search

conglu1997 / v-d4rl

Challenges and Opportunities in Offline Reinforcement Learning from Visual Observations
MIT License
94 stars 9 forks source link

Questions about the number of eval episodes #15

Closed return-sleep closed 8 months ago

return-sleep commented 8 months ago

As the paper noted, the experimental results is averaged over six random seeds. Can I ask how many eval_episodes were used for each methods (DV2, CQL,et al) in the evaluation phase, as I found the visual input settings（V-D4RL） are more unstable compared to proprioceptive states (D4RL).

conglu1997 commented 8 months ago

Please refer to these lines where this is set: https://github.com/conglu1997/v-d4rl/blob/9dcca0984faaddaaedce6975f4d9e6ef5977ad6f/drqbc/cfgs/config.yaml#L16

https://github.com/conglu1997/v-d4rl/blob/9dcca0984faaddaaedce6975f4d9e6ef5977ad6f/offlinedv2/dreamerv2/configs.yaml#L18