Functions — github.com/rlcode/reinforcement-learning

hub / github.com/rlcode/reinforcement-learning / functions

Functions200 in github.com/rlcode/reinforcement-learning

⨍ Functions 200◇ Types & classes 36

↓ 1 callersFunctionload_demo

(path)

4-atari-hard/env_robustify.py:227

↓ 1 callersMethodload_state

(self, st)

4-atari-hard/2-go-explore.py:182

↓ 1 callersFunctionmain

()

4-atari-hard/3-robustify.py:125

↓ 1 callersFunctionmain

()

4-atari-hard/extract_demo.py:48

↓ 1 callersFunctionmake_vec_env

Bundle n_envs copies of make_env into a SyncVectorEnv.

3-atari/env.py:108

↓ 1 callersFunctionmake_vec_env

envpool vector env. Returns (n_envs, 4, 84, 84) uint8 obs and accepts int32 actions of shape (n_envs,). `info` is a single dict of per-env arr

4-atari-hard/env.py:184

↓ 1 callersFunctionparse_args

()

4-atari-hard/env_go_explore.py:117

↓ 1 callersFunctionparse_args

()

4-atari-hard/env.py:141

↓ 1 callersFunctionpick_device

(arg="auto")

4-atari-hard/3-robustify.py:38

↓ 1 callersFunctionpick_device

(arg="auto")

4-atari-hard/env.py:209

↓ 1 callersMethodpolicy_evaluation

(self)

1-grid-world/1-policy_iteration.py:18

↓ 1 callersMethodpolicy_improvement

(self)

1-grid-world/1-policy_iteration.py:41

↓ 1 callersMethodpush

(self, frame, action, reward, done)

3-atari/1-dqn.py:72

↓ 1 callersFunctionquit_if_window_closed

(env)

4-atari-hard/env.py:219

↓ 1 callersMethodreconstruct_actions

Walk the prev_id chain back to the root (-1); return actions in forward order. This is how demos are rebuilt for replay/Phase 2.

4-atari-hard/2-go-explore.py:164

↓ 1 callersMethodrecord

(self, starting_point, success)

4-atari-hard/env_robustify.py:204

↓ 1 callersFunctionseed_all

Seed python / numpy / torch for reproducibility (MPS is not bit-exact).

4-atari-hard/env.py:52

↓ 1 callersMethodseed_root

(self, key, snapshot, lives)

4-atari-hard/2-go-explore.py:205

↓ 1 callersMethodstep

(self, action)

4-atari-hard/env_robustify.py:150

↓ 1 callersMethodtrain_model

(self, state, action, reward, next_state, done)

2-cartpole/2-a2c.py:83

↓ 1 callersMethodtrain_model

(self)

2-cartpole/1-dqn.py:98

↓ 1 callersMethodtrain_model

(self)

1-grid-world/6-reinforce.py:82

↓ 1 callersMethodtrain_model

(self, state, action, reward, next_state, next_action, done)

1-grid-world/5-deep_sarsa.py:70

↓ 1 callersMethodunroll

Recompute a (T,B) rollout's logits/values with done-masked GRU state for BPTT. obs_seq (T,B,4,105,80), hx0 (B,gru), done_seq (T,B).

4-atari-hard/3-robustify.py:107

↓ 1 callersMethodupdate

Move rule (atari-reset ResetManager.proc_infos): forward-cumsum the per-index success rates from index 0; the new max starting point is the

4-atari-hard/env_robustify.py:208

↓ 1 callersMethodupdate_from_trajectory

Walk one exploration episode (master-side, serial): append to the experience log, accumulate raw score from the SAMPLING-TIME capture

4-atari-hard/2-go-explore.py:240

↓ 1 callersMethodvalue_iteration

(self)

1-grid-world/2-value_iteration.py:12

↓ 1 callersFunctionwarmup_obs_rms

Step a random agent so obs running stats are realistic before training. Without this, the first intrinsic rewards are wildly scaled and the pr

4-atari-hard/1-ppo-rnd.py:171

(self, capacity, frame_shape=(84, 84), stack=4)

2-cartpole/2-a2c.py:49

Method__init__

(self, state_size, action_size)

2-cartpole/2-a2c.py:61

Method__init__

(self, state_size, action_size)

2-cartpole/3-ppo.py:67

Method__init__

(self, state_size, action_size)

2-cartpole/1-dqn.py:36

Method__init__

(self, state_size, action_size)

2-cartpole/1-dqn.py:55

Method__init__

(self, state_size, action_size)

1-grid-world/6-reinforce.py:33

Method__init__

(self)

1-grid-world/6-reinforce.py:48

Method__init__

(self, state_size, action_size)

1-grid-world/5-deep_sarsa.py:30

Method__init__

(self)

1-grid-world/5-deep_sarsa.py:45

Method__init__

(self, actions)

1-grid-world/4-q_learning.py:8

Method__init__

(self, actions)

1-grid-world/3-sarsa.py:10

Method__init__

(self, title="GridWorld")

