Simple neural network computation NOT working

I've used the "Neural Net Fitting" App as well as my own code to compute weights for a very simple 1 hidden layer (w/ 1 neuron).

== my own code ==

h1 = X_input * w1 + b1;

o1 = tanh(h1);

h2 = o1 * w2 + b2;

o2 = tanh(h2);

=====

However, it does not output a correct output compared to the "Neural Net Fitting" App.

Is there anything wrong with the above algorithm?

Thanks.

Answer 1

[Steven Lord]

Do you have any pre- or post-processing steps?

[Jacob Jeong]

%% Linear function for testing NN %%

function [x_input, y_true] ...

= test_lin(N_samples)

a = 3;

b = -5;

% preallocated for speed

x_input = zeros(1, N_samples);

y_true = zeros(1, N_samples);

% generate x (within [-10, 10])

for i = 1:N_samples

x_input(i) = unifrnd(-1, 1);

y_true(i) = a * x_input(i) + b;

end

%%

%% Testbench %%

clear; clc;

N_neuron = 1; % number of neurons

K_max = 1000;

Y_max = 100;

eta = 0.01;

[x_input, y_true] = test_lin(K_max);

X_input = x_input;

Y_true = y_true/Y_max;

Y_pred = zeros(1,K_max);

%%

%% Trained Neural Net Fitting using X_input and Y_true

got the following from Neural Net Fitting

% Layer 1

b1 = -0.0034309766062626356579;

IW1_1 = 0.22567942543499841523;

% Layer 2

b2 = 0.015126064794135411773;

LW2_1 = 4.4754110209172548451;

%% Neural Net Fitting app uses tanh for hidden layer and linear function for output layer

h1 = X_input * w1 + b1;

o1 = tanh(h1);

h2 = o1 * w2 + b2;

o2 = h2

o2 should be same as when I input X_input to Neural Net Fitting function, but is NOT the same....