[情報] DeepMind 和 Blizzard 發表 AI 環境
看板StarCraft (星海爭霸2 - SC2)作者LCamel (LCamel)時間8年前 (2017/08/10 09:47)推噓42(42推 0噓 20→)留言62則, 33人參與討論串1/1
DeepMind and Blizzard open StarCraft II as an AI research environment
Wednesday, 9 August 2017 (縮 https://goo.gl/KSfFFa )
DeepMind (世界最強圍棋 AI 那家) 和 Blizzard 共同發表新的 AI 開發環境,
"SC2LE" , 其中包括了:
* A Machine Learning API developed by Blizzard that gives researchers
and developers hooks into the game. This includes the release of tools
for Linux for the first time.
Blizzard 公開一套給機器學習用的接口
https://github.com/Blizzard/s2client-proto
The StarCraft II API is an interface that provides full external control
of StarCraft II.
可以讓外部程式控制 SC2.
This API exposes functionality for developing software for:
這套接口可以用來開發:
* Scripted bots. 寫死規則的傳統 bot
* Machine-learning based bots. 機器學習的 bot
* Replay analysis. 分析 replay 錄影 (才知道高手怎麼玩)
* Tool assisted human play. 人玩, 同時工具在旁協助 (這沒問題嗎?)
The API is available in the retail Windows and Mac clients.
There are also Linux clients available at the download links below.
零售版的 Windows / Mac 星海可用. Linux 也有.
* A dataset of anonymised game replays, which will increase from 65k
to more than half a million in the coming weeks.
將陸續公開超過 50 萬場 replay 來讓大家有得學.
* An open source version of DeepMind’s toolset, PySC2,
to allow researchers to easily use Blizzard’s feature-layer API
with their agents.
https://github.com/deepmind/pysc2
"... It exposes Blizzard Entertainment's StarCraft II
Machine Learning API as a Python RL Environment."
把上述接口包成 python 語言可以用的強化學習環境.
* A series of simple RL mini-games to allow researchers to test
the performance of agents on specific tasks.
一些讓研究者可測試的小任務.
採礦採礦 生兵生兵
https://www.youtube.com/watch?v=6L448yg0Sm0
* A joint paper that outlines the environment, and reports initial
baseline results on the mini-games, supervised learning from replays,
and the full 1v1 ladder game against the built-in AI.
https://deepmind.com/documents/110/sc2le.pdf
一篇兩家公司共同發表的目前成果, 作為大家開始的起點.
影片中右邊的訓練成果看起來還有模有樣的呢:
https://www.youtube.com/watch?v=WEOzide5XFc
LCamel
--
※ 發信站: 批踢踢實業坊(ptt.cc), 來自: 36.227.89.51
※ 文章網址: https://www.ptt.cc/bbs/StarCraft/M.1502329625.A.C72.html
※ 編輯: LCamel (36.227.89.51), 08/10/2017 09:47:55
推
08/10 09:48, , 1F
08/10 09:48, 1F
推
08/10 09:54, , 2F
08/10 09:54, 2F
→
08/10 09:54, , 3F
08/10 09:54, 3F
說不定喔... 像 AlphaGo 下圍棋會開局點三三取實利,
一般人不會那樣下, 可是後來研究以後, 柯潔也模仿了這種下法
又, 是不是很久以前有蓋建築改變採氣路徑來加速的作法?
推
08/10 09:56, , 4F
08/10 09:56, 4F
推
08/10 10:12, , 5F
08/10 10:12, 5F
→
08/10 10:12, , 6F
08/10 10:12, 6F
電腦大局觀沒有人強 不過最佳化這種事情倒是可以期待...
推
08/10 10:17, , 7F
08/10 10:17, 7F
推
08/10 10:26, , 8F
08/10 10:26, 8F
→
08/10 10:29, , 9F
08/10 10:29, 9F
推
08/10 10:58, , 10F
08/10 10:58, 10F
→
08/10 11:15, , 11F
08/10 11:15, 11F
原來如此 XD
推
08/10 11:17, , 12F
08/10 11:17, 12F
推
08/10 11:19, , 13F
08/10 11:19, 13F
推
08/10 11:22, , 14F
08/10 11:22, 14F
看來還有得學 ^^
推
08/10 11:25, , 15F
08/10 11:25, 15F
→
08/10 11:26, , 16F
08/10 11:26, 16F
我想電腦雖然沒有直覺 不過或許可以透過近乎完美的計算和偵查
找到雙方兵力差距的窗口捅穿對手...
