How to run algorithms where there isn't a need for dataloaders?

Lightning-AI / pytorch-lightning

Pretrain, finetune and deploy AI models on multiple GPUs, TPUs with zero code changes.

https://lightning.ai

Apache License 2.0

27.93k stars 3.34k forks source link

How to run algorithms where there isn't a need for dataloaders? #2182

Closed nsidn98 closed 4 years ago

nsidn98 commented 4 years ago

What is your question?

In on-policy algorithms in reinforcement learning, rollouts are generated on the fly and there is no need for a replay buffer and consequently a dataloader. In these cases, the loss function is calculated according to the current states obtained (generally by using multiple parallel environments).

Hence, is it possible to bypass the dataloader requirement for lightning?

williamFalcon commented 4 years ago

Check these examples out:

https://github.com/djbyrne/core_rl

nsidn98 commented 4 years ago

@williamFalcon the examples in the linked repository does not have on-policy algorithms. Is there a way to bypass the need of train_dataloader() in the lightning module?

stale[bot] commented 4 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.