how neural network outputs are kept constant for each time running the program

I used to traingdm algorithm and whenever restart my computer all outputs are changes. Because weights are different. How can I save idealized weights ?

Answer 1

[Greg Heath]

Random weight initialization and trn/val/tst data division are defaults.

If you wish to duplicate a design you have to reset the state of the RNG. I always use rng(0) before designing the first of multiple designs in a double-loop design

 rng(0)
 for h = 1:numH      % Loop over number of hidden nodes
   for n = 1:Ntrials % Loop over number of weight initializations
      s(n,h) = rng  % save the initial RNG state
      net = ...
      ......
   end
 end

If the best design occurs for h= Hb, n=Nb it can be recreated via

 h = Hb
 n = Nb
 rng(s(h,n))
 net = ...

If you wish to reuse a design, save the net

 net1 =net;
 save net1;

Then,later

 load net1
 net = net1

Finally, if you just want to save the weights

 wb(h,n) = getwb(net).

NOTE: THE MULTIPLE INPUT COMMANDS IN HELP GETWB AND DOC GETWB ARE ERRONEOUS

Then later you can reconstruct the net via

 set(net,wb)

Hope this helps.

Thank you for formally accepting my answer

Greg