How to format the inputs and targets for a feedforward neural network?

I have to create a feedforward neural netweok in order to classify some signal data in matlab. Below is the dataset I have been given. I'm not sure what parameters from this dataset I should use as my inputs and also how to create the targets from this as well. I am using MatLab 2020a and the deep learning toolbox.

rng(2);

sample_rate = 6700; % Samples measured per second
T = 1; % Time in seconds
num_classes = 3;

tt = 0:1/sample_rate:T-1/sample_rate;
num_samples = length(tt);

m = 1;
k = 400:50:90000;
num_freq = length(k);

A = randperm(num_freq) / num_freq;

phi = 0;

% Generate U1ndamped (c = 0)

U1 = zeros([num_samples, 1, 1, num_freq]);
c = 0;

gamma = c / (2 * m);

omega0 = sqrt(k / m);
omega1 = sqrt(omega0.^2 - gamma^2);
for i = 1:num_freq
    U1(:,:,:,i) = A(i) * exp(-1 * gamma * tt) .* cos(omega1(i) * tt - phi) + 0.01*randn(1,num_samples);
end
figure;
plot(tt,U1(:,:,:,1));
hold on;
plot(tt,U1(:,:,:,end));

% Generate D1amped

D1 = zeros([num_samples, 1, 1, num_freq]);

for i = 1:num_freq
    maxx_c = 2*sqrt(k(i));
    c = 2*sqrt(k(i))- maxx_c + 10;

    gamma = c / (2 * m);

    omega0 = sqrt(k(i) / m);
    omega1 = sqrt(omega0.^2 - gamma^2);
    
    D1(:,:,:,i) = A(i) * exp(-1 * gamma * tt) .* cos(omega1 * tt - phi) + 0.01*randn(1,num_samples);
end 

figure;
plot(tt,D1(:,:,:,1));
hold on;
plot(tt,D1(:,:,:,end));


% Generate C1ritically D1amped

C1 = zeros([num_samples, 1, 1, num_freq]);

for i = 1:num_freq
    maxx_c = 2*sqrt(k(i));
    c = maxx_c;

    gamma = c / (2 * m);

    omega0 = sqrt(k(i) / m);
    omega1 = sqrt(omega0.^2 - gamma^2);
    
    C1(:,:,:,i) = A(i) * exp(-1 * gamma * tt) .* cos(omega1 * tt - phi) + 0.01*randn(1,num_samples);
end 

figure;
plot(tt,C1(:,:,:,1));
hold on;
plot(tt,C1(:,:,:,end));

% Generate U1ndamped (c = 0)

U2 = zeros([num_samples, 1, 1, num_freq]);
c = 0;

gamma = c / (2 * m);

omega0 = sqrt(k / m);
omega1 = sqrt(omega0.^2 - gamma^2);
for i = 1:num_freq
    U2(:,:,:,i) = A(i) * exp(-1 * gamma * tt) .* cos(omega1(i) * tt - phi) + 0.05*randn(1,num_samples);
end
figure;
plot(tt,U2(:,:,:,1));
hold on;
plot(tt,U2(:,:,:,end));

% Generate D1amped

D2 = zeros([num_samples, 1, 1, num_freq]);

for i = 1:num_freq
    maxx_c = 2*sqrt(k(i));
    c = 2*sqrt(k(i))- maxx_c + 10;

    gamma = c / (2 * m);

    omega0 = sqrt(k(i) / m);
    omega1 = sqrt(omega0.^2 - gamma^2);
    
    D2(:,:,:,i) = A(i) * exp(-1 * gamma * tt) .* cos(omega1 * tt - phi) + 0.05*randn(1,num_samples);
end 

figure;
plot(tt,D2(:,:,:,1));
hold on;
plot(tt,D2(:,:,:,end));


