Small vocab errors? about char-rnn HOT 3 CLOSED

See this line:

data load done. Number of batches in train: 2, val: 0, test: 1

Somehow only 3 distinct batches of data were read in, and validation set has 0 (which is why you're seeing NaN - there is no validation data).

This makes it seem like you have a very tiny dataset, and that it's somehow not being read in correctly.

from char-rnn.

Ah, thanks for the pointer. I had inadvertently started training with a very small input.txt, and did not realize I had to delete the *.t7 files once I switched to a new, larger dataset. Once I deleted those, training worked as expected. (Though now I'm thinking back to other times I started with smaller datasets and then retrained with a larger set without deleting those files... ouch.)

from char-rnn.

Doh, see I tried to be clever and cache the files, I didn't think people would be swapping out the data in place. Hmmmmmm. Interesting, thanks for reporting back. Maybe I should do some checksumy things.

from char-rnn.