How to get the performance of more neural networks at once using a for-loop?

Question

Jingyuan Yao 2021-7-19

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/881983-how-to-get-the-performance-of-more-neural-networks-at-once-using-a-for-loop

回答： Vineet Joshi 2021-7-28

Hi, I am trying to get the performance of more neural networks. So I created 100 networks at first.

% Train the Network
%[net,tr] = train(net,x,t);
% Train more networks for better performance 
numNN = 100;
network_array = cell(1, numNN);
for i = 1:numNN
  fprintf('Training %d/%d\n', i, numNN)
  network_array{i} = train(network, x, t,'CheckpointFile','MyCheckpoint','CheckpointDelay',120);
  save 20212101_Workspace_NN_Torsion % Biegung und Torsion
end

For one network the code for test and recalculation is

% Test the Network
y = net_hiddenlayersize6_sortiert(x);
e = gsubtract(t,y);
performance = perform(net_hiddenlayersize6_sortiert,t,y)
% Recalculate Training, Validation and Test Performance
trainTargets = t .* tr.trainMask{1};
valTargets = t .* tr.valMask{1};
testTargets = t .* tr.testMask{1};
trainPerformance = perform(net_hiddenlayersize6_sortiert,trainTargets,y)
valPerformance = perform(net_hiddenlayersize6_sortiert,valTargets,y)
testPerformance = perform(net_hiddenlayersize6_sortiert,testTargets,y)

So now I have to do it for 100 networks. Shall I use a for-loop here? My idea is like this:

% Test the Network, but for more networks
for l = 1:numNN
y{l} = network{1,l}(x);
end
e = gsubtract(t,y);
for l = 1:numNN
performance{l} = perform(network{1,l},t(l),y{1,l})
end
% Recalculate Training, Validation and Test Performance
trainTargets = t .* tr.trainMask{1};
valTargets = t .* tr.valMask{1};
testTargets = t .* tr.testMask{1};
trainPerformance = perform(netzwerk,trainTargets,y)
valPerformance = perform(netzwerk,valTargets,y)
testPerformance = perform(netzwerk,testTargets,y)

I am kinda stuck here. How can I test and do the recalculation for 100 networks at once? I found an example here

https://ch.mathworks.com/help/deeplearning/ug/improve-neural-network-generalization-and-avoid-overfitting.html

%Then, ten neural networks are trained.
net = feedforwardnet(10);
numNN = 10;
nets = cell(1, numNN);
for i = 1:numNN
  fprintf('Training %d/%d\n', i, numNN)
  nets{i} = train(net, x1, t1);
end
%Next, each network is tested on the second dataset with both individual performances and the performance for the average output calculated.
perfs = zeros(1, numNN);
y2Total = 0;
for i = 1:numNN
  neti = nets{i};
  y2 = neti(x2);
  perfs(i) = mse(neti, t2, y2);
  y2Total = y2Total + y2;
end
perfs
y2AverageOutput = y2Total / numNN;
perfAveragedOutputs = mse(nets{1}, t2, y2AverageOutput) 