opendilab / di-star Goto Github PK

看大小都是150M左右的小模型，打了几场感觉应变很差，套路也很固定也基本没侦察。放点战术就死了，比如两矿堵口速飞龙，看到飞龙才知道补防空，而且打哪个矿补哪个，2矿*了一波去主矿竟然都没补。看到堵口以后完全没有应对，没下三矿也没侦察，也没预判飞龙。不知道原始的大模型是否也是这样？（ps：我zvz很菜，正常打是打不过，只能放战术）

Getting this error when trying to run: module 'numpy' has no attribute 'int'. Did you mean: 'inf'?.

Looks like the latest version of numpy, np.int was removed, and it's just int now, or potentially np.int32 and np.int64. I don't know enough about the codebase to do a PR with the updates, but a simple fix for now is to just pin the numpy version with numpy==1.23.5 in setup.py.

SL training works fine if I turn on cuda = True although I am running a little short of budget so I wanted to try if running on CPU works.

If I turn off cuda, here is the exception I get. Any idea how to fix this? I tried to change that part of the code to with same issue.

if torch.cuda.is_available():
embedding = torch.zeros(bs * shape_y * shape_x, device=x[k].device)
bias = torch.arange(bs, device=x[k].device).unsqueeze(dim=1) * shape_y * shape_x
else:
embedding = torch.zeros(bs * shape_y * shape_x)
bias = torch.arange(bs).unsqueeze(dim=1) * shape_y * shape_x

[2022-12-04 14:37:58,901][checkpoint_helper.py][line: 140][ INFO] save checkpoint in /home/mark/projects/DI-star/experiments/sl_train_protoss/checkpoint/sl_train_protoss_iteration_1.pth.tar
[2022-12-04 14:37:58,902][ log_helper.py][line: 161][ INFO] BaseLearner139853483483344 save checkpoint in /home/mark/projects/DI-star/experiments/sl_train_protoss/checkpoint/sl_train_protoss_iteration_1.pth.tar
Traceback (most recent call last):
File "/home/mark/projects/DI-star/distar/ctools/torch_utils/checkpoint_helper.py", line 364, in wrapper
return func(*args, **kwargs)
File "/home/mark/projects/DI-star/distar/ctools/worker/learner/base_learner.py", line 266, in run
self._train(data)
File "/home/mark/projects/DI-star/distar/ctools/worker/learner/base_learner.py", line 129, in wrapper
ret = fn(*args, **kwargs)
File "/home/mark/projects/DI-star/distar/agent/default/sl_learner.py", line 50, in _train
logits, infer_action_info, hidden_state = self._model.sl_train(**data, hidden_state=self.hidden_state)
File "/home/mark/projects/DI-star/distar/agent/default/model/model.py", line 182, in sl_train
self.encoder(spatial_info, entity_info, scalar_info, entity_num)
File "/home/mark/anaconda3/envs/distar/lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/mark/projects/DI-star/distar/agent/default/model/encoder.py", line 43, in forward
embedded_spatial, map_skip = self.spatial_encoder(spatial_info, scatter_map)
File "/home/mark/anaconda3/envs/distar/lib/python3.8/site-packages/torch/nn/modules/module.py", line 727, in _call_impl
result = self.forward(*input, **kwargs)
File "/home/mark/projects/DI-star/distar/agent/default/model/obs_encoder/spatial_encoder.py", line 74, in forward
embedding[x[k].long()] = 1.
IndexError: index 9728000 is out of bounds for dimension 0 with size 9728000

The link in the last update is invalid now.

I'm a new fan(SC2 newbie). Wanna learn more. -v-

想问一下有没有最新版本的5.0.9的replay数据，或者如何去获得这些replay数据呢？

The Discord invite is invalid.

I'm glad to see that this repo is continuously maintained this year, and still hope to make contribution.
But I find it's quite difficult to get started, and good docs/tutorials are in strong demand.

If you can make a short introduction for the code architecture, I can help construct a doc template (like this: trojanzoo), so that we can together work on it.

Of course, this must be based on the assumption that the api architecture is already fixed.

I've read the code and found terran and protoss in the encoding. Is there a way to build a agent for terran or protoss?

I can hear the sound of the agent's window while playing against it. Such as the sound of buildings and units.

