checkDeepOneWave.m

close all
clear
addpath("./functions/")

% %% --prepare for calc 
% clc
% num_days     = 2;
% num_subjects = 9;
% vClass = [];
% Events = {};
% 
% for subject = 1:num_subjects
%     for day = 1:num_days
%         [tmp_Events, tmp_vClass]  = GetEvents( subject, day );
%         Events = combineTwoCellArrays(Events,tmp_Events);
%         vClass = [vClass; tmp_vClass];
%     end
% end
%% extract and save

%% --prepare for calc 
clc
num_days     = 2;
num_subjects = 9;

for subject = 1:num_subjects
    vClass = [];
    Events = {};
    for day = 1:num_days
        [tmp_Events, tmp_vClass]  = GetEvents( subject, day );
        Events = combineTwoCellArrays(Events,tmp_Events);
        vClass = [vClass; tmp_vClass];
    end
    
    m_data   = cell2mat(Events);
    v_lables = kron(vClass, ones(22,1));

    start_of_train = 1;
    end_of_train   = int32(size(m_data,2)*0.8);
    start_of_test  = end_of_train + 1;
    end_of_test    = size(v_lables,1);
    
    m_train_data  = m_data(:,start_of_train:end_of_train);
    v_train_lable = v_lables(start_of_train:end_of_train);
    m_test_data  = m_data(:,start_of_test:end_of_test);
    v_test_lable = v_lables(start_of_test:end_of_test);
    
    train_perm = randperm(length(v_train_lable));
    test_perm  = randperm(length(v_test_lable));

    %--change the order
    m_train_data  = m_train_data(:,train_perm);
    v_train_lable = v_train_lable(train_perm);
    
    m_test_data  = m_test_data(:,test_perm);
    v_test_lable = v_test_lable(test_perm);
    
    %save train
    for ii = 1:size(m_train_data,2)
        x = m_train_data(:,ii);
        save("../tmp/train/" + num2str(subject) + "/" + num2str(ii), 'x');
    end 
    lable = v_train_lable;
    save("../tmp/train/" + num2str(subject) + "/lable" , 'lable');

    %save test
    for ii = 1:size(m_test_data,2)
        x = m_test_data(:,ii);
        save("../tmp/test/" + num2str(subject) + "/" + num2str(ii), 'x');
    end 
    lable = v_train_lable;
    save("../tmp/test/" + num2str(subject) + "/lable" , 'lable');
end

%% --Devide to test training and validation          
m_data   = cell2mat(Events);
v_lables = kron(vClass, ones(22,1));

start_of_train = 1;
% end_of_train   = int32(size(v_lables,1) * 0.8);
end_of_train   = int32(500);
% start_of_val   = end_of_train + 1 ;
% end_of_val     = int32(size(v_lables,1) * 0.8);
% start_of_test  = end_of_train + 1;
% end_of_test    = size(v_lables,1);



m_train_data  = m_data(:,start_of_train:end_of_train);
v_train_lable = v_lables(start_of_train:end_of_train);

% m_val_data    = m_data(:,start_of_val:end_of_val);
% v_val_lable   = v_lables(start_of_val:end_of_val);

% m_test_data   = m_data(:,start_of_test:end_of_test);
% v_test_lable  = v_lables(start_of_test:end_of_test);

%-- shuffel data
train_perm = randperm(length(v_train_lable));
% val_perm   = randperm(length(v_val_lable));
test_perm  = randperm(length(v_test_lable));

%--change the order
m_train_data  = m_train_data(:,train_perm);
v_train_lable = v_train_lable(train_perm);

% m_val_data   = m_val_data(:,val_perm);
% v_val_lable  = v_val_lable(val_perm);

% m_test_data   = m_test_data(:,test_perm);
% v_test_lable  = v_test_lable(test_perm);
%% save all the data

for ii = 1:size(m_train_data,2)
    x = m_train_data(:,ii);
    save("../new_tmp/train/" + num2str(ii), 'x');
end 
lable = v_train_lable;
save("../new_tmp/train/lable" , 'lable');

