In Minibatch you can pass a seed to make sure at each iteration it use the same index to get subset of data (that’s how you can make sure X_minibatch and Y_minibatch is the same row in a regression).
But if you are having doubt about the indexing, you can also following the docstring in minibatch to make it explicit:
To be more concrete about how we get minibatch, here is a demo
create shared variable >>> shared = theano.shared(data)
create random slice of size 10 >>> ridx = pm.tt_rng().uniform(size=(10,), low=0, high=data.shape[0]-1e-10).astype(‘int64’)
take that slice >>> minibatch = shared[ridx]
That’s done. Next you can use this minibatch somewhere else. You can see that implementation does not require fixed shape for shared variable. Feel free to use that if needed.