v0.2.0

Environment

1. Add Metadrive environment and its configurations (#192)
2. Add Sampled MuZero/UniZero and DMC environment with related configurations (#260)
3. Polish Chess environment and its render method; add unit tests and configurations (#272)
4. Add Jericho environment and its related configurations (#307)

Algorithm

1. Add Harmony Dream loss balance in MuZero (#242)
2. Adopt AlphaZero for non-zero-sum games (#245)
3. Add AlphaZero CTree unittest (#306)
4. Add recent MCTS-related papers (#324)
5. Introduce rope to use the true timestep index as pos_index (#266)
6. Add Jericho DDP configuration (#337)

Enhancement

1. Add LightZero Sphinx documentation (#237)
2. Add Wandb support (#294)
3. Add Atari100k metric utilities (#295)
4. Add eval_benchmark tests (#296)
5. Include save_replay and collect_episode_data options in Jericho (#333)
6. Add an MCTS TicTacToe demo in a single file (#315)

Polish

1. Polish efficiency and performance on Atari and DMC (#292)
2. Update requirements (#298)
3. Optimize reward/value/policy_head_hidden_channels (#314)
4. Update configuration and log instructions in tutorials (#330)

Fix

1. Fix DownSample issues for different observation shapes (#254)
2. Fix the wrong chance values in Stochastic MuZero (#275)
3. Use display_frames_as_gif in CartPole (#288)
4. Fix the chance encoder in stochastic_muzero_model_mlp.py (#284)
5. Correct typo in model/utils.py (#290)
6. Fix SMZ compile_args and num_simulations bug in world_model (#297)
7. Fix reward type bug in 2048 and OS import issue in CartPole (#304)
8. Switch to macos-13 in action (#319)
9. Fix SMZ & SEZ config for pixel-based DMC (#322)
10. Fix update_per_collect in DDP setting (#321)
11. Fix bug with obs_shape tuple in initialize_zeros_batch (#327)
12. Fix prepare_obs_stack_for_unizero issue (#328)
13. Fix random_policy when len(ready_env_id) < collector_env_num (#335)
14. Fix timestep compatibility issues (#339)

CI & Test

1. Add self-hosted Linux (Ubuntu) CI runner (#259)
2. Add self-hosted Linux runner for CI tests (#323)

---

Full Changelog: v0.1.0...v0.2.0

Contributors: @ruiheng123 @TuTuHuss @HarryXuancy @ShivamKumar2002 @Roland0511 @cmarlin @xiongjyu @PaParaZz1 @puyuan1996

v0.1.0

Environment

1. SumToThree Env from pooltool (#227)

Algorithm

1. UniZero (#232)
2. ReZero (#238)

Enhancement

1. add logs and config documentations (#220)
2. polish atari_env_action_space_map, fix test_muzero_game_buffer
3. polish release.yml

Style

1. update discord link and add a new badge in readme (#221)

Full Changelog: v0.0.5...v0.1.0

Contributors: @ekiefl @TuTuHuss @HarryXuancy @PaParaZz1 @puyuan1996

v0.0.5

Environment

1. MemoryEnv (#197)
2. MountainCar (#181)

Algorithm

1. Gumbel AlphaZero in ctree (#212)

Enhancement

1. add eval_offline option (#188)
2. save the updated searched policy and value to the buffer during reanalyze (#190)
3. add muzero visualization (#181)
4. add efficientzero tictactoe configs (#204)
5. add 2 mcts related iclr2024 papers
6. add load pretrained model option in test_game_segment (#194)
7. polish _forward_learn() and some data process operations (#191)

Fix

1. fix sync_gradients and log in DDP settings (#200)
2. fix channel_last bug
3. fix total_episode_count bug in collector
4. fix memory_lightzero_env return bug
5. fix obs_max_scale bug in memory_env

Style

1. add ZeroPal and discord link (#209)
2. add unittest for game_buffer_muzero (#186)
3. add customization documentation section in readme

Full Changelog: v0.0.4...v0.0.5

Contributors: @karroyan @HarryXuancy @nighood @puyuan1996

v0.0.4

Enhancement

1. add agent configurations & polish replay video saving method (#184)
2. polish comments in worker files
3. polish comments in tree search files (#185)
4. rename mcts_mode to battle_mode_in_simulation_env, add sampled alphazero config for tictactoe (#179)
5. polish redundant data squeeze operations (#177)
6. polish the continuous action process in sez model
7. polish bipedalwalker env

