online-x

A bunch of online algorithms for matrix factorization and collaborative filtering (Online algorithm miX for machine learning).

Overview

This repo provides implementation of Passive-Aggresive Gibbs sampling LDA Online Collaborative Topic Regression, see below for deatils.

Quick guide

1. install cmake
2. navigate to the root of the source tree
3. mkdir build && cd build && cmake ..
4. if cmake throw no error, then make.
5. modify run/call.sh which will invoke the program.

Parameter setting

Every parameter is fed into the program in a fixed position. If the parameter is not used in a specific algorithm, it will be ignored. This package only deal with diagonal value of a covariance matrix.

synposis: ./cppWrapper $algo $K $U $V $T $I $J $Jburnin $e $c $a0 $b0 $lambda_u $lambda_v $sigma_u $sigma_v $train_file $test_file $learn_cnt $test_cnt $basedir $test_interval $cdk_file $ofm_method

$algo: this parameter is to set the algorithm to be run. Possible values are ostmr(obi-ctr) osgd-topic-flow(odi-ctr) osgd-topic-fixed(bdi-ctr) gibbs-lda pa-i. For difference of these methods please see our paper at arXiv.

K: length of the latent variable or the length of the feature vector.

U: number of users.

V: number of items.

T: number of vocabulary in text corpus.

I: reserved for limiting iterations.

J: number of iterations for each Jibbs round (including burn-in samples)

JB: number of burn-in samples for each Jibbs round

e: reserved for limiting errors

c: only used when algo is pa-i. The 'step-size' in Passive-Aggresive.

a0: $\alpha_0$, hyper-parameter for word prior distribution.

b0: $\beta_0$, hyper-parameter for topic prior distribution.

lu lv: regularization $\lambda_v$, $\lambda_u$ for CTR (bdi-ctr, obi-ctr). Or $\sigma_r$ (please set these two params as identical values) for obi-ctr.

su sv: $\sigma_u$ $\sigma_v$ for obi-ctr.

train test: training and test filename.

learn_cnt train_cnt: number of tranining and test samples.

base_dir: base directory path for the above two files.

test_interval: test performance every test_interval samples seen.

cdk_file: import pretrained LDA model result for bdi-ctr.

Data format

All IDs count from zero.

train test files should contain each sample in one line seperated by one space: userid itemid score.

A file named _docs file should exist, each line contains a list of wordid represents each word in the corresponding document. If a word appear several times, repeat its wordid.

License (GPL V3)

This program is free software: you can redistribute it and/or modify it under the terms of the GNU General Public License as published by the Free Software Foundation, either version 3 of the License, or (at your option) any later version.

This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with this program. If not, see http://www.gnu.org/licenses/.