From 97c327f9963947481475c4ec7eb4960c02a1d702 Mon Sep 17 00:00:00 2001 From: MishaLaskin Date: Sat, 8 Feb 2020 17:51:38 -0800 Subject: [PATCH] readme mods --- README.md | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/README.md b/README.md index 819027c..5aad476 100644 --- a/README.md +++ b/README.md @@ -17,10 +17,10 @@ CUDA_VISIBLE_DEVICES=0 python train.py \ In your console, you should see printouts that look like: ``` -| train | E: 221 | S: 28000 | D: 18.1 s | R: 785.2634 | BR: 3.8815 | ALOSS: -305.7328 | CLOSS: 190.9854 | RLOSS: 0.0000 -| train | E: 225 | S: 28500 | D: 18.6 s | R: 832.4937 | BR: 3.9644 | ALOSS: -308.7789 | CLOSS: 126.0638 | RLOSS: 0.0000 -| train | E: 229 | S: 29000 | D: 18.8 s | R: 683.6702 | BR: 3.7384 | ALOSS: -311.3941 | CLOSS: 140.2573 | RLOSS: 0.0000 -| train | E: 233 | S: 29500 | D: 19.6 s | R: 838.0947 | BR: 3.7254 | ALOSS: -316.9415 | CLOSS: 136.5304 | RLOSS: 0.0000 +| train | E: 221 | S: 28000 | D: 18.1 s | R: 785.2634 | BR: 3.8815 | A_LOSS: -305.7328 | CR_LOSS: 190.9854 | CU_LOSS: 0.0000 +| train | E: 225 | S: 28500 | D: 18.6 s | R: 832.4937 | BR: 3.9644 | A_LOSS: -308.7789 | CR_LOSS: 126.0638 | CU_LOSS: 0.0000 +| train | E: 229 | S: 29000 | D: 18.8 s | R: 683.6702 | BR: 3.7384 | A_LOSS: -311.3941 | CR_LOSS: 140.2573 | CU_LOSS: 0.0000 +| train | E: 233 | S: 29500 | D: 19.6 s | R: 838.0947 | BR: 3.7254 | A_LOSS: -316.9415 | CR_LOSS: 136.5304 | CU_LOSS: 0.0000 ``` The maximum score for cartpole swing up is around 845 pts. Notice how CURL solves visual cartpole in 30k steps! This takes about and hour of training depending on your GPU. For reference, the state-state-of-the-art end-to-end method D4PG takes 50,000,000 timesteps to solve the same problem. CURL is ~1000x more efficient!