search

macournoyer / neuralconvo

Neural conversational model in Torch
776 stars 347 forks source link

fix for issue 52: cpu mode epochs #76

Open TTN- opened 7 years ago

TTN- commented 7 years ago

Applying fix for issues: https://github.com/macournoyer/neuralconvo/issues/52 https://github.com/macournoyer/neuralconvo/issues/75

TTN- commented 7 years ago

EDIT: ignore that last commit for now, its semi broken.

I've just added one more commit to my fork, not sure if you'd like to have it in yours too. I find that the --dataset size parameter isn't listened to at all if the dataset has already been generated. So if I run it once with --dataset 1000, then later again with --dataset 80000 later, I will still actually be using the 1000 dataset. Generating the dataset takes only a fraction of the time needed to train a network, so it is easiest to fix this by simply generating the dataset on each run as a workaround. Like I said maybe you dont want to include it here, but its there if you like it.. Just want to say though, this project is amazing, I'm toying around with it a lot. Very interesting to test on.