DeepLearningTutorials

I changed the code to do a validation test every epoch. While this is not the best option, it was straight forward to implement, and from the discussion in the emails I got the feeling there is no one way of doing this well .. I think we should add all the comments we had in the emails in the tutorial .. as rules of thumb [in the early stopping section] of what one should consider and leave aceeptable parameters in the code. Does anyone has a strong feeling of certain parameters to which I need to change the code ?

Minibatches of size 20

minor fixes in text

function([x,y], y_pred) --> function(x, y_pred); and other minor fixes