Same neural network training result each time at "dividerand" option

I wonder why my NN results shows same performance(error) every run even though I separated data set training, validation and test sets randomly by using 'deviderand'.

I understand the weight and bias are always same because I use rng('default') code, however, I think the results must be different every time due to its randomness of data selection.

Is it because I misunderstand the concept of NN?

The code is the following:

% Solve an Input-Output Fitting problem with a Neural Network
% Script generated by Neural Fitting app
% Created 28-Dec-2016 14:15:01
%
% This script assumes these variables are defined:
%
%   InputT12 - input data.
%   TargetT12 - target data.

rng('default')

x = InputT12;
t = TargetT12;

% Choose a Training Function
% For a list of all training functions type: help nntrain
% 'trainlm' is usually fastest.
% 'trainbr' takes longer but may be better for challenging problems.
% 'trainscg' uses less memory. Suitable in low memory situations.
trainFcn = 'trainlm';  % Levenberg-Marquardt backpropagation.

% Create a Fitting Network
hiddenLayerSize = 10;
net = fitnet(hiddenLayerSize,trainFcn);

% Setup Division of Data for Training, Validation, Testing

net.divideFcn = 'dividerand';
net.divideParam.trainRatio = 70/100;
net.divideParam.valRatio = 15/100;
net.divideParam.testRatio = 15/100;

% Train the Network
[net,tr] = train(net,x,t);

% Test the Network
y = net(x);
e = gsubtract(t,y);
performance = perform(net,t,y)

% View the Network
view(net)

% Plots
% Uncomment these lines to enable various plots.
%figure, plotperform(tr)
%figure, plottrainstate(tr)
%figure, ploterrhist(e)
%figure, plotregression(t,y)
%figure, plotfit(net,x,t)

[Brent McWatters]

How did you solve this?

Answer 1

[Greg Heath]

When you reboot, the RNG goes into the default state.

Therefore begin with states different from default and 0.

Hope this helps

Thank you for formally accepting my answer

Greg

[Dongwoo Kim]

I finally fixed the problem. Thank you for your help.

Answer 2

[Greg Heath]

You are mistaken.

When the rng is repeatedly set to a certain value, The resulting string of PSEUDO-RANDOM numbers is repeated.

HOPE THIS HELPS

THANK YOU FOR FORMALLY ACCEPTING MY ANSWER

GREG

[Dongwoo Kim]

I understand your point. However, after I delete the rng('default') code, still same results are shown.

For example,

• 1st result:67.6067
• 2nd result:46.0002 -> different result due to deleated RNG code.
• 3rd result:26.4202

after reboot the matlab, same results are shown

• 1st result:67.6067
• 2nd result:46.0002
• 3rd result:26.4202

Something wrong in my code?

[Merged from duplicate question]

I wonder why my NN results shows same performance(error) every run even though I separated data set training, validation and test sets randomly by using 'deviderand'.

For example,

• 1st run result: 67.6067
• 2nd run result: 46.0002
• 3rd run result: 26.4202

however, same results are shown after rebooting the matlab

• 1st run result: 67.6067
• 2nd run result: 46.0002
• 3rd run result: 26.4202

I think the results must be different due to its randomness of data selection.

The code is the following:

% Solve an Input-Output Fitting problem with a Neural Network
% Script generated by Neural Fitting app
% Created 28-Dec-2016 14:15:01
%
% This script assumes these variables are defined:
%
%   InputT12 - input data.
%   TargetT12 - target data.

x = InputT12;
t = TargetT12;

% Choose a Training Function
% For a list of all training functions type: help nntrain
% 'trainlm' is usually fastest.
% 'trainbr' takes longer but may be better for challenging problems.
% 'trainscg' uses less memory. Suitable in low memory situations.
trainFcn = 'trainlm';  % Levenberg-Marquardt backpropagation.

% Create a Fitting Network
hiddenLayerSize = 10;
net = fitnet(hiddenLayerSize,trainFcn);

% Setup Division of Data for Training, Validation, Testing

net.divideFcn = 'dividerand';
net.divideParam.trainRatio = 70/100;
net.divideParam.valRatio = 15/100;
net.divideParam.testRatio = 15/100;

% Train the Network
[net,tr] = train(net,x,t);

% Test the Network
y = net(x);
e = gsubtract(t,y);
performance = perform(net,t,y)

% View the Network
view(net)

% Plots
% Uncomment these lines to enable various plots.
%figure, plotperform(tr)
%figure, plottrainstate(tr)
%figure, ploterrhist(e)
%figure, plotregression(t,y)
%figure, plotfit(net,x,t)