search

georgeliu233 / Scene-Rep-Transformer

[T-IV] Augmenting Reinforcement Learning with Transformer-based Scene Representation Learning for Decision-making of Autonomous Driving
https://georgeliu233.github.io/Scene-Rep-Transformer/
48 stars 10 forks source link

How to train? #5

Closed zygalaxy closed 4 weeks ago

zygalaxy commented 1 month ago

How should I train? I don't see the corresponding train.py

12jenny commented 1 month ago

Have you solved the problem yet?

Cyanopica321 commented 4 weeks ago

I also can't find the training code, have you solved it?

12jenny commented 4 weeks ago

Not yet.

georgeliu233 commented 4 weeks ago

Hi all, Sorry for the late reply by other projects I am recently busy at!

Yes, currently the repo only contains the testing part of all algos, and I plan to collate the rest part of the project by the next couple of months when available. It is based on the original trainer from tf2rl.

Bset, Haochen

georgeliu233 commented 4 weeks ago

You may access the training pipelines in ./envs/runners, and full algorithm in ./algos first, for your own development.

12jenny commented 4 weeks ago

Hi all, Sorry for the late reply by other projects I am recently busy at!

Yes, currently the repo only contains the testing part of all algos, and I plan to collate the rest part of the project by the next couple of months when available. It is based on the original trainer from tf2rl.

Bset, Haochen

And may I ask how your model was trained?

georgeliu233 commented 4 weeks ago

Hi Jenny,

Pretty straightforward: 1) initialize env (smarts/carla) 2) initalize the algos as in ./algo, and 3) wrap both env and algo in the corresponding trainer (./envs/runners). call the trainer to train the model.

12jenny commented 4 weeks ago

Hi Jenny,

Pretty straightforward: 1) initialize env (smarts/carla) 2) initalize the algos as in ./algo, and 3) wrap both env and algo in the corresponding trainer (./envs/runners). call the trainer to train the model.

Okay, thanks for the answer. I'll give it a try, wish me luck! Hopefully I can contact you afterwards!