Patternnet for 14 -class classification problem

Hi,

I am trying to classify the number of 0 and 1 appearing after a binary time series. What i have is a very large sequence of [ 0 1 1 0 1 ] which precedes say a continuous sequence of 3 zeros. ie [ 0 1 1 0 1 0 0 0 ]. I want to use the patternet to do this ie input [0 1 1 0 1] and the network should classify it as 3. From the binary series i have extracted the Input and the corresponding target. I want to classify upto 7 continuous zeros (1 class for 1 zero, 2nd class for 2 continuous 0s and so) and ones (8th class for 1 ones,..14th class for 7+ ones). ie 14 classifications. The problem is that it is not able to do so accurately. There is a confusion of more that 0.8. It classifies most as 5, 6 (zeros of length 5 and 6) and 14 - ones of length 7 or more (the most). Other classes aren't classified.

I am using the default parameters (trainscg, max_fails =6) and for hidden layers i have [20 30]. I generate around 30k Input Target values. Train:Validate:test = 0.6:0.2:0.2. I am using eye(14) to create the targets.

My question is - Is what i am trying to do even possible using patternet? If yes, what are the things i need to take care of so i get better results. and if no which network should i use?

Thank you very much.

[Greg Heath]

Please try to explain your problem a little better. For example,

 size(input)  = [I N ] = ?
 size(target) = [ O N] = ?

 inputexample =   ...
 targetexample =  ...

[Gmoj Goo]

size(input) = [I N ] = 30*28008

size(target) = [ O N] = 14*28008

 inputexample =   [2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1; .. ]'

 targetexample =  [0 0 0 0 0 0 0 0 0 0 0 0 0 1; ..]'

I am using eye(14) to create the target values. Here I have replaced the [0 1] with [1 2] in the input.

[Greg Heath]

I do not understand why that target is associated with that input.