search

rpryzant / delete_retrieve_generate

PyTorch implementation of the Delete, Retrieve Generate style transfer algorithm
MIT License
132 stars 26 forks source link

The bleu metric value is always 0 #1

Closed Enihsuns closed 5 years ago

Enihsuns commented 5 years ago

When using the sample_config.json to run the code, the bleu metric value is always 0. Why?

I use the same command provided in Readme to run the code and I didn't alter the code.

My log: 2019-03-11 13:22:21,646 - INFO - EPOCH: 15 ITER: 48.0/50.0 WPS: 175.62 LOSS: 6.4455 METRIC: 0.0000 2019-03-11 13:22:21,759 - INFO - EPOCH: 15 ITER: 49.0/50.0 WPS: 178.09 LOSS: 6.6800 METRIC: 0.0000 2019-03-11 13:22:21,760 - INFO - EPOCH 15 COMPLETE. EVALUATING... 80/100...2019-03-11 13:22:26,620 - INFO - METRIC: 0. TIME: 4.86s CHECKPOINTING...

Thank you!

rpryzant commented 5 years ago

This is because the sample data (and training steps) are too small to learn anything useful from. If you look at losses (6.6800) the model hadn't converged either.

I've updated the readme with a more substantial example experiment. Marking as this as closed.