1-grid-world/env.py:65

Method__init__

(self, title="DynamicGridWorld", step_penalty=0.0, render_mode="human")

1-grid-world/env.py:158

Method__init__

(self)

1-grid-world/env.py:247

Method__init__

(self, agent, title, buttons=None)

1-grid-world/env.py:280

Method__init__

(self, n_actions, gru_dim=GRU_DIM)

4-atari-hard/3-robustify.py:77

Method__init__

(self, p=0.25)

4-atari-hard/env_robustify.py:46

Method__init__

(self, demo, seed, *, sticky=0.25, allowed_lag=50, allowed_score_deficit=0, reset_steps_ignor

4-atari-hard/env_robustify.py:64

Method__init__

(self, demo, n_envs, *, move_threshold=0.1, nudge=100, window=None)

4-atari-hard/env_robustify.py:185

Method__init__

(self, run_dir, ckpt_every)

4-atari-hard/env_go_explore.py:50

Method__init__

(self, n_actions)

4-atari-hard/1-ppo-rnd.py:71

Method__init__

(self)

4-atari-hard/1-ppo-rnd.py:113

Method__init__

(self, shape=())

4-atari-hard/1-ppo-rnd.py:129

Method__init__

(self, run_dir, ckpt_every)

4-atari-hard/env.py:72

Method__init__

(self, snapshot, score, traj_len, traj_last, lives)

4-atari-hard/2-go-explore.py:83

Method__init__

(self, log_dir, chunk_size=EXPLOG_CHUNK, ancestor_dir=None)

4-atari-hard/2-go-explore.py:99

Method__init__

(self)

4-atari-hard/2-go-explore.py:200

Function_explore_task

task = (snapshot bytes | None for root reset, lives, k, seed). Restore -> up to k steps of repeated random actions; abort on life loss / game

4-atari-hard/2-go-explore.py:313

Function_state

()

4-atari-hard/3-robustify.py:174

Function_state_fn

Full checkpoint state — normalizers / int_filter / update too, so resume is exact.

4-atari-hard/1-ppo-rnd.py:249

Function_state_fn

()

4-atari-hard/2-go-explore.py:371

Function_worker_init

(env_key)

4-atari-hard/2-go-explore.py:308

Methodbest_done_score

(self)

4-atari-hard/2-go-explore.py:209

Methodcheckpoint

Periodic 'latest', 50M-step milestone, and best-gate checkpoints. state_fn() builds the dict only when a save actually happens.

4-atari-hard/env_go_explore.py:78

Methodfinalize

Final 'latest' checkpoint + a final.json result summary.

4-atari-hard/env_go_explore.py:93

2-cartpole/2-a2c.py:43

Methodforward

(self, x)

2-cartpole/2-a2c.py:56

Methodforward

(self, x)

2-cartpole/3-ppo.py:81

Methodforward

(self, x)

2-cartpole/1-dqn.py:50

Methodforward

(self, x)

1-grid-world/6-reinforce.py:43

Methodforward

(self, x)

1-grid-world/5-deep_sarsa.py:40

Methodforward

(self, x)

4-atari-hard/1-ppo-rnd.py:84

Methodforward

(self, x)

4-atari-hard/1-ppo-rnd.py:106

Methodforward

(self, x)

4-atari-hard/1-ppo-rnd.py:122

Methodget_action

(self, state)

1-grid-world/1-policy_iteration.py:78

Methodget_transition_prob

(self, state, action)

1-grid-world/env.py:264

Methodlog

Append one structured row (frames + sps + caller's scalars) to metrics.jsonl.

4-atari-hard/env_go_explore.py:60

Functionmake_env

Single gymnasium env with the standard Atari preprocessing. Used for `--test` rendering; envpool has no human render mode.

4-atari-hard/env.py:169

Methodmax_archive_score

(self)

4-atari-hard/2-go-explore.py:214

Functionon_calculate

()

1-grid-world/2-value_iteration.py:64

Functionon_clear

()

1-grid-world/2-value_iteration.py:84

Functionon_evaluate

()

1-grid-world/1-policy_iteration.py:103

Functionon_improve

()

1-grid-world/1-policy_iteration.py:107

Functionon_move

()

1-grid-world/1-policy_iteration.py:111

Functionon_move

()

1-grid-world/2-value_iteration.py:81

Functionon_print_policy

()

1-grid-world/2-value_iteration.py:68

Functionon_reset

()

1-grid-world/1-policy_iteration.py:114

Functionpick

(state)

2-cartpole/3-ppo.py:116

Functionpolicy_action

1-grid-world/env.py:168

Methodreset

(self)

4-atari-hard/env_robustify.py:50

Methodresolve_resume

auto' -> run_dir/ckpt/latest.pt, else a path, else None.

4-atari-hard/env_go_explore.py:69

Functionrun_test_loop

Replay episodes forever using the supplied action picker (single env).

4-atari-hard/env.py:228

Methodstate

(self)

4-atari-hard/2-go-explore.py:176

Methodstep

(self, action)

1-grid-world/env.py:179

← previous101–200 of 200, ranked by callers