Inverse Regression for Text Analysis

Multinomial (inverse) regression inference for text documents and associated attributes. For details see: Taddy (2013 JASA) Multinomial Inverse Regression for Text Analysis and Taddy (2015, AoAS), Distributed Multinomial Regression, . A minimalist partial least squares routine is also included. Note that the topic modeling capability of earlier 'textir' is now a separate package, 'maptpx'.


textir v2.0

This is the textir package for R, implementing the MNIR routines of "multinomial inverse regression for text analysis". It also provides a minimalist partial least squares algorithm.

The cran page is at https://CRAN.R-project.org/package=textir.

Versions 2+ make use of the distrom package, for DMR as in "distributed multinomial regression". These algorithms differ from those in the original MNIR in significant ways: penalties are chosen from full regularization paths (instead of being fixed), via the gamma lasso algorithm as implemented in gamlr (instead of exact log penalties), in parallel for independent Poisson log regressions (instead of jointly for a full multinomial likelihood). It also uses replaces the slam library for sparse simple triplet matrices with the more common Matrix library.

I have kept the mnlm function in textir for backwards compatability, but for simplicity recommend that you use distrom's dmr instead. The argument list is exactly the same (mnlm just calls dmr).

The last pre-2.0 version of textir, matching the implementation in the original MNIR paper, is textir_1.8-8. This is available in archives on the cran page.

News

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.

install.packages("textir")

2.0-5 by Matt Taddy, a year ago


http://taddylab.com


Browse source code at https://github.com/cran/textir


Authors: Matt Taddy <[email protected]>


Documentation:   PDF Manual  


Task views: Natural Language Processing


GPL-3 license


Depends on distrom, gamlr, Matrix, stats, graphics

Suggests MASS


Imported by politeness.

Suggested by distrom.


See at CRAN