MATLAB: How to find the value used to seed neural networks in Matlab

neural networkrandom number generator

Hi,

I am using this in Matlab to build and train a neural network, network = fitnet(15); [network, model] = train(network, trainData, trainLabels);

The training results are different each time I run the code. I know that this is due to the random generation of initial weights and biases seeded by some value.

Is it possible for me to know this seed value? If yes, then how can I get it?

Best Answer

Store the state of the random number generator using rng BEFORE you create or initialize your network. Then you can restore the state (again using rng) before the next time you create or initialize a network.

Related Solutions

MATLAB: Is good to repeat the same sample in the training of the neural network

Generally, it does not make it worse.

Hope this helps.

Thank you for formally accepting my answer

Greg

MATLAB: Rand(‘seed’,1)

Seeding the random number generator means initializing it to a certain status. Seeding inside the loop means, that all "random" numbers created inside the loop will be the same in each iteration:

for i = 1:3
  rand('seed', 1);
  disp(rand);
end

Result: 0.51291 0.51291 0.51291

This is not very useful. Seeding RAND outside the loop allows you to reproduce the results:

for j = 1:2
  rand('seed', 1);
  for i = 1:3
    disp(rand);
  end
end

Now you get two series of 3 different numbers:

 0.51291  0.46048  0.3504  0.51291  0.46048  0.3504

Best Answer

Related Solutions

MATLAB: Is good to repeat the same sample in the training of the neural network

MATLAB: Rand(‘seed’,1)

Related Question