Neural Networks for integer values -
i have approximately 5000 integer vectors (=size) like:
[1 0 4 2 0 1 3 ...]
they have same length n=32 , values ranges 0 4 let's [0 max].
created nn takes vectors inputs , outputs binary array corresponding 1 of desired output(number of possible outputs = m):
instance [0 1 0 0 ...0] => 2nd output. array_length = m
used multi layer perceptron in neuroph integer values did not converge.
guessing problem using integer values or using mlp 3 layers: input, hidden , output.
can advise me on network structure? type of nn suitable? should remodel input , output simplify learning process? have been thinking gray encoding integers input.
Comments
Post a Comment