delta_W_rec dimensions

[eyaler]

in rnn, shouldn't: self.delta_W_rec = theano.shared(value = numpy.zeros((n_in,n_out), dtype=theano.config.floatX), name='delta_W_rec') be: self.delta_W_rec = theano.shared(value = numpy.zeros((n_out,n_out), dtype=theano.config.floatX), name='delta_W_rec') ?

[ghost]

The computation is T.dot(input, W), so the dimension is (n_in, n_out)

[eyaler]

not T.dot(h_tm1, self.W_rec) ?

[ghost]

"input" symbolically means inputs to a layer, not just the original features. It of course can refer to h_tm1