machine learning - Sequential training data for XOR Recurrent Neural Network -

i trying implement recurrent neural network , trying learn xor function petty example.

as recurrent network, thought have work 1 input unit in order see how remembering previous state; is, implementing xor function based on sequential input:

input(t-1) = 0 input(t)   = 1 output(t)  = 1 

or

input(t-1) = 1 input(t)   = 1 output(t)  = 0 

so input training data presented 1 bit @ time in order:

inputs = { 0, 0, 1, 1, 0 } 

and corresponding target output

targets = { 0, 0, 1, 0, 1 }. 

but not learning and, though know there can many reasons that, wondering maybe did not define dataset , wouldn't presenting right problem network. come here looking ideas on right training set supervised learning of "sequential" xor function.

the implementation working on similar elman rnn, if need details on it, please ask.

the following link: https://code.google.com/p/encog-java/source/browse/trunk/encog-examples/src/org/encog/examples/neural/recurrent/elman/elmanxor.java?r=2423

contains recurrent elman network implementation works xor problem. uses encog (which relatively easy use) handle implementation of network itself. able see difference in implementation , correct.