lad

Least absolute deviations with L1 regularization using majorization-minimization


Keywords
statistics, optimization
License
MIT
Install
pip install lad==0.1.dev0

Documentation

lad

ci-badge appveyor-badge cov-badge Code Health zenodo-badge

Linear least absolute deviations with L1 regularization.

In estimation theory terms, this is the Maximum A Posterior (MAP) estimator for a Laplacian likelihood with Laplacian prior, i.e.

lad.png

The algorithm yield by the Majorization-Minimization framework turns out to be an iteratively reweighted least-squares. See notes/notes.pdf.

Python Version

To install the development version, proceed as follows:

git clone https://github.com/mirca/lad.git
pip install -e lad

Or install the lastest version on PyPi:

pip install lad

Installation dependencies:

- tensorflow

Test dependencies:

- numpy
- tensorflow
- pytest
- pytest-cov

R version

Inside the R console, type:

devtools::install_github("mirca/lad/r/lad")