search

masato-ka / airc-rl-agent

AI RC Car Agent that using deep reinforcement learning on Jetson Nano
MIT License
86 stars 24 forks source link

steps per second e.g. FPS = 0 while training #23

Closed newten82 closed 4 years ago

newten82 commented 4 years ago

While training with jetbot the FPS drops to 0. At most I'm only seeing 2 FPS. Both seem low. What is an expected FPS? Or how can I increase it?

`----------------------------------------- | current_lr | 0.0003 | | ent_coef | 0.023513975 | | ent_coef_loss | 0.19573997 | | entropy | 1.6002797 | | episodes | 35 | | fps | 0 | | mean 100 episode reward | 131 | | n_updates | 19800 | | policy_loss | -19.272217 | | qf1_loss | 2.0728447 | | qf2_loss | 1.9528334 | | time_elapsed | 8822 | | total timesteps | 4635 | | value_loss | 0.9012158 |

`

masato-ka commented 4 years ago

FPS is calculated including the time for slope calculation. Jetson nano requires a lot of time to calculate the gradient, so the FPS is very low. This value is calculated by stable_baseline2. If JetBot is working fine then don't worry.