How to determine if the simulated result is matched?

Here is my code:

inputData = rand(5, 56);
targetData = eye(56);
simInput = [a;b;c;d;e]; %%assume the value match 1 of the column in inputData
net = feedforwardnet(5);
net = train(net, inputData, targetData);
y1 = sim(net, simInput);
y2 = find(max(y1));

y2 should return a value that tell me which column it matched in inputData. However it only return 1 regardless the simInput. So am I using it right?

Answer 1

[Greg Heath]

You have used random data. For a more reasonable example, go to the NN data library

 help nndatasets

 doc nndatasets

A simpler alternative is to just use the example in the patternnet classifier documentation

 help patternnet

 doc patternnet

Hope this helps.

Greg

PS: also lookup the functions IND2VEC and VEC2IND to transfer between target matrices and class indices.

Answer 2

[Walter Roberson]

You probably want

[maxy1, y2] = max(y1);

If y1 is a vector then max(y1) is going to be a scalar, and find() applied to a scalar is going to return [] if the scalar is 0 and is going to return 1 (the location of the non-zero element in the scalar) otherwise. find(max(y1)) does not mean to search y1 for its maximum and return the location in y2. Or maybe you just wanted to know what the maximum was; if so then just max() works.

[Vishnu Rajen]

Thanks for the reply, now I am able to get the result. However, the result is not matched for most of the time, do you have any idea?

[Walter Roberson]

My idea is that having only 1 sample per class is not adequate to train a neural network.