[theano-users] Minibatch size and RAM usage

[Luca Morselli]

Hi to all, a theano newbie here.

I'm trying to training a MLP on MNIST dataset with mini-batch sgd (using
cpu) following deeplearning.net tutorial.
The default batch-size is set to 20 but when I launch the script after 2
epoch I'm out of ram.
I noticed that every time train_model(index) is called something is stored
in RAM. If I use all of training data this doesn't happen.

What I'm missing?

Thanks.