dictlearn 0.2.2

Dictionary Learning Toolbox

Dictionary Learning Toolbox

Official documentation here.

This work was supported in part by the Graphomaly Research Grant and is based on the Matlab Dictionary Learning Toolbox that accompanied the Springer book mentioned bellow.

If you use our work in your research, please cite as:

B. Dumitrescu and P. Irofti, Dictionary Learning Algorithms and Applications, Springer, 2018

@book{DL_book,
author    = {Dumitrescu, B. and Irofti, P.},
title     = {Dictionary Learning Algorithms and Applications},
year      = {2018},
publisher = {Springer},
}

Installation and setup

Install via pip from the PyPi repository:

pip install dictlearn

or for the latest changes not yet in the official release:

pip install git+https://gitlab.com/unibuc/graphomaly/dictionary-learning

Usage

The package follows the sklearn API and can be included in your projects via

from dictlearn import DictionaryLearning

which will provide you with a standard scikit-learn estimator that you can use in your pipeline.

Example

import matplotlib.pyplot as plt

from dictlearn import DictionaryLearning
from sklearn.datasets import make_sparse_coded_signal

n_components = 50      # number of atoms
n_features = 20        # signal dimension
n_nonzero_coefs = 3    # sparsity
n_samples = 100        # number of signals
n_iterations = 20      # number of dictionary learning iterations

max_iter = 10
fit_algorithm = "aksvd"
transform_algorithm = "omp"

Y, D_origin, X_origin = make_sparse_coded_signal(
    n_samples=n_samples,
    n_components=n_components,
    n_features=n_features,
    n_nonzero_coefs=n_nonzero_coefs,
    random_state=0
)

dl = DictionaryLearning(
    n_components=n_components,
    max_iter=max_iter,
    fit_algorithm=fit_algorithm,
    transform_algorithm=transform_algorithm,
    n_nonzero_coefs=n_nonzero_coefs,
    code_init=None,
    dict_init=None,
    verbose=False,
    random_state=None,
    kernel_function=None,
    params=None,
    data_sklearn_compat=False
)

dl.fit(Y)

plt.plot(range(max_iter), dl.error_, label=fit_algorithm)
plt.legend()
plt.show()

For configuration and tweaks please consult the documentation.

Development and testing

First clone the repository and change directory to the root of your fresh checkout.

0. Install Prerequisites

Install PyPA’s build:

python3 -m pip install --upgrade build

1. Build

Inside the dictonary-learning directory

python -m build

2. Virtual Environment

Create a virtual environment with Python:

python -m venv venv

Activate the environment:

source venv/bin/activate

For Windows execute instead:

venv\Scripts\activate

3. Install

Inside the virutal environment execute:

pip install dist/dictlearn-*.whl