Fix

1. fix completed value inf bug when zero exists in action_mask in gumbel muzero (#178)
2. fix render settings when using gymnasium (#173)
3. fix lstm_hidden_size in sampled_efficientzero_model.py
4. fix action_mask in bipedalwalker_cont_disc_env, fix device bug in sampled efficientzero (#168)

Full Changelog: v0.0.3...v0.0.4

Contributors: @karroyan @HarryXuancy @puyuan1996 @zjowowen

v0.0.3

Env

1. MiniGrid env (#110)
2. Bsuite env (#110)
3. GoBigger env (#39)

Algorithm

1. Sampled AlphaZero (#141)
2. MuZero+RND (#110)
3. Multi-Agent MuZero/EfficientZero (#39)

Enhancement

1. add ctree version of mcts in alphazero (#142)
2. upgrade the dependency on gym with gymnasium (#150)
3. add agent class to support LightZero's HuggingFace Model Zoo (#163)
4. add recent MCTS-related papers in readme (#159)
5. add muzero config for connect4 (#107)
6. add CONTRIBUTING.md (#119)
7. add .gitpod.yml and .gitpod.Dockerfile (#123)
8. add contributors subsection in README (#132)
9. add CODE_OF_CONDUCT.md (#127)
10. polish comments and render_eval configs for various common envs (#154) (#161)
11. polish action_type and env_type, fix test.yml, fix unittest (#160)
12. update env and algo tutorial doc (#106)
13. polish gomoku env (#141)
14. add random_policy support for continuous env (#118)
15. polish simulation method of ptree_az (#120)
16. polish comments of game_segment_to_array

Fix

1. fix render method for various common envs (#154) (#161)
2. fix gumbel muzero collector bug, fix gumbel typo (#144)
3. fix assert bug in game_segment.py (#138)
4. fix visit_count_distributions name in muzero_evaluator
5. fix mcts and alphabeta bot unittest (#120)
6. fix typos in ptree_mz.py (#113)
7. fix root_sampled_actions_tmp shape bug in sez ptree
8. fix policy utils unittest
9. fix typo in readme and add a 'back to top' button in readme (#104) (#109) (#111)

Style

1. add NeurIPS 2023 paper link

News

1. NeurIPS 2023 Spotlight: LightZero: A Unified Benchmark for Monte Carlo Tree Search in General Sequential Decision Scenarios

Full Changelog: v0.0.2...v0.0.3

Contributors: @PaParaZz1 @karroyan @nighood @jayyoung0802 @timothijoe @TuTuHuss @HarryXuancy @puyuan1996 @HansBug @mohitd404 @@PentesterPriyanshu @0Armaan025 @prajjwalyd @suravshresth @sohamtembhurne @eltociear

v0.0.2

Env

1. MuJoCo env (#50)
2. 2048 env (#64)
3. Connect4 env (#63)

Algorithm

1. Gumbel MuZero (#22)
2. Stochastic MuZero (#64)

Enhancement

1. polish mcts and ptree_az (#57) (#61)
2. polish readme (#36) (#47) (#51) (#77) (#95) (#96)
3. update paper notes (#89) (#91)
4. polish model and configs (#26) (#27) (#50)
5. add Dockerfile and its usage instructions (#95)
6. add doc about how to customize envs and algos (#78)
7. add pytorch ddp support (#68)
8. add eps greedy and random collect option in train_muzero_entry (#54)
9. add atari visualization option (#40)
10. add log_buffer_memory_usage utils (#30)

Fix

1. fix priority bug in muzero collector (#74)

Style

1. update github action (#71) (#72) (#73) (#81) (#83) (#84) (#90)

Full Changelog: v0.0.1...v0.0.2

Contributors: @PaParaZz1 @karroyan @nighood @jayyoung0802 @timothijoe @TuTuHuss @HarryXuancy @puyuan1996 @HansBug

v0.0.1

Full Changelog: https://github.com/opendilab/LightZero/commits/v0.0.1