Hello IA大哥，我在本地运行Windows版的DI-star并尝试与AI对抗时发现一个问题
基于这个环境下启动的星际争霸2没有办法正确读取我账号对应的自定义快捷键文件
请问这是我设置的问题吗？还是说是一个还没有实现的功能

期待回复

1.我的笔记本是m1 max 是否可以使用m1 max的gpu 进行训练

may i know your suggestion for gpu on laptop computer?

您好，请问如果想用4.10.0版本的replays进行监督学习，要去哪里找更多对应版本（4.10.0）的replays呢？我已经试过在s2client-proto通过终端命令下载了4.10.0版本的replays，但却无法用4.10.0版本的客户端进行播放（提示游戏执行所参照的模块或地图关联内容已不可用，如下图），貌似外服录像无法用国服客户端打开？

I tryed blizard offical SC2 and pysc2, neither can open replay file generated by 'python -m distar.bin.play'.

If I open with pysc2, it turns out that has an error
'pysc2.lib.remote_controller.RequestError: SC2APIProtocol.ResponseReplayInfo.Error.ParsingError: 'Could not open initData for the replay: C:\Users***\AppData\Local\Temp\StarCraft II\TempReplayInfo.SC2Replay'.
google-deepmind/pysc2#25, probably suggests that the error comes from a broken replay file.

If need me to provide the broken replay file, pls let me know.

打开程序后直接就进入游戏，无法联网与真人对决吗？

I don't know

Hi!
After checking this repo, it seems quite some codes are from pysc2. Those codes might better serve as an installed package, rather than cloned in this repo.

And I hope that there could be an abstract to illustrate the contribution of this repo, such as the code structures, algorithm and mechanism figures. Docs and test CI workflows are also important. Currently it's a little difficult to catch up without a clear guidance, but I'm quite willing to contribute in the future.

Btw, is IA himself the maintainer of this repo? Quite amazing.

I'm a terran player and it will be perfect if human can choose races.

Hi, I am trying to train a playable the AI to learn ZvT, but am encountering some errors.

1. I made a trainingreplays folder and put some replays from Patch 4.9. Command I ran was python -m distar.bin.sl_train --data ./trainingreplays:

Start decoding replay with player 1, path: ./trainingreplays\reynorvssoul1.SC2Replay
12380 [ERROR] parse replay error SC2APIProtocol.ResponseReplayInfo.Error.InvalidReplayPath: ''
  File "C:\Users\Steve Lu\Downloads\di-star\di-star\distar\agent\default\replay_decoder.py", line 380, in run
    self._replay_info = self._parse_replay_info()
  File "C:\Users\Steve Lu\Downloads\di-star\di-star\distar\agent\default\replay_decoder.py", line 343, in _parse_replay_info
    replay_info = self._controller.replay_info(replay_path=self._replay_path)
  File "C:\Users\Steve Lu\Downloads\di-star\di-star\distar\pysc2\lib\remote_controller.py", line 69, in _check_error
    return check_error(func(*args, **kwargs), error_enum)
  File "C:\Users\Steve Lu\Downloads\di-star\di-star\distar\pysc2\lib\remote_controller.py", line 60, in check_error
    raise RequestError("%s.%s: '%s'" % (enum_name, error_name, details))

2. I tried the reinforcement learning selfplay command python -m disatr.bin.rl_train --task selfplay. Am I understanding correctly that selfplay is when it trains by playing against me?

