search

rail-berkeley / serl

SERL: A Software Suite for Sample-Efficient Robotic Reinforcement Learning
https://serl-robot.github.io/
MIT License
375 stars 42 forks source link

Question about the hyperparameters used in the experiments #55

Closed BurakDmb closed 5 months ago

BurakDmb commented 5 months ago

Hi,

Thank you for your efforts in the SERL paper and the repo.

I want to ask if you used the same hyperparameters in the experiments in the paper with the examples in this repo.

If different, what values did you use for these parameters: Network size, discount rate, temperature_init, utd_ratio, and the encoder type?

youliangtan commented 5 months ago

IIRC all the hyperparameters provided here are default used in those experiments. @Leo428 can correct me if Im wrong.

Leo428 commented 5 months ago

hi @BurakDmb, the hyperparams for each task and experiment is specified in the .sh files. For example, here