is designed to reproduce the original AMICA algorithm while providing a native Python implementation, optional JAX acceleration, and seamless integration with the scientific Python ecosystem.

Validation experiments comparing jamica with the Fortran AMICA 1.7 reference are available in the documentation. Agreement is established for single-model configurations against a locally patched build of the reference; multi-model parity is not covered.

Where can I find examples?#

See the Examples section of the documentation for:

  • MNE-Python integration

  • Pure NumPy/JAX workflows

  • Validation examples

  • HPC/Slurm execution

I found a bug. Where should I report it?#

Please open an issue on GitHub:

snesmaeili/jamica#issues

When possible, include:

  • your operating system

  • Python version

  • jamica version

  • backend (NumPy or JAX)

  • a minimal reproducible example

Can I contribute?#

Absolutely!

Please read the Contributing Guide before opening a pull request.