ray/rllib/tuned_examples/impala/pong-impala-vectorized.yaml

# This can reach 18-19 reward within 10 minutes on a Tesla M60 GPU (e.g., G3 EC2 node)
# with 32 workers and 10 envs per worker. This is more efficient than the non-vectorized
# configuration which requires 128 workers to achieve the same performance.
pong-impala-vectorized:
    env: PongNoFrameskip-v4
    run: IMPALA
    config:
        rollout_fragment_length: 50
        train_batch_size: 500
        num_workers: 32
        num_envs_per_worker: 10
[rllib] Basic IMPALA implementation (using deepmind's reference vtrace.py) (#2504) Rename AsyncSamplesOptimizer -> AsyncReplayOptimizer Add AsyncSamplesOptimizer that implements the IMPALA architecture integrate V-trace with a3c policy graph audit V-trace integration benchmark compare vs A3C and with V-trace on/off PongNoFrameskip-v4 on IMPALA scaling from 16 to 128 workers, solving Pong in <10 min. For reference, solving this env takes ~40 minutes for Ape-X and several hours for A3C. 2018-08-01 20:53:53 -07:00			`# This can reach 18-19 reward within 10 minutes on a Tesla M60 GPU (e.g., G3 EC2 node)`
			`# with 32 workers and 10 envs per worker. This is more efficient than the non-vectorized`
			`# configuration which requires 128 workers to achieve the same performance.`
			`pong-impala-vectorized:`
			`env: PongNoFrameskip-v4`
			`run: IMPALA`
			`config:`
[rllib] Rename sample_batch_size => rollout_fragment_length (#7503) * bulk rename * deprecation warn * update doc * update fig * line length * rename * make pytest comptaible * fix test * fi sys * rename * wip * fix more * lint * update svg * comments * lint * fix use of batch steps 2020-03-14 12:05:04 -07:00			`rollout_fragment_length: 50`
[rllib] Basic IMPALA implementation (using deepmind's reference vtrace.py) (#2504) Rename AsyncSamplesOptimizer -> AsyncReplayOptimizer Add AsyncSamplesOptimizer that implements the IMPALA architecture integrate V-trace with a3c policy graph audit V-trace integration benchmark compare vs A3C and with V-trace on/off PongNoFrameskip-v4 on IMPALA scaling from 16 to 128 workers, solving Pong in <10 min. For reference, solving this env takes ~40 minutes for Ape-X and several hours for A3C. 2018-08-01 20:53:53 -07:00			`train_batch_size: 500`
			`num_workers: 32`
			`num_envs_per_worker: 10`