gseqNMF

This package is a re-implementation of the seqNMF algorithm described in Mackevicius et al., 2019, and provided as a MATLAB toolbox here. It utilizes standard sklearn syntax for easy plug-and-play usage.

Features

• Compatible with scikit-learn pipelines
• Significant performance optimizations (benchmarks pending)
• Drop-in GPU acceleration via CuPy
• Comprehensive test suite
• Linted with fully-typed codebase
• Optional visualization module

Installation

The package is available on PyPI and can be installed via pip.

pip install gseqnmf

GPU acceleration can be enabled by installing the package with the cuda12 extra. Development dependencies can be installed with the dev extra.

pip install gseqnmf[cuda12,dev]

Usage Example

import numpy as np
from gseqnmf import GseqNMF

# Load synthetic dataset (samples x neurons)
data = np.load("your_data.npy")
n_components = 3
seqeuence_length = 50
lam = 5e-2
model = GseqNMF(
    n_components=n_components,
    sequence_length=seqeuence_length,
    lam=lam,
)
model.fit(data)

License

This project is licensed under the terms of the MIT license.