train !!!!!!!!!!!!!!!!!!!
Traceback (most recent call last):
  File "C:\Users\Steve Lu\Downloads\miniconda\lib\multiprocessing\process.py", line 315, in _bootstrap
    self.run()
  File "C:\Users\Steve Lu\Downloads\miniconda\lib\multiprocessing\process.py", line 108, in run
    self._target(*self._args, **self._kwargs)
TypeError: learner_run() takes 2 positional arguments but 5 were given
{'checkpoint_paths': ['C:\\Users\\Steve Lu\\Downloads\\di-star\\di-star\\experiments\\test\\league_models\\MP0H1_MP0_ckpt.pth.tar', 'C:\\Users\\Steve Lu\\Downloads\\di-star\\di-star\\experiments\\test\\league_models\\MP0_ckpt.pth.tar'], 'env_info': {'map_name': 'random', 'player_ids': ['MP0H1', 'MP0'], 'side_id': [0, 1]}, 'frac_ids': [1, 1], 'pipelines': ['default', 'default'], 'player_ids': ['MP0H1', 'MP0'], 'send_data_players': ['MP0'], 'side_ids': [0, 1], 'successive_ids': ['none', 'MP0'], 'teacher_checkpoint_paths': ['none', 'C:\\Users\\Steve Lu\\Downloads\\di-star\\di-star\\distar\\bin\\sl_model.pth'], 'teacher_player_ids': ['none', 'model'], 'update_players': ['MP0'], 'z_path': ['3map.json', '3map.json'], 'z_prob': [0.25, 0.25]}
run worker for token: MP0model at 127.0.0.1: 18992
 * Serving Flask app 'distar.ctools.worker.coordinator.coordinator' (lazy loading)
 * Environment: production
   WARNING: This is a development server. Do not use it in a production deployment.
   Use a production WSGI server instead.
 * Debug mode: off
z_type 1 cum False bo False
Map: NewRepugnancy Race: zerg, Born location: (16, 101), loop: 4840, idx: None
Building order:
  Build_Extractor_unit, (0, 0)
  Build_SpawningPool_pt, (21, 106)
  Train_Zergling_quick, (0, 0)
  Train_Zergling_quick, (0, 0)
  Train_Zergling_quick, (0, 0)
  Research_ZerglingMetabolicBoost_quick, (0, 0)
  Train_Zergling_quick, (0, 0)
  Train_Zergling_quick, (0, 0)
  Build_BanelingNest_pt, (16, 111)
  Train_Zergling_quick, (0, 0)
  Train_Zergling_quick, (0, 0)
  Build_Hatchery_pt, (16, 78)
  Train_Zergling_quick, (0, 0)
  Train_Baneling_quick, (0, 0)
  Train_Baneling_quick, (0, 0)
  Train_Queen_quick, (0, 0)
Cumulative stat:
  Build_BanelingNest_pt
  Build_Extractor_unit
  Build_Hatchery_pt
  Build_SpawningPool_pt
  Research_ZerglingMetabolicBoost_quick
  Train_Baneling_quick
  Train_Queen_quick
  Train_Zergling_quick

[EPISODE LOOP ERROR] 'terran'
  File "C:\Users\Steve Lu\Downloads\di-star\di-star\distar\actor\actor.py", line 140, in _inference_loop
    self.agents[idx].reset(map_name, race, game_info[idx], observations[idx])
  File "C:\Users\Steve Lu\Downloads\di-star\di-star\distar\agent\default\agent.py", line 191, in reset
    z = random.choice(z_data[self._map_name][self._race][born_location_str])

z_type 2 cum False bo False
Map: NewRepugnancy Race: zerg, Born location: (135, 18), loop: 9473, idx: None
Building order:
  Build_Hatchery_pt, (135, 43)
  Build_Extractor_unit, (0, 0)
  Build_SpawningPool_pt, (134, 30)
  Train_Queen_quick, (0, 0)
  Train_Queen_quick, (0, 0)
  Research_ZerglingMetabolicBoost_quick, (0, 0)
  Build_EvolutionChamber_pt, (126, 43)
  Train_Queen_quick, (0, 0)
  Research_ZergMeleeWeapons_quick, (0, 0)
  Train_Zergling_quick, (0, 0)
  Train_Zergling_quick, (0, 0)
  Train_Zergling_quick, (0, 0)
  Train_Zergling_quick, (0, 0)
  Train_Zergling_quick, (0, 0)
  Train_Zergling_quick, (0, 0)
  Train_Zergling_quick, (0, 0)
  Train_Zergling_quick, (0, 0)
  Build_RoachWarren_pt, (125, 40)
  Build_Hatchery_pt, (104, 27)
  Build_Extractor_unit, (0, 0)
Cumulative stat:
  Build_EvolutionChamber_pt
  Build_Extractor_unit
  Build_Hatchery_pt
  Build_RoachWarren_pt
  Build_SpawningPool_pt
  Morph_Ravager_quick
  Research_ZergGroundArmor_quick
  Research_ZerglingMetabolicBoost_quick
  Research_ZergMeleeWeapons_quick
  Train_Queen_quick
  Train_Roach_quick
  Train_Zergling_quick

