search

MishaLaskin / curl

CURL: Contrastive Unsupervised Representation Learning for Sample-Efficient Reinforcement Learning
MIT License
561 stars 88 forks source link

WARN precision #6

Closed Rivendile closed 4 years ago

Rivendile commented 4 years ago

I got WARN: Box bound precision lowered by casting to float32 when I run the code. Should this be fixed?

MishaLaskin commented 4 years ago

Should be ok, I get this warning as well on some machines - the method is not sensitive to float64 -> 32 casting

On Sun, May 10, 2020 at 9:48 PM Ian Zhao notifications@github.com wrote:

I got WARN: Box bound precision lowered by casting to float32 when I run the code. Should this be fixed?

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/MishaLaskin/curl/issues/6, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABHWQWJQFZ6PXF5HQCZYASTRQ5KOXANCNFSM4M5QBMVA .

Rivendile commented 4 years ago

Thanks a lot :)

Rivendile commented 4 years ago

Hi, when I run the task "Cheetah, Run", it costs about 65s for every four episodes. Is it ok or did I make some mistakes? I run the code on TITAN RTX. The run.sh is: CUDA_VISIBLE_DEVICES=2 python train.py \ --domain_name cheetah \ --task_name run \ --encoder_type pixel \ --action_repeat 4 \ --save_tb --pre_transform_image_size 100 --image_size 84 \ --work_dir ./tmp \ --agent curl_sac --frame_stack 3 \ --seed -1 --critic_lr 2e-4 --actor_lr 2e-4 --encoder_lr 2e-4 --eval_freq 10000 --batch_size 512 --num_train_steps 100000

MishaLaskin commented 4 years ago

That's normal - large batch sizes take a long time to train

On Thu, May 14, 2020 at 1:48 AM Ian Zhao notifications@github.com wrote:

Hi, when I run the task "Cheetah, Run", it costs about 65s for every four episodes. Is it ok or did I make some mistakes? I run the code on TITAN RTX. The run.sh is: CUDA_VISIBLE_DEVICES=2 python train.py --domain_name cheetah --task_name run --encoder_type pixel --action_repeat 4 --save_tb --pre_transform_image_size 100 --image_size 84 --work_dir ./tmp --agent curl_sac --frame_stack 3 --seed -1 --critic_lr 2e-4 --actor_lr 2e-4 --encoder_lr 2e-4 --eval_freq 10000 --batch_size 512 --num_train_steps 100000

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/MishaLaskin/curl/issues/6#issuecomment-628401763, or unsubscribe https://github.com/notifications/unsubscribe-auth/ABHWQWMIPKW2KTW7OT3CDU3RROA37ANCNFSM4M5QBMVA .

Rivendile commented 4 years ago

Thanks for your reply!