-
Notifications
You must be signed in to change notification settings - Fork 0
/
mlp_P300.m
280 lines (235 loc) · 10.8 KB
/
mlp_P300.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
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
% TP "Réseaux de Neurones Formels"
% M2R CNA
%
% Apprentissage sur un réseau MLP pour la classification
% Base P300 Speller
close all
clear
clc
disp(datetime)
addpath('.\Netlab')
addpath('.\P300')
addpath('.\func')
dir_data_path = 'P300\long';
save_log_NET = 0;
%% Configuration
% Paramètres d'étude principaux
n_sources_tab = [1 4]; % assert(max(n_sources_tab) <= 4)
n_hidden_tab = [1 2 5 10 20]; % Différentes tailles pour la couche cachée
idx_subjects = 14; % assert(max(idx_subjects) <= 20))
percentS_learn_valid_test = {[60 30 10]}; % assert(sum == 100)
% percentS_learn_valid_test = [[60 30 10] ; ...
% [50 25 25] ; ...
% [40 20 40] ; ...
% [30 15 55] ; ...
% [20 10 70] ; ...
% [10 5 85] ; ...
% [6 3 91] ; ...
% [4 2 94] ; ...
% [2 1 97] ; ...
% [1 0.5 98.5] ; ...
% [0.4 0.2 99.4] ]';
% Paramètres d'étude secondaires
optimizer = 'graddesc'; % 'conjgrad', 'scg', 'graddesc'
type_err = 'auc'; % Stop criteria : 'auc', 'costfunc'
n_output_neurons = 1; % 1, 2
if n_output_neurons == 1
outfunc = 'logistic'; % 'logistic', 'linear'
elseif n_output_neurons == 2
outfunc = 'softmax'; % 'softmax', 'logistic', 'linear'
end
% Paramètres à fixer
n_iter_max = 100;
delta_iter_valid = 5; % Intervalle entre deux validations
flag_CR = 1; % Données centrées réduites (1) ou non (0)
% Paramètres d'affichage
disp_learning_curves = 1;
% disp_roc = 0; % TODO
% disp_conf_mat = 0; % TODO
RGB_colors = {[0, 0.4470, 0.7410], ...
[0.8500, 0.3250, 0.0980], ...
[0.4660, 0.6740, 0.1880], ...
[0.4940, 0.1840, 0.5560], ...
[0.9290, 0.6940, 0.1250], ...
[0.3010, 0.7450, 0.9330] };
%% Strcuture de savegarde des données
cnt_save = 0;
log_NET = struct('CR', {}, ...
'idx_subject', {}, ...
'n_sources', {}, ...
'percent_learn', {}, ...
'percent_valid', {}, ...
'percent_test', {}, ...
'net', {}, ...
'n_hidden', {}, ...
'n_output', {}, ...
'outfunc', {}, ...
'optimizer', {}, ...
'type_err', {}, ...
'n_iter_max', {}, ...
'validation_num_iter', {}, ...
'validation_err', {}, ...
'test_err', {} );
%% Main loops
for idx_per = size(percentS_learn_valid_test, 2)
percent_learn_valid_test = percentS_learn_valid_test{idx_per};
for idx_subject = idx_subjects
if disp_learning_curves
fig_handle = figure(idx_subject);
[subplot_dim1, subplot_dim2] = subplotDim(size(n_sources_tab, 2));
set(fig_handle, 'Name', ['Subject ' num2str(idx_subject)], 'NumberTitle', 'off')
end
cnt_source = 0; % Juste pour l'indice de subplot
for n_source = n_sources_tab
cnt_source = cnt_source + 1;
cnt_plot = 0;
p = []; % plot's handle
legends = {};
% ----------------------------------
% Chargement de la base de données, répartition en Learn, Valid et Test
% ----------------------------------
[data, class, n_data] = fLoadP300DataSet(dir_data_path, idx_subject, n_source, percent_learn_valid_test, flag_CR);
if n_output_neurons == 1
class.Learn = class.Learn(:, 1);
class.Valid = class.Valid(:, 1);
class.Test = class.Test(:, 1);
end
for idx_hidden = 1 : size(n_hidden_tab, 2)
n_hidden = n_hidden_tab(idx_hidden);
if (disp_learning_curves) RGB_color = RGB_colors{idx_hidden}; end
%-------------------------
% Création et initialisation du MLP
%-------------------------
net = mlp(size(data.Learn, 2), n_hidden, n_output_neurons, outfunc, 0);
options = zeros(1,18);
options(1) = -1; % This provides or nor display of error values.
options(2) = -1; % Stopping criterion of the weights gradient
options(3) = -1; % Stopping criterion of the error gradient
options(14) = 1; % Number of training cycles. assert(options(14)==1)
options(17) = 0.6; % Momentum.
options(18) = 0.001; % Learning rate.
