Installation

For now, AMICA-Python should be installed from source, and you will have to manually install PyTorch (see below) yourself.

Clone the repository, and install it using pip:

git clone https://github.com/scott-huberty/amica-python.git
cd amica-python
pip install -e .

Important

AMICA-Python requires PyTorch. Please install PyTorch by following the instructions at pytorch.org before using AMICA-Python.

Installing PyTorch

To install PyTorch, please follow the instructions at pytorch.org:

Pip

pip install torch

Conda

conda install -c conda-forge pytorch

If you wish, you can install the CPU-only version of PyTorch:

Pip

pip install torch --index-url https://download.pytorch.org/whl/cpu

Conda

conda install -c conda-forge pytorch cpuonly

If you use UV, you can install Pytorch at the same time as AMICA-Python:

Torch with GPU support

uv pip install -e ".[torch-cuda]"

CPU-only Torch

uv pip install -e ".[torch-cpu]"

GPU-accelerated

Once PyTorch with CUDA support is installed, you can run AMICA on GPU:

import numpy as np
from scipy import signal
from amica import AMICA


rng = np.random.default_rng(0)
n_samples = 2000
time = np.linspace(0, 8, n_samples)

s1 = np.sin(2 * time)                     # Sinusoidal
s2 = np.sign(np.sin(3 * time))            # Square wave
s3 = signal.sawtooth(2 * np.pi * time)    # Sawtooth

S = np.c_[s1, s2, s3]
S += 0.2 * rng.standard_normal(S.shape)   # Add noise
S /= S.std(axis=0)                        # Standardize

A = np.array([[1, 1, 1],
            [0.5, 2, 1.0],
            [1.5, 1.0, 2.0]])           # Mixing matrix

X = S @ A.T                               # Observed mixtures

ica = AMICA(random_state=0, device="cuda")  # default is "cpu"
X_new = ica.fit_transform(X)