Dimensionality problem

[AlbertoArdu]

Hello! I' trying to modify a bit you're code to make the RNN return not an index of a word, but an array. That is, my input will be an array of floats and I want the RNN estimate the next one. I encountered some problems: there is a warning compiling the forard_prop_step function inside theano.scan that tells that all parameters should be specified in the sequence or non_sequence field. In fact I can't see where the function gets s_t_prev when executes. Can you clarify this point?