% for ii = 1:size(m_val_data,2)
%     x = m_val_data(:,ii);
%     save("../tmp/val/" + num2str(ii), 'x');
% end 
% lable = v_val_lable;
% save("../tmp/val/lable" , 'lable');

% for ii = 1:size(m_test_data,2)
%     x = m_test_data(:,ii);
%     save("../tmp/test/" + num2str(ii), 'x');
% end 
% lable = v_test_lable;
% save("../tmp/test/lable" , 'lable');


%% cut the data

% m_train_data  = m_train_data(:,1:500);
% v_train_lable = v_train_lable(1:500);
% m_val_data  = m_train_data;
% v_val_lable = v_train_lable;



%% set the data to network
v_train_lable_categorical = discretize(v_train_lable,[0.5 1.5 2.5 3.5 4.5],...
                                       'categorical', {'1' '2' '3' '4'});
tmp_m_train_data           = reshape(m_train_data,750,1,1,[]);     



v_val_lable_categorical = discretize(v_val_lable,[0.5 1.5 2.5 3.5 4.5],...
                                       'categorical', {'1' '2' '3' '4'});
tmp_m_val_data          = reshape(m_val_data,750,1,1,[]);     



v_test_lable_categorical = discretize(v_test_lable,[0.5 1.5 2.5 3.5 4.5],...
                                       'categorical', {'1' '2' '3' '4'});
tmp_m_test_data          = reshape(m_test_data,750,1,1,[]);     



%% define network structure

layers = [
    
    imageInputLayer([int32(size(m_train_data,1)), int32(1)])
    %-- layer 1
    convolution2dLayer([7 1], 4,'Padding','same')
    batchNormalizationLayer
    reluLayer
    dropoutLayer
    maxPooling2dLayer([2 1],'Stride',2)
    
    %-- layer 2
    convolution2dLayer([7 1], 8,'Padding','same')
    batchNormalizationLayer
    reluLayer
    dropoutLayer
    maxPooling2dLayer([2 1],'Stride',2)
    
    
    %-- layer 3
    convolution2dLayer([7 1], 16,'Padding','same')
    batchNormalizationLayer
    reluLayer

    
    %-- fully connected
    fullyConnectedLayer(4)
    softmaxLayer
    classificationLayer];




% layers = [
%     
%     imageInputLayer([int32(size(m_train_data,1)), int32(1)])
%     %-- layer 1
%     convolution2dLayer([60 1], 80,'Padding','same')
%     batchNormalizationLayer
%     reluLayer
%     maxPooling2dLayer([4 1],'Stride',1)
%     dropoutLayer
%     
%     %-- layer 2
%     convolution2dLayer([1 1], 80,'Padding','same')
%     batchNormalizationLayer
%     reluLayer
%     maxPooling2dLayer([2 1],'Stride',2)
%     
% 
%     
%     %-- fully connected
%     fullyConnectedLayer(5000)

%     dropoutLayer
%     fullyConnectedLayer(5000)
%     dropoutLayer
%     
%     
%     %--output
%     fullyConnectedLayer(4)    
%     softmaxLayer
%     
%     classificationLayer];


    







%% training options


options = trainingOptions(  'sgdm', ...
                            'InitialLearnRate',0.01, ...
                            'MaxEpochs',300, ...
                            'Shuffle','every-epoch', ...
                            'ValidationData',{tmp_m_val_data, v_val_lable_categorical}, ...
                            'ValidationFrequency',30, ...
                            'Verbose',false, ...
                            'Plots','training-progress');

%% train 

net = trainNetwork( tmp_m_train_data, v_train_lable_categorical, layers, options );
%%
view(net)

%% test accuracy
YPred       = classify(net,tmp_m_test_data);
YValidation = v_test_lable_categorical;

test_accuracy = sum(YPred == YValidation)/numel(YValidation)
%% accuracy on val
YPred       = classify(net, tmp_m_val_data);
YValidation = v_val_lable_categorical;

val_accuracy = sum(YPred == YValidation)/numel(YValidation)
%% accuracy on train
YPred       = classify(net, tmp_m_train_data);
YValidation = v_train_lable_categorical;

train_accuracy = sum(YPred == YValidation)/numel(YValidation)