Stuck on an issue?

Lightrun Answers was designed to reduce the constant googling that comes with debugging 3rd party libraries. It collects links to all the places you might be looking at while hunting down a tough bug.

And, if you’re still stuck at the end, we’re happy to hop on a call to see how we can help out.

what is the world_size argument in SerialSampler?

See original GitHub issue

SerialSampler.initialize takes the argument world_size what does this do and how do you set it? It seems to have something to do with DistributedDataParallel in agent…

Issue Analytics

State:
Created 4 years ago
Comments:5 (2 by maintainers)

Top GitHub Comments

2reactions

astookecommented, Oct 25, 2019

That’s right, world size is relevant when using the SyncRl runner, which runs multi-GPU.

1reaction

renqibingcommented, Oct 25, 2019

By the way, can you tell me what does bootstrap_timelimit do? Does it correspond with pretraining? I see all of q function policy gradient algorithms set it as true.

Yes this is from the original SAC implementation, it makes the algorithm use the next Q value in the Q-learning target when the episodes ends due to timeout. Normally, when an episode is done, the next Q value would not be used, only the reward. Intuitively, the time limit is an arbitrary implementation decision, but the agent should learn as if it can run forever. Not to do with pretraining. Try it with and without, it makes a difference on e.g. HalfCheetah. (good idea to ask in separate thread in the future, so others can find it 😃

Thanks for your sincere help!