During DQN training, it is possible to configure a certain number of workers to fill in the replay_buffer until reaching learning_starts, and after that decrease the number of workers?
Related topics
Topic | Replies | Views | Activity | |
---|---|---|---|---|
DQN (and maybe other algo) should take into account the "num_envs_per_worker" config when computing the round robin native_ratio used to determined the number of steps to use for training | 3 | 477 | April 21, 2021 | |
Accessing the memory buffer dqn | 10 | 1002 | January 16, 2022 | |
[RLlib] Effect of num_cpus_for_driver? | 1 | 750 | February 24, 2021 | |
Training steps for DQN | 3 | 113 | April 18, 2024 | |
Change learning rete for DQN | 6 | 508 | February 25, 2022 |