err_learn_tot = [];
err_valid_tot = [];
resnet = [];
%-------------------------
% Apprentissage et validation
%-------------------------
cnt_iter = 0;
for i = 1 : round(n_iter_max / delta_iter_valid)
for j = 1 : delta_iter_valid
cnt_iter = cnt_iter + 1;
% Apprentissage
[net, options] = netopt(net, options, data.Learn, class.Learn, optimizer);
err_learn = computeErr(net, data.Learn, class.Learn, type_err);
err_learn_tot = [err_learn_tot err_learn];
end
clear j
resnet = [resnet net];
% Erreur de validation durant l'apprentissage
err_valid = computeErr(net, data.Valid, class.Valid, type_err);
err_valid_tot = [err_valid_tot err_valid];
% Affichage des profils d'erreur d'apprentissage et de validation
if disp_learning_curves
figure(idx_subject)
subplot(subplot_dim1, subplot_dim2, cnt_source)
title([num2str(n_source) ' sources'])
cnt_plot = cnt_plot + 1;
p(cnt_plot) = plot(err_learn_tot, '-', 'Color', RGB_color);
legends = [legends, 'noLegend'];
hold on
cnt_plot = cnt_plot + 1;
p(cnt_plot) = plot(cnt_iter, err_valid, 'x', 'Color', RGB_color);
legends = [legends, 'noLegend'];
drawnow
end
end
clear i
%-------------------------
% Récupération du paramétrage "optimal" du réseau
%-------------------------
if strcmp(type_err, 'auc')
[err_valid_best, idx_best] = max(err_valid_tot);
elseif strcmp(type_err, 'costfunc')
[err_valid_best, idx_best] = min(err_valid_tot);
end
net_end = resnet(idx_best);
idx_best = idx_best * delta_iter_valid;
% S'assurer que l'on récupère le bon état du net
err_tmp = computeErr(net_end, data.Valid, class.Valid, type_err);
if strcmp(type_err, 'costfunc')
if min(err_valid_tot) ~= err_tmp
disp('Problème de récupération du net optimal')
return
end
elseif strcmp(type_err, 'auc')
if max(err_valid_tot) ~= err_tmp
disp('Problème de récupération du net optimal')
return
end
end
clear err_tmp
%-------------------------
% Evaluation sur la base de test
%-------------------------
err_test = computeErr(net_end, data.Test, class.Test, type_err);
if disp_learning_curves
cnt_plot = cnt_plot + 1;
p(cnt_plot) = plot(idx_best, err_test, '.', 'Color', RGB_color, 'MarkerSize', 30);
legends = [legends ['nbHidden = ' num2str(n_hidden_tab(idx_hidden)) ' | Test = ' num2str(err_test, '%.3f')] ];
end
%-------------------------
% Sauvegarde des données
%-------------------------
cnt_save = cnt_save + 1;
if mod(cnt_save, 100) == 0 % Affichage de la progression
disp(cnt_save)
end
log_NET(cnt_save).CR = flag_CR;
log_NET(cnt_save).idx_subject = idx_subject;
log_NET(cnt_save).n_sources = n_source;
log_NET(cnt_save).percent_learn = percent_learn_valid_test(1);
log_NET(cnt_save).percent_valid = percent_learn_valid_test(2);
log_NET(cnt_save).percent_test = percent_learn_valid_test(3);
log_NET(cnt_save).net = net_end;
log_NET(cnt_save).n_hidden = net_end.nhidden;
log_NET(cnt_save).n_output = net_end.nout;
log_NET(cnt_save).outfunc = net_end.outfn;
log_NET(cnt_save).optimizer = optimizer;
log_NET(cnt_save).type_err = type_err;
log_NET(cnt_save).n_iter_max = n_iter_max;
log_NET(cnt_save).validation_num_iter = idx_best;
log_NET(cnt_save).validation_err = err_valid_best;
log_NET(cnt_save).test_err = err_test;
end %end loop on n_hidden
clear err_learn err_valid err_test
clear idx_best
clear n_hidden idx_hidden cnt_iter
clear err_learn_tot err_valid_tot
clear RGB_color
% clear net resnet
if disp_learning_curves
[p, legends] = formatLegends(p, legends, 'noLegend');
lgd = legend(p, legends);
if strcmp(type_err, 'auc') location = 'southeast'; else location = 'northeast'; end
set(lgd, 'Location', location)
xlabel("Iterations")
ylabel("Area under ROC curve (AUC)")
end
% if disp_roc
% scores = mlpfwd(net, data.Test);
% [X, Y, ~, ~] = perfcurve(class.Test, scores, 1);
% figure(100)
% plot(X,Y)
% hold on
% end
clear lgd location legends cnt_plot
end %end loop on idx_source
clear cnt_source n_source p
end %end loop on idx_subject
clear idx_subject
clear fig_handle subplot_dim1 subplot_dim2
end %end loop on percent_learn_valid_test
clear RGB_colors
if save_log_NET
save(['log_NET_' datestr(datetime, 'mm-dd_HH-MM-SS') '.mat'], 'log_NET')
end