FeDis

FeDis is a fast and scalable logistic regression algorithm with data/feature distributed stochastic coordinate descent.

People

FeDis is submitted to the following conference. The supplementary document includes full theoretical proofs omitted from the main paper.

FeDis: Fast and Scalable Logistic Regression with Feature Distributed Stochastic Coordinate Descent
Dongyeop Kang, Woosang Lim, Kijung Shin, U Kang
Submitted to 20th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD'14) [pdf | supplementary document]

FeDis is written in Java and runs on top of Hadoop. The code and executing scripts are here.

Here are some of the real and synthetic dataset referenced in the FeDis paper.

Name	N	D	S (%)	Size (MB)	Description	Download
RCV1	20,242	47,236	0.5	36	An archive of over 800,000 manually categorized newswire stories recently made by Reuters, Ltd.	link
News20	19,996	1,355,191	0.03	137	Size-balanced two-class variant of the UCI 20Newsgroups dataset	link
kdda	8,407,752	19,264,097	0.0002	2,607	KDD Cup 2010 competition data come by Carnigie Learning and DataShop.	link
kddb	20,216,830	29,890,095	0.0001	5,011	KDD Cup 2010 competition data come by Carnigie Learning and DataShop.	link