Making predictions with a different number of data points than used for estimating the parameters

[junpenglao]

If you are using a theano.shared variable as input, you can do that by input_X.set_value, as long as there is no shape information related to this input_X is hard coded in your model.