-
Notifications
You must be signed in to change notification settings - Fork 1
/
train_HMM.m
164 lines (141 loc) · 4.64 KB
/
train_HMM.m
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
cd '/home/ting/Documents/eeg_hmm';
addpath('/home/ting/Documents/eeglab')
addpath(genpath('HMM-MAR'))
addpath('/data/projects/Shawn/2019_HMM/data')
addpath('/data/projects/Shawn/2016 JNE/dataset')
%% Indexing datafile
data_base_dir = '/data/projects/Shawn/2019_HMM/data/';
% data_base_dir = '/data/projects/Shawn/2016 JNE/dataset/';
data_filelist = dir(strcat(data_base_dir, '*.set'));
data_filenames = {};
output_filenames = {};
for i = 1:length(data_filelist)
filename = data_filelist(i).name;
output_filename = split(filename, '.');
output_filename = output_filename{1};
if ismember(output_filename, {'s54_081209n'})
data_filenames{i} = filename;
output_filenames{i} = output_filename;
end
end
data_filenames = data_filenames(~cellfun('isempty', data_filenames));
output_filenames = output_filenames(~cellfun('isempty', output_filenames));
n_of_files = length(data_filenames);
%% Prepare raw data
eegdata_list = cell(1, n_of_files);
eeglab;
for i = 1:n_of_files
filename = data_filenames{i};
eegdata_list{i} = pop_loadset(filename);
end
%% Prepare global training parameters
% Regarding T, it can be either a (N X 1) vector (where N is the total number of trials
% or segments for all subjects) containing the length of each trial/segment/subject, or
% a (no. of subjects X 1) cell with each element containing a vector (no. of trials X 1)
% reflecting the length of the trials for that particular subject.
n_epoch = 1;
select_start = 0;
select_end = 1;
Fs = 250;
K = 10;
use_stochastic = 0;
method = 'GAU';
% cyc = 1000;
% tol = 1e-5;
% initrep = 3;
% initcycle = 50;
options = struct();
options.K = K; % number of states
options.Fs = Fs;
options.verbose = 1;
options.useParallel = 0;
options.standardise = 0;
options.onpower = 0;
% options.cyc = cyc;
% options.tol = tol;
% options.initrep = initrep;
% options.initcyc = initcycle;
if strcmp(method,'MAR')
options.order = 2;
options.zeromean = 1;
options.covtype = 'diag';
options.DirichletDiag = 100;
elseif strcmp(method, 'crossMAR')
options.order = 1;
options.zeromean = 1;
options.covtype = 'uniquefull';
options.DirichletDiag = 100;
elseif strcmp(method, 'TDE')
% For TDE: order = 0, zeromean = 1, covtype = 'full'
options.embeddedlags = -7:7;
options.order = 0; % no autoregressive components
options.zeromean = 1; % model the mean
options.covtype = 'full'; % full covariance matrix
elseif strcmp(method, 'GAU')
options.order = 0;
options.zeromean = 0;
options.covtype = 'full';
options.onpower = 0;
options.DirichletDiag = 100;
elseif strcmp(method, 'MIX')
% default
end
options.DirichletDiag = 100;
if use_stochastic && n_epochs > 1
options.BIGNinitbatch = 15;
options.BIGNbatch = 15;
options.BIGtol = tol;
options.BIGcyc = cyc;
options.BIGinitrep = initrep;
options.BIGundertol_tostop = 5;
options.BIGforgetrate = 0.7;
options.BIGbase_weights = 0.9;
end
%% Train
delete(gcp('nocreate')); % shut down any current pool
npar = 27;
parpool(npar); % request workers from the cluster
options_list = repmat(options, 1, n_of_files);
results_list = cell(1, n_of_files);
parfor (idx = 1:n_of_files, npar)
X = transpose(eegdata_list{idx}.data);
T = eegdata_list{idx}.pnts;
Fs = eegdata_list{idx}.srate;
options_list(idx).Fs = Fs;
start_time = tic;
[hmm, Gamma, ~, vpath, ~, residuals, fehist] = hmmmar(X, T, options_list(idx));
time_elapsed = toc(start_time)
% save output and meta data
result = struct();
result.hmm = hmm;
result.Gamma = Gamma;
% result.Xi = Xi;
result.vpath = vpath;
result.fehist = fehist;
% result.GammaInit = GammaInit;
result.residuals = residuals;
result.fehist = fehist;
result.time_elapsed = time_elapsed;
result.select_start = select_start;
result.select_end = select_end;
result.training_data_size = size(X);
results_list{idx} = result;
end
fprintf('Training done');
results_dir = strcat('/home/ting/Documents/eeg_hmm/HMM_results/results_K', num2str(K), '/');
if ~exist(results_dir, 'dir')
mkdir(results_dir);
end
for idx = 1:n_of_files
% Find the first non-duplicate filename
fileindex = 0;
result_filename_prefix = strcat(method, '_', output_filenames{idx});
result_filename = strcat(result_filename_prefix, '_', num2str(fileindex), '.mat');
while isfile(strcat(results_dir, result_filename))
fileindex = fileindex + 1;
result_filename = strcat(result_filename_prefix, '_', num2str(fileindex), '.mat');
end
result = results_list{idx};
save(strcat(results_dir, result_filename), '-struct', 'result');
end
delete(gcp('nocreate'));