Skip to content

Latest commit

 

History

History
418 lines (323 loc) · 16.2 KB

README.md

File metadata and controls

418 lines (323 loc) · 16.2 KB
Raw

Honor of Kings AI Open Environment of Tencent(腾讯王者荣耀AI开放环境)

avatar

Update: 3v3 Mode Now Available

python3.8 -c "from hok.hok3v3.unit_test.test_env import run_test; run_test()"

Please refer to hok3v3 for further information.

Please consult cluster.md document for instructions on cluster training utilizing the hok_env environment and the integrated rl_framework.

Introduction

PyPI Image

  • Hok_env is the open environment of the MOBA game: Honor of kings.

  • This repository mainly includes Hok_env SDK, a reinforcement learning training framework and an implementation of ppo algorithm based on the training framework. Hok_env SDK is used to interact with the gamecore of Honor of Kings.

  • This repository also contains the implementation code for the paper:

    Honor of Kings Arena: an Environment for Generalization in Competitive Reinforcement Learning.
    Hua Wei*, Jingxiao Chen*, Xiyang Ji*, Hongyang Qin, Minwen Deng, Siqin Li, Liang Wang, Weinan Zhang, Yong Yu, Lin Liu, Lanxiao Huang, Deheng Ye, Qiang Fu, Wei Yang. (*Equal contribution)
    NeurIPS Datasets and Benchmarks 2022
    Project Page: https://github.com/tencent-ailab/hok_env
    arXiv: https://arxiv.org/abs/2209.08483

    Abstract: This paper introduces Honor of Kings Arena, a reinforcement learning (RL) environment based on Honor of Kings, one of the world’s most popular games at present. Compared to other environments studied in most previous work, ours presents new generalization challenges for competitive reinforcement learning. It is a multiagent problem with one agent competing against its opponent; and it requires the generalization ability as it has diverse targets to control and diverse opponents to compete with. We describe the observation, action, and reward specifications for the Honor of Kings domain and provide an open-source Python-based interface for communicating with the game engine. We provide twenty target heroes with a variety of tasks in Honor of Kings Arena and present initial baseline results for RL-based methods with feasible computing resources. Finally, we showcase the generalization challenges imposed by Honor of Kings Arena and possible remedies to the challenges. All of the software, including the environment-class, are publicly available at: https://github.com/tencent-ailab/hok_env. The documentation is available at: https://aiarena.tencent.com/hok/doc/.

  • Current supported heroes in hok_env:

    • lubanqihao
    • miyue
    • libai
    • makeboluo (Marco Polo)
    • direnjie
    • guanyu
    • diaochan
    • luna
    • hanxin
    • huamulan
    • buzhihuowu (Mai Shiranui)
    • jvyoujing (Ukyou Tachibana)
    • houyi
    • zhongkui
    • ganjiangmoye
    • kai
    • gongsunli
    • peiqinhu
    • shangguanwaner

Running Requirement

  • python >= 3.6, <= 3.9.

  • Windows 10/11 or Linux wine (to deploy windows gamecore server)

  • Docker (to deploy hok_env on Linux containers)

    • For windows, WSL 2 is required. (Windows Subsystem for Linux Version 2.0)

The gamecore of hok_env runs on the Windows platform, and the package hok_env needs to be deployed in linux platforms to interact with the gamecore.

We also provided a docker image for training on your computer. In a further version, we will release a gamecore server compatible with linux.

To enable cluster training, here is a workaround by running Windows gamecore on Linux: run windows gamecore on linux.

Gamecore Installation

Download the hok gamecore

You need to apply for the license and gamecore on this page: https://aiarena.tencent.com/aiarena/en/open-gamecore

Please put the license.dat under the folder:hok_env_gamecore/gamecore/core_assets.

avatar

Test the gamecore

open CMD

 cd gamecore\bin
 set PATH=%PATH%;..\lib\
 .\sgame_simulator_remote_zmq.exe .\sgame_simulator.common.conf

sgame_simulator_remote_zmq.exe requires one parameters: sgame_simulator.common.conf the config file path

You can see the following message:

