Parsimonious Topic Model
For details of the algorithm, please check the paper, Hossein Soleimani and David J. Miller, "Parsimonious Topic Models with Salient Word Discovery", arXiv:1401.6169.
(C) Copyright 2014, Hossein Soleimani David J. Miller
This program is free program; 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 2 of the License, or any later version. This program is distributed in the hope that it will be useful, but WITHOUT ANY WARRANTY; without even he implied warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.
Compile the program in Linux-based system. type: make
--task training/test, (default: training ) --num_topics number of topics --directory directory to save the output --corpus corpus file, in lda-c format; i.e. each line is of the form [L] [term_1]:[count] ... [term_L]:[count] where L is the number of unique terms in the document, and the [count] associated with each term is the number of times that term appears in the document. --init initialization method. seeded/random/load seeded: see the paper for details of this method random: random initialization load: load word probabilities and randomly initialize topic proportions --model name of the model to load --max_iter maximum iterations after which we stop the EM algorithm. (default: 100) --convergence If increase in the log-likelihood is less than "convergence", EM is terminated. (default: 5e-3) --save_lag Save the model at every "save_lag" step. (default: -1) --step Number of topics to remove for next steps' initialization. See the paper for model order selection. (default 0)
Output format: Training phase saves the follwong files in the directory:
final.alpha: Contains topic proportions, where each line corresponds to a document in the format: [alpha_1] [alpha_2] ... [alpha_M] where M is the number of topics final.v: Binary switches for topic proportions (i.e. v switches) in the same format as in final.alpha. final.beta Contains M+1 columns and N rows where each row corresponds to a term (N: total # unique words) First column is the shared model, and each of the next M columns indicates probability of words under that topic. final.u Contains u switches in M columns and N rows final.other First row is the number of topics and the second number of terms likelihood.dat: Contains bic, log-likelihood, and convergence values at each iteration of EM. nbar.txt: Indicates total number of topic-specific words at each iteration of EM.
Test step saves the follwong files in the directory: test-alpha: Similar to final.alpha. test-lhood: similar to likelihood.dat