search

yaserkl / RLSeq2Seq

Deep Reinforcement Learning For Sequence to Sequence Models
https://arxiv.org/abs/1805.09461
MIT License
767 stars 160 forks source link

can you offer your pretrain model? #11

Open 457138317 opened 6 years ago

457138317 commented 6 years ago

i have tried your project for rl-learning, however,it's result is not so good. so, can you offer a pretrained model of rl-learning here?

yaserkl commented 6 years ago

I'm not expecting to see an improvement in the results if you use the default hyperparameters in this package. This package is based on Pointer-Generator model. Please not that Paulus et al. (https://arxiv.org/pdf/1705.04304.pdf) is using a completely different set of hyperparameters (check out the section at the very end of the above paper) than the default parameters used in this package. So, if you want to replicate their model, you have to use the same hyperparamters that they used in the paper. We will publish our own trained models (using the default parameter in this package) this coming week. We will also provide another model which uses exactly the same set of hyperparameters that Paulus et al. used in their paper in the next few weeks.

457138317 commented 6 years ago

thank u for your reply.

jonaseltes commented 5 years ago

Any pretrained models shared yet? :)

sudharavali commented 5 years ago

I just wanted to follow up on if you have the pretrained models up for it. Let me know !

git4sun commented 5 years ago

Any updates on the pretrained model?