PlayerNum:2
AbsPath:../scene/1V1.abs
PlayerInfo [CampID:0][HeroID:199][Skill:80104][AutoAi:1][AiServer::0:100] [Symbol 0 0 0] [Request:-1]
PlayerInfo [CampID:1][HeroID:199][Skill:80104][AutoAi:1][AiServer::0:100] [Symbol 0 0 0] [Request:-1]
SGame Simulator Begin
init_ret:0
seed: 3417
Symbols:
inHeroId: 199
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
inHeroId: 199
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
start_ret:0
Hero[0] inHeroId:199; outPlayerId:148
Hero[1] inHeroId:199; outPlayerId:149
[Hero Info] [HeroID:199] [RuntimeID:8] client_id:0.0.0.0_1658_0_20221202193534_148
[Hero Info] [HeroID:199] [RuntimeID:9] client_id:0.0.0.0_1658_0_20221202193534_149
boost_ret finished: 8, gameover_ai_server: 0
close_ret:0
uninit_ret:0
SGame Simulator End [FrameNum:8612][TimeUsed:7580ms]

The gamecore has started successfully!

Here is the content of sgame_simulator.common.conf:

{
    "abs_file": "../scene/1V1.abs",
    "core_assets": "../core_assets",
    "game_id": "kaiwu-base-35401-35400-6842-1669963820108111766-217",
    "hero_conf": [
        {
            "hero_id": 199
        },
        {
            "hero_id": 199
        }
    ]
}

output files:

AIOSS_221202-1935_linux_1450111_1450123_1_1_20001_kaiwu-base-35401-35400-6842-1669963820108111766-217.abs
kaiwu-base-35401-35400-6842-1669963820108111766-217.json
kaiwu-base-35401-35400-6842-1669963820108111766-217.stat
kaiwu-base-35401-35400-6842-1669963820108111766-217_detail.stat

1v1

Observation and action spaces

Please refer to https://aiarena.tencent.com/hok/doc/quickstart/index.html

Usage

Please refer to hok1v1/unit_test for the basic usage of the hok1v1.

Please refer to aiarena/1v1 for the training code of hok1v1.

Test the gamecore with the demo script in WSL

You can test gamecore with a simple python script in wsl.

Make sure your pc supports wsl2

For the installation and upgrade of wsl2, please refer to the link: https://docs.microsoft.com/zh-cn/windows/wsl/install-manual#step-4---download-the-linux-kernel-update-package`

You need to install python3.6 and some required dependencies in wsl.

Run the test script in wsl2

  1. Start the gamecore server outside wsl2

    cd gamecore
gamecore-server.exe server --server-address :23432

  2. Install hok_env in python

    ## after git clone this repo 
cd hok_env/hok_env
pip install -e .

  3. Run the test script

    cd /hok_env/hok/hok1v1/unit_test
python test_env.py

If you see the following message, you have successfully established a connection with Hok_env and have completed a game. Congratulations!

# python test_env.py
127.0.0.1:23432 127.0.0.1
======= test_send_action
camp_config {'mode': '1v1', 'heroes': [[{'hero_id': 132}], [{'hero_id': 133}]]}
common_ai [False, True]
try to get first state...
first state:  dict_keys(['observation', 'legal_action', 'reward', 'done', 'model_output_name', 'game_id', 'player_id', 'frame_no', 'sub_action_mask', 'req_pb', 'sgame_id'])
first frame: 0
----------------------run step  0
----------------------run step  100
----------------------run step  200
----------------------run step  300
----------------------run step  400
----------------------run step  500
----------------------run step  600
----------------------run step  700
----------------------run step  800
----------------------run step  900
----------------------run step  1000
----------------------run step  1100
2023-08-23 13:13:57.782 | INFO     | hok.common.log:info:85 - game not end, send close game at first: cur_frame_no(3525)
[{
    "player_id": 8,
    "frame_no": 30,
    "observation": array([1.00000000e00, 1.00000000e00, 0.00000000e00, 0.00000000e00, ...]),
    "reward": (0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0),
    "legal_action": array([1.0, 1.0, 1.0, 0.0, 0.0, ...]),
    "sub_action_mask": {
        0: array([1.0, 0.0, 0.0, 0.0, 0.0, 0.0]),
        1: array([1.0, 0.0, 0.0, 0.0, 0.0, 0.0]),
        ...
        11: array([1.0, 0.0, 0.0, 0.0, 0.0, 0.0]),
    },
    "req_pb": hok.hok1v1.lib.interface.AIFrameState,
}, ...]

