FloatingPointError while training about dup-ocropy HOT 5 CLOSED

The default parameters should work without problems. Can you say more about your environment (Python version, OS, compiler version, etc.)?

You might want to give CLSTM a try as well; it's the same algorithm in C++.

from dup-ocropy.

My environment:
python 2.7.6
Ubuntu 14.04.1 LTS \n \l
gcc version 4.8.4

from dup-ocropy.

I'm getting the same errors. I've tried clipping the values, but the model seems keeping very high error rates (I'm experimenting with training samples taken from a mobile camera, since we want to recognize receipts uploaded with mobiles)

def ffunc(x):
    "Nonlinearity used for gates."
    # cliping to avoid overflows
    return 1.0/(1.0+exp(clip(-x, None, 700)))

from dup-ocropy.

Sorry for not responding earlier. Can you provide a backtrace?

You might also add a line of python to the main program to turn overflows into errors (when I get around to it, I may do this myself).

from dup-ocropy.

Should be fixed with #201.

from dup-ocropy.