When I set net.trainP​aram.epoch​s=5000, the max epoch is still 1000 in GUI

I can't change the max epoch using net.trainParam.epochs.

[Elizabeth Reese]

Can you post how you created the network? It would be helpful to see the code that reproduces this issue. Sometimes the order of changing different parameters and options can reset values back to their defaults and 1000 is the default epoch value for the training functions.

[DAR VON]

function net = network_train(train_data,train_label)

layer=25;

net=newff(train_data,train_label,layer);

net.trainParam.epochs=5000;

net.trainParam.lr=0.01;

%net.trainParam.lr=lr;

net.trainParam.goal=0.001;

net.trainParam.show=50;

net.trainParam.max_fail=500;

net.trainFcn='trainscg';

net=train(net,train_data,train_label);

end

Thanks a lot.

Answer 1

[Elizabeth Reese]

Thank you for posting your code.

Based on the Neural Network Toolbox documentation here, updating the net.trainFcn will reset the net.trainParam to the default values for the new function. So changing the net.trainFcn to trainscg changes the net.trainParam to the defaults listed here. These defaults include having the epochs set to 1000.

To fix this, just move the assignment to the net.trainFcn to before any updates to the net.trainParam.

[DAR VON]

Would you please tell me how can I use GPU to accelerate computation in projects like BP， thanks a lot.

[Greg Heath]

Why didn't you post this as a new question?