Modify 1v1 Game Config

Before running the game Env, you need to create file config.json at the running path.

An example:

{
  "reward_money": "0.006",
  "reward_exp": "0.006" ,
  "reward_hp_point": "2.0",
  "reward_ep_rate": "0.75",
  "reward_kill": "-0.6",
  "reward_dead": "-1.0",
  "reward_tower_hp_point": "5.0",
  "reward_last_hit": "0.5",
  
  "log_level": "4"
}

This config file includes sub-reward factors and log_level of protobuf processing part. The file is only loaded when creating instances of HoK1v1, and any modifications would not be reloaded even if you call HoK1v1.reset.

In most cases, log_level should be set as 4 to avoid useless log information. Only if you meet some error when using our environment, log_level may need a lower value to help us get more information about the error your meet.

3v3

Observation and action spaces

Please refer to hok3v3 for further information.

Usage

Assuming you have started your gamecore server at 127.0.0.1:23432 and your IP running the hok_env is 127.0.0.1.

Here is the basic usage of the hok3v3 environment:

  • Get the environment instance:

    GC_SERVER_ADDR = os.getenv("GAMECORE_SERVER_ADDR", "127.0.0.1:23432")
AI_SERVER_ADDR = os.getenv("AI_SERVER_ADDR", "127.0.0.1")
reward_config = RewardConfig.default_reward_config.copy()

env = get_hok3v3(GC_SERVER_ADDR, AI_SERVER_ADDR, reward_config)

  • Reset env and start a new game

    use_common_ai = [True, False]
camp_config = {
    "mode": "3v3",
    "heroes": [
        [{"hero_id": 190}, {"hero_id": 173}, {"hero_id": 117}],
        [{"hero_id": 141}, {"hero_id": 111}, {"hero_id": 107}],
    ],
}
env.reset(use_common_ai, camp_config, eval_mode=True)

  • Game loop and predictions

    gameover = False
while not gameover:
    for i, is_comon_ai in enumerate(use_common_ai):
        if is_comon_ai:
            continue

        continue_process, features, frame_state = env.step_feature(i)
        gameover = frame_state.gameover
        # only predict every 3 frame
        if not continue_process:
            continue

        probs = random_predict(features, frame_state)
        ok, results = env.step_action(i, probs, features, frame_state)
        if not ok:
            raise Exception("step action failed")

env.close_game(force=True)

You can get the default reward config by:

reward_config = RewardConfig.default_reward_config.copy()

Here is the reward config example:

reward_config = {
    "whether_use_zero_sum_reward": 1,
    "team_spirit": 0,
    "time_scaling_discount": 1,
    "time_scaling_time": 4500,
    "reward_policy": {
        "hero_0": {
            "hp_rate_sqrt_sqrt": 1,
            "money": 0.001,
            "exp": 0.001,
            "tower": 1,
            "killCnt": 1,
            "deadCnt": -1,
            "assistCnt": 1,
            "total_hurt_to_hero": 0.1,
            "atk_monster": 0.1,
            "win_crystal": 1,
            "atk_crystal": 1,
        },
    },
    "policy_heroes": {
        "hero_0": [169, 112, 174],
    },
}

You can get the hero_id by hero_name via HERO_DICT:

from hok.common.camp import HERO_DICT
print(HERO_DICT)

Please refer to hok3v3/test_env for the full code introduced above.

