INTRODUCTION
This is an implementation of ContraVis - contrastive and visual topic modeling for comparing document collections.
Usage:
perl contravis.pl --num_topics $num_topics
--num_labels $num_labels
--dim $dim
--lambda $lambda
--varphi $varphi
--gamma $gamma
--sigma_0 $sigma_0
--EM_iter $EM_iter
--Quasi_iter $Quasi_iter
--data $data
--output_file $output_file
Arguments:
$num_topics: number of topics
$num_labels: number of labels (include the common label)
$dim: number of dimensions
$lambda: Dirichlet parameter
$varphi: covariance for Gaussian prior of topic coordinates
$gamma: covariance for Gaussian prior of document coordinates
$sigma_0: covariance for Gaussian prior of label coordinates
$EM_iter: number of iterations for EM
$Quasi_iter: maximum iterations of Quasi-Newton
$data: input data
$output_file: output file
- Data: each line is a document containing ids of words. The first $num_labels numbers constitute a 0-1 vector l telling which labels are assigned to that document. l[i] = 1 if document d has label i.