Neural Network Toolbox > Is there anybody who is using 'newff' for multiple inputs?

Neural Network Toolbox > Is there anybody who is using 'newff' for multiple inputs?

I'm using a newff to forecast Interchange Scheduling from multiple inputs, around 10, but the results were too poor (MAPE is about 50%). Following is the functions I used.

net = newff(P,T,[90],{'logsig'},'traincgb','learngdm','mse');
net.trainParam.epochs = 2000;
net.trainParam.goal = 0.001;

net = train(net,P,T);

I need to find the right functions to get a reasonable result.

Would you let me know if you have any idea to improve the accuracy? Also, I'll be happy if you show your function example.

Answer 1

[Greg Heath]

Using 90 hidden nodes is probably 10 orders of magnitude too high.

Use as many defaults as possible.

Vary the number of hidden nodes.

For each choice try 10 different sets of initial weights.

Search the NEWSGROUP and ANSWERS using

 greg newff Ntrials