MCPcopy Create free account

hub / github.com/rlcode/reinforcement-learning / types & classes

Types & classes36 in github.com/rlcode/reinforcement-learning

⨍ Functions 200◇ Types & classes 36
↓ 4 callersClassEnv
Static 5x5 grid for tabular SARSA / Q-learning. agent=[col,row].
1-grid-world/env.py:60
↓ 3 callersClassCell
Archive entry. snapshot/lives describe the state AT this cell so a worker can restore and keep exploring; traj_last points into the experience
4-atari-hard/2-go-explore.py:76
↓ 2 callersClassActorCritic
3-atari/2-ppo.py:41
↓ 2 callersClassDynamicEnv
5x5 grid with horizontally-bouncing obstacles (Deep SARSA, REINFORCE). Goal at (4,4) terminates; obstacle hit costs -1 but continues. State:
1-grid-world/env.py:147
↓ 2 callersClassGraphicDisplay
Pygame button-driven viewer for policy / value iteration. Set `display.buttons = [(label, handler[, enabled]), ...]` (up to 4). `enabled` is
1-grid-world/env.py:269
↓ 2 callersClassQNetwork
3-atari/1-dqn.py:39
↓ 2 callersClassQNetwork
2-cartpole/1-dqn.py:35
↓ 2 callersClassReplayResetEnv
One raw ALE env that starts episodes from demo states. Not a gym env — the vectorized loop in 3-robustify.py drives it directly.
4-atari-hard/env_robustify.py:60
↓ 2 callersClassRunLogger
Optional run-directory outputs: metrics.jsonl, periodic / milestone / best checkpoints, resume, and a final.json summary. Inert when run_dir is
4-atari-hard/env_go_explore.py:38
↓ 2 callersClassRunningMeanStd
Welford / Chan parallel algorithm. Used for both obs (last frame) and intrinsic-return scaling.
4-atari-hard/1-ppo-rnd.py:126
↓ 1 callersClassA2CAgent
2-cartpole/2-a2c.py:60
↓ 1 callersClassActor
2-cartpole/2-a2c.py:35
↓ 1 callersClassActorCritic
2-cartpole/3-ppo.py:66
↓ 1 callersClassActorCriticRND
4-atari-hard/1-ppo-rnd.py:70
↓ 1 callersClassArchive
Cell store + novelty-weighted selection + the accept rule. All updates happen serially in the master process.
4-atari-hard/2-go-explore.py:196
↓ 1 callersClassCritic
2-cartpole/2-a2c.py:48
↓ 1 callersClassDQNAgent
2-cartpole/1-dqn.py:54
↓ 1 callersClassDeepSARSAgent
1-grid-world/5-deep_sarsa.py:44
↓ 1 callersClassExperienceLog
Append-only step log as a prev_id linked list (design note 5). RAM holds only the active chunk; full chunks flush to <dir>/chunk_NNNNN.npz (c
4-atari-hard/2-go-explore.py:92
↓ 1 callersClassFireResetEnv
3-atari/env.py:24
↓ 1 callersClassGRUActorCritic
conv 8/4/3 -> fc + LayerNorm -> GRUCell -> pi, V. Input = 4-stacked 105x80 grayscale frames (4 channels).
4-atari-hard/3-robustify.py:73
↓ 1 callersClassLifeLossTerminalEnv
3-atari/env.py:37
↓ 1 callersClassPolicyEnv
Pure-data MDP for policy/value iteration. state = [row, col].
1-grid-world/env.py:242
↓ 1 callersClassPolicyIteration
1-grid-world/1-policy_iteration.py:6
↓ 1 callersClassPolicyNetwork
1-grid-world/6-reinforce.py:32
↓ 1 callersClassQLearningAgent
1-grid-world/4-q_learning.py:7
↓ 1 callersClassQNetwork
1-grid-world/5-deep_sarsa.py:29
↓ 1 callersClassRNDPredictor
Slightly deeper than target — two extra ReLU FCs so it has the capacity to actually fit the target's random projection.
4-atari-hard/1-ppo-rnd.py:110
↓ 1 callersClassRNDTarget
4-atari-hard/1-ppo-rnd.py:100
↓ 1 callersClassReinforceAgent
1-grid-world/6-reinforce.py:47
↓ 1 callersClassReplayBuffer
Single-frame uint8 buffer — stacks of 4 are reconstructed at sample time, cutting RAM ~4x vs. storing the full stack per slot.
3-atari/1-dqn.py:58
↓ 1 callersClassResetManager
Owns the shared curriculum across N envs. The trainer calls assign() once to stagger starting points, and update() each time a batch of episodes
4-atari-hard/env_robustify.py:180
↓ 1 callersClassRunLogger
Optional run-directory outputs: metrics.jsonl, periodic / 5M-milestone / best checkpoints, resume, and a final.json summary. Inert when run_dir is
4-atari-hard/env.py:67
↓ 1 callersClassSARSAgent
1-grid-world/3-sarsa.py:9
↓ 1 callersClassStickyActionEnv
repeat_action_probability applied BELOW frameskip — sticky at the raw action level, the standard v5 stochasticity. We build the ALE env with s
4-atari-hard/env_robustify.py:40
↓ 1 callersClassValueIteration
1-grid-world/2-value_iteration.py:4