%% main.m 
% Main function of SVD2 code.
% Prepares the corpus and calls SVD2.

% If you use this code, please cite:
%   Michael Lamar*, Yariv Maron*, Mark Johnson, Elie Bienenstock. 
%   SVD and clustering for unsupervised POS tagging. (ACL 2010) 
% For more information, see the above reference.


% Prepare corpus
diary on
fprintf(sprintf('\n'))
disp(['**********************   START   ' datestr(now) '   **********************'])
fprintf(sprintf('\n'))

% Load corpus
[tokens, tags17, tags45, N_types, freqs] = load_wsj;
% tokens  - The corpus. Array of word ids.
% tags17 - The tags. Reduced tag ste of Smith &Eisner
% tags45 - The tags (Penn Tag set).
% N_types- Numnber of different word types in the corpus.
% freqs  - Frequency of every word type (fraction).

%% Find 17 tags

% Set parameters
svd_R1   = 100; % Number of singular vectors for first SVD
svd_R2   = 300; % Number of singular vectors for second SVD
cl_num_1 = 500; % Number of clusters after first clustering
cl_num_2 = 17 ; % Number of clusters after second clustering

% Tag corpus with SVD2 (takes three minutes on my desktop computer)
tic
labels17 = SVD2(tokens, N_types, freqs, svd_R1, cl_num_1, svd_R2, cl_num_2);
toc

% Evaluate induced labels against gold-standard tags
[MTO_score, OTO_score, NVI_score] = eval_labels(labels17, tags17);

%% Find 50 tags

% Set parameters
svd_R1   = 100; % Number of singular vectors for first SVD
svd_R2   = 300; % Number of singular vectors for second SVD
cl_num_1 = 500; % Number of clusters after first clustering
cl_num_2 = 50 ; % Number of clusters after second clustering

% Tag corpus with SVD2 (takes three minutes on my desktop computer)
tic
labels50 = SVD2(tokens, N_types, freqs, svd_R1, cl_num_1, svd_R2, cl_num_2);
toc

% Evaluate induced labels against gold-standard tags
[MTO_score, OTO_score, NVI_score] = eval_labels(labels50, tags45);


fprintf(sprintf('\n'))
disp(['**********************   END   ' datestr(now) '   **********************'])
fprintf(sprintf('\n'))
diary off