% Generate C1ritically D1amped

C2 = zeros([num_samples, 1, 1, num_freq]);

for i = 1:num_freq
    maxx_c = 2*sqrt(k(i));
    c = maxx_c;

    gamma = c / (2 * m);

    omega0 = sqrt(k(i) / m);
    omega1 = sqrt(omega0.^2 - gamma^2);
    
    C2(:,:,:,i) = A(i) * exp(-1 * gamma * tt) .* cos(omega1 * tt - phi) + 0.05*randn(1,num_samples);
end 

figure;
plot(tt,C2(:,:,:,1));
hold on;
plot(tt,C2(:,:,:,end));

% Generate U1ndamped (c = 0)

U3 = zeros([num_samples, 1, 1, num_freq]);
c = 0;

gamma = c / (2 * m);

omega0 = sqrt(k / m);
omega1 = sqrt(omega0.^2 - gamma^2);
for i = 1:num_freq
    U3(:,:,:,i) = A(i) * exp(-1 * gamma * tt) .* cos(omega1(i) * tt - phi) + 0.1*randn(1,num_samples);
end
figure;
plot(tt,U3(:,:,:,1));
hold on;
plot(tt,U3(:,:,:,end));

% Generate D1amped

D3 = zeros([num_samples, 1, 1, num_freq]);

for i = 1:num_freq
    maxx_c = 2*sqrt(k(i));
    c = 2*sqrt(k(i))- maxx_c + 10;

    gamma = c / (2 * m);

    omega0 = sqrt(k(i) / m);
    omega1 = sqrt(omega0.^2 - gamma^2);
    
    D3(:,:,:,i) = A(i) * exp(-1 * gamma * tt) .* cos(omega1 * tt - phi) + 0.1*randn(1,num_samples);
end 

figure;
plot(tt,D3(:,:,:,1));
hold on;
plot(tt,D3(:,:,:,end));


% Generate C1ritically D1amped

C3 = zeros([num_samples, 1, 1, num_freq]);

for i = 1:num_freq
    maxx_c = 2*sqrt(k(i));
    c = maxx_c;

    gamma = c / (2 * m);

    omega0 = sqrt(k(i) / m);
    omega1 = sqrt(omega0.^2 - gamma^2);
    
    C3(:,:,:,i) = A(i) * exp(-1 * gamma * tt) .* cos(omega1 * tt - phi) + 0.1*randn(1,num_samples);
end 

figure;
plot(tt,C3(:,:,:,1));
hold on;
plot(tt,C3(:,:,:,end));

% C1ombine D1ata
x_data = cat(4,U1,U2,U3,D1,D2,D3,C1,C2,C3);
y_data = [zeros(1, 3*length(k)) ones(1,3*length(k)) 2*ones(1,3*length(k))];
y_data = categorical(y_data);

p = randperm(num_classes*3*num_freq);
x = x_data(:,:,:,p);
y = y_data(p);
figure;

hist(y);

N = length(y);

train_split = round(0.8 * N, 0);
test_split = round(0.1 * N, 0);

x_train = x(:,:,:,1:train_split);
x_val = x(:,:,:,train_split:train_split+test_split);
x_test = x(:,:,:,train_split+test_split:end);

y_train = y(1:train_split);
y_val = y(train_split:train_split+test_split);
y_test = y(train_split+test_split:end);

figure;
subplot(1,3,1)
hist(y_train);
xlabel('Train');
subplot(1,3,2)
hist(y_val);
xlabel('Val');
subplot(1,3,3)
hist(y_test);
xlabel('Test');

save('ff_cnn_data.mat', 'x_train', 'y_train', 'x_val', 'y_val', 'x_test', 'y_test');

Answer 1

[Swetha Polemoni]

Hi

It is my understanding that you are doing classification. As it is clearly mentioned in the code x_train and y_tran are for traning i.e., x_train is input to the network and y_train values are target values. You may find the following link helpful for classification in Matlab.

https://www.mathworks.com/help/stats/classification.html