[EPISODE LOOP ERROR] 'terran'
  File "C:\Users\Steve Lu\Downloads\di-star\di-star\distar\actor\actor.py", line 140, in _inference_loop
    self.agents[idx].reset(map_name, race, game_info[idx], observations[idx])
  File "C:\Users\Steve Lu\Downloads\di-star\di-star\distar\agent\default\agent.py", line 191, in reset
    z = random.choice(z_data[self._map_name][self._race][born_location_str])

Process SpawnProcess-4:
Traceback (most recent call last):
  File "C:\Users\Steve Lu\Downloads\miniconda\lib\runpy.py", line 194, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "C:\Users\Steve Lu\Downloads\miniconda\lib\runpy.py", line 87, in _run_code
    exec(code, run_globals)
  File "C:\Users\Steve Lu\Downloads\di-star\di-star\distar\bin\rl_train.py", line 153, in <module>
    actor_run(config, args)
  File "C:\Users\Steve Lu\Downloads\di-star\di-star\distar\bin\rl_train.py", line 65, in actor_run
Traceback (most recent call last):
  File "C:\Users\Steve Lu\Downloads\di-star\di-star\distar\actor\actor.py", line 140, in _inference_loop
    self.agents[idx].reset(map_name, race, game_info[idx], observations[idx])
  File "C:\Users\Steve Lu\Downloads\di-star\di-star\distar\agent\default\agent.py", line 191, in reset
    z = random.choice(z_data[self._map_name][self._race][born_location_str])
    KeyError: 'terran'
actor.run()

I see there are many from __future__ import XXX which supports python2.

Since pytorch has terminated the support for py2, we might remove those compatible sentences.

Hey! First of all thanks for the great project!
I just tried to run to play against the bot but I'm getting an error. Here's what I did:

• Followed the github tutorial and was able to:
• Install python 3.10.4
• Installed CUDA 11
• Installed pytorch 1.7.1
• I think I successfully downloaded the models
• Downloaded old sc2 version with the replay

Here's what I'm getting when I run:

python -m distar.bin.play

Traceback (most recent call last):
  File "C:\Users\yurim\AppData\Local\Programs\Python\Python310\lib\runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "C:\Users\yurim\AppData\Local\Programs\Python\Python310\lib\runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "c:\users\yurim\projects\di-star\distar\bin\play.py", line 86, in <module>
    actor = Actor(user_config)
  File "c:\users\yurim\projects\di-star\distar\actor\actor.py", line 38, in __init__
    self._setup_agents()
  File "c:\users\yurim\projects\di-star\distar\actor\actor.py", line 65, in _setup_agents
    state_dict = torch.load(self._cfg.model_paths[player_id], map_location='cpu')
  File "C:\Users\yurim\AppData\Local\Programs\Python\Python310\lib\site-packages\torch\serialization.py", line 705, in load
    with _open_zipfile_reader(opened_file) as opened_zipfile:
  File "C:\Users\yurim\AppData\Local\Programs\Python\Python310\lib\site-packages\torch\serialization.py", line 243, in __init__
    super(_open_zipfile_reader, self).__init__(torch._C.PyTorchFileReader(name_or_buffer))
RuntimeError: PytorchStreamReader failed reading zip archive: failed finding central directory

Sorry I'm not familiar with python at all, anything I'm missing?
Thanks!

Hey! Awesome work so far!
It would be really cool if (a fork of) the DI-star environment also had support for training Starcraft 1 Broodwar AIs using BWAPI ( https://github.com/bwapi/bwapi ).
The Broodwar AI community has been very competitive for over a decade now and the current top bots are already at or above the playing strength of average human players.
Some more references if you are interested:
https://sscaitournament.com/
https://www.basil-ladder.net/

Keep up the good work! :)

李培楠nb！@upia99
咱这啥时候能开发好啊🤔CN星际都要画上完美句号啦
对目前的开发进度不是很清楚。我作为external contributor能做些什么吗？

I got an error when initializing the ai:

ImportError: cannot import name 'container_abcs' from 'torch._six' (C:\ProgramData\Miniconda3\lib\site-packages\torch_six.py)

Pytorch Version (Cuda v11)

torch 1.9.0
torchaudio 0.9.0
torchvision 0.10.0

Did some search and think this is relevant to torch version. What is the version that dev is using? Thanks!

Hi, when running python -m distar.bin.download_model --name rl_model i am getting Operation timed out error. Can you please check if your servers are up?

Following the windows software test installation, pressing the activate 4.10.0 button the replay does not play or download version 4.10.0