推
08/10 11:37, , 17F
08/10 11:37, 17F
推
08/10 11:41, , 18F
08/10 11:41, 18F
推
08/10 11:47, , 19F
08/10 11:47, 19F
推
08/10 11:48, , 20F
08/10 11:48, 20F
→
08/10 11:49, , 21F
08/10 11:49, 21F
→
08/10 11:50, , 22F
08/10 11:50, 22F
→
08/10 11:50, , 23F
08/10 11:50, 23F
這個看來要等更多資訊公開了 ^^
推
08/10 11:59, , 24F
08/10 11:59, 24F
推
08/10 12:00, , 25F
08/10 12:00, 25F
謝謝資訊 ~
推
08/10 12:02, , 26F
08/10 12:02, 26F
歡迎 XD
推
08/10 12:03, , 27F
08/10 12:03, 27F
推
08/10 12:50, , 28F
08/10 12:50, 28F
推
08/10 12:55, , 29F
08/10 12:55, 29F
※ 編輯: LCamel (202.174.4.5), 08/10/2017 13:22:26
→
08/10 13:17, , 30F
08/10 13:17, 30F
推
08/10 13:34, , 31F
08/10 13:34, 31F
→
08/10 13:35, , 32F
08/10 13:35, 32F
基於這個接口應該做的出來
推
08/10 13:38, , 33F
08/10 13:38, 33F
基於這個接口應該做的出來
推
08/10 14:37, , 34F
08/10 14:37, 34F
→
08/10 14:38, , 35F
08/10 14:38, 35F
了解 ~
推
08/10 14:57, , 36F
08/10 14:57, 36F
→
08/10 14:58, , 37F
08/10 14:58, 37F
因為不是一條條的規則 可能我們也很難描述AI怎麼學和學到了什麼 :p
推
08/10 15:21, , 38F
08/10 15:21, 38F
推
08/10 15:29, , 39F
08/10 15:29, 39F
推
08/10 16:23, , 40F
08/10 16:23, 40F
我家AI本來很乖的 都是上網才學壞的
推
08/10 18:05, , 41F
08/10 18:05, 41F
推
08/10 19:08, , 42F
08/10 19:08, 42F
推
08/10 20:13, , 43F
08/10 20:13, 43F
→
08/10 20:13, , 44F
08/10 20:13, 44F
看在限制 apm 的條件下爬天梯可以爬到哪裡... ?
推
08/10 22:34, , 45F
08/10 22:34, 45F
→
08/10 22:34, , 46F
08/10 22:34, 46F
好的不學學壞的 TIMBA 啦
※ 編輯: LCamel (36.227.89.51), 08/10/2017 22:47:36
後來想想不管是電腦操控的 T 贏還是人類贏 都是人類贏. TIMBA 啦
推
08/10 22:45, , 47F
08/10 22:45, 47F
比方說發現其實光砲快攻必勝之類的 (翻桌)
推
08/10 23:24, , 48F
08/10 23:24, 48F
太壞 XD
推
08/10 23:34, , 49F
08/10 23:34, 49F
→
08/10 23:34, , 50F
08/10 23:34, 50F
阿法狗可以讓人類玩家有顏面的輸半目 XD
推
08/11 01:29, , 51F
08/11 01:29, 51F
結果 AI 一怒就毀滅人類了 XD
推
08/11 11:40, , 52F
08/11 11:40, 52F
結果 apm 完全等於有效apm (抖)
推
08/11 12:54, , 53F
08/11 12:54, 53F
→
08/11 15:31, , 54F
08/11 15:31, 54F
前一陣子鋼鐵人和臉書老闆才在吵這個...
推
08/11 16:48, , 55F
08/11 16:48, 55F
推
08/11 20:12, , 56F
08/11 20:12, 56F
推
08/11 21:00, , 57F
08/11 21:00, 57F
→
08/11 21:01, , 58F
08/11 21:01, 58F
→
08/11 21:02, , 59F
08/11 21:02, 59F
→
08/11 21:04, , 60F
08/11 21:04, 60F
→
08/11 21:05, , 61F
08/11 21:05, 61F
這樣 O_O
※ 編輯: LCamel (36.227.77.123), 08/12/2017 21:49:31
推
08/13 16:16, , 62F
08/13 16:16, 62F
StarCraft 近期熱門文章
PTT遊戲區 即時熱門文章
16
27