alphaCSC: Convolution sparse coding for time-series

This is a library to perform shift-invariant sparse dictionary learning, also known as convolutional sparse coding (CSC), on time-series data. It includes a number of different models:

1. univariate CSC
2. multivariate CSC
3. multivariate CSC with a rank-1 constraint[1]
4. univariate CSC with an alpha-stable distribution[2]

A mathematical descriptions of these models is available in the documentation.

Installation

To install this package, the easiest way is using pip. It will install this package and its dependencies. The setup.py depends on numpy and cython for the installation so it is advised to install them beforehand. To install this package, please run one of the two commands:

(Latest stable version)

pip install alphacsc

(Development version)

pip install git+https://github.com/alphacsc/alphacsc.git#egg=alphacsc

(Dicodile backend)

pip install numpy cython
pip install alphacsc[dicodile]

To use dicodile backend, do not forget to set MPI_HOSTFILE environment variable.

If you do not have admin privileges on the computer, use the --user flag with pip. To upgrade, use the --upgrade flag provided by pip.

To check if everything worked fine, you can run:

python -c 'import alphacsc'

and it should not give any error messages.

Quickstart

Here is an example to present briefly the API:

import numpy as np
import matplotlib.pyplot as plt
from alphacsc import BatchCDL

# Define the different dimensions of the problem
n_atoms = 10
n_times_atom = 50
n_channels = 5
n_trials = 10
n_times = 1000

# Generate a random set of signals
X = np.random.randn(n_trials, n_channels, n_times)

# Learn a dictionary with batch algorithm and rank1 constraints.
cdl = BatchCDL(n_atoms, n_times_atom, rank1=True)
cdl.fit(X)

# Display the learned atoms
fig, axes = plt.subplots(n_atoms, 2, num="Dictionary")
for k in range(n_atoms):
    axes[k, 0].plot(cdl.u_hat_[k])
    axes[k, 1].plot(cdl.v_hat_[k])

axes[0, 0].set_title("Spatial map")
axes[0, 1].set_title("Temporal map")
for ax in axes.ravel():
    ax.set_xticklabels([])
    ax.set_yticklabels([])

plt.show()

Dicodile backend

AlphaCSC can use a dicodile-based backend to perform sparse encoding in parallel.

To install dicodile, run pip install alphacsc[dicodile].

Known OpenMPI issues

When self-installing OpenMPI (for instance to run dicodile on a single machine, or for continuous integration), running the dicodile solver might end up causing a deadlock (no output for a long time). It is often due to communication issue between the workers. This issue can often be solved by disabling Docker-related virtual NICs, for instance by running export OMPI_MCA_btl_tcp_if_exclude="docker0".

Bug reports

Use the github issue tracker to report bugs.

Cite our work

If you use this code in your project, please consider citing our work:

[1] Dupré La Tour, T., Moreau, T., Jas, M., & Gramfort, A. (2018). Multivariate Convolutional Sparse Coding for Electromagnetic Brain Signals. Advances in Neural Information Processing Systems (NIPS).

[2] Jas, M., Dupré La Tour, T., Şimşekli, U., & Gramfort, A. (2017). Learning the Morphology of Brain Signals Using Alpha-Stable Convolutional Sparse Coding. Advances in Neural Information Processing Systems (NIPS), pages 1099--1108.