And you can run the test code by following script:

python3.8 -c "from hok.hok3v3.unit_test.test_env import run_test; run_test()"

You will get the following message if works correctly:

2023-12-28 12:54:28.106 | INFO     | hok.hok3v3.unit_test.test_env:get_hok3v3:14 - Init libprocessor: /usr/local/lib/python3.8/dist-packages/hok/hok3v3/config.dat
2023-12-28 12:54:28.106 | INFO     | hok.hok3v3.unit_test.test_env:get_hok3v3:15 - Init reward: {'whether_use_zero_sum_reward': 1, 'team_spirit': 0.2, 'time_scaling_discount': 1, 'time_scaling_time': 4500, 'reward_policy': {'policy_name_0': {'hp_rate_sqrt': 1, 'money': 0.001, 'exp': 0.001, 'tower': 1, 'killCnt': 1, 'deadCnt': -1, 'assistCnt': 1, 'total_hurt_to_hero': 0.1, 'ep_rate': 0.1, 'win_crystal': 1}}, 'hero_policy': {1: 'policy_name_0'}, 'policy_heroes': {'policy_name_0': [1, 2]}}
2023-12-28 12:54:28.107 | INFO     | hok.hok3v3.unit_test.test_env:get_hok3v3:16 - Init gamecore environment: 127.0.0.1:23432 127.0.0.1
2023-12-28 12:54:28.107 | INFO     | hok.hok3v3.reward:update_reward_config:116 - Update reward config: time_scaling_time:4500, time_scaling_discount:1, team_spirit:0.2, whether_use_zero_sum_reward:1
2023-12-28 12:54:28.107 | INFO     | hok.hok3v3.reward:update_reward_config:124 - Update hero reward config: 1 -> {'hp_rate_sqrt': 1, 'money': 0.001, 'exp': 0.001, 'tower': 1, 'killCnt': 1, 'deadCnt': -1, 'assistCnt': 1, 'total_hurt_to_hero': 0.1, 'ep_rate': 0.1, 'win_crystal': 1}
2023-12-28 12:54:28.107 | INFO     | hok.hok3v3.reward:update_reward_config:124 - Update hero reward config: 2 -> {'hp_rate_sqrt': 1, 'money': 0.001, 'exp': 0.001, 'tower': 1, 'killCnt': 1, 'deadCnt': -1, 'assistCnt': 1, 'total_hurt_to_hero': 0.1, 'ep_rate': 0.1, 'win_crystal': 1}
2023-12-28 12:54:28.136 | INFO     | hok.hok3v3.server:start:31 - Start server at tcp://0.0.0.0:35151
2023-12-28 12:54:28.139 | INFO     | hok.hok3v3.env:reset:85 - Reset info: agent:0 is_common_ai:True
2023-12-28 12:54:28.139 | INFO     | hok.hok3v3.env:reset:85 - Reset info: agent:1 is_common_ai:False
2023-12-28 12:54:30.212 | INFO     | hok.hok3v3.unit_test.test_env:run_test:78 - ----------------------run step 0
2023-12-28 12:54:30.673 | INFO     | hok.hok3v3.unit_test.test_env:run_test:78 - ----------------------run step 100
2023-12-28 12:54:30.945 | INFO     | hok.hok3v3.unit_test.test_env:run_test:78 - ----------------------run step 200

Cluster training

Please consult cluster.md document for instructions on cluster training utilizing the hok_env environment and the integrated rl_framework.

Replay software: ABS Parsing Tool (will be provided along with the gamecore)

Watching the game is a direct way to see the performance of your agent throughout a match. We provide a replay tool to visualize the matches.

This is an official replay software which parses the ABS file generated by the gamecore and outputs the videos in the game UI of Honor of Kings. The ABS file generated by the gamecore could be found under the gamecore folder. You can visualize the matches by putting ABS files under the Replays folder and running ABSTOOL.exe. avatar

Pre-built image

https://hub.docker.com/r/tencentailab/hok_env

See also: run_with_prebuilt_image

Install from PyPI

pip install hok

Citation

If you use the gamecore of hok_env or the code in this repository, please cite our paper as follows.

@inproceedings{wei2022hok_env,
  title={Honor of Kings Arena: an Environment for Generalization in Competitive Reinforcement Learning},
  author={Wei, Hua and Chen, Jingxiao and Ji, Xiyang and Qin, Hongyang and Deng, Minwen and Li, Siqin and Wang, Liang and Zhang, Weinan and Yu, Yong and Liu, Lin and Huang, Lanxiao and Ye, Deheng and Fu, Qiang and Yang, Wei},
  booktitle={Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks},
  year={2022}
}