Category: Game

KataGo 推出了人類棋譜訓練出來的 model

KataGo v1.15.0 的「New Human-like Play and Analysis」推出了用人類棋譜訓練出來的 model：

This release adds a new human supervised learning ("Human SL") model trained on a large number of human games to predict human moves across players of different ranks and time periods! Not much experimentation with it has been done yet and there is probably low-hanging fruit on ways to use and visualize it, open for interested devs and enthusiasts to try.

lightvector 列出的一些 screenshot 看起來像是試著去猜測人類的棋力可能會選擇的點，列出了同樣的棋局在 20 級、1 段與 9 段會考慮的點的差異，算是一種幫助人類理解的方式？

LinkedIn 決定要在平台上面弄出遊戲...

也是清連結的消息，三月中的消息，LinkedIn 想要在平台上面弄出遊戲：「LinkedIn plans to add gaming to its platform」。

會是 puzzle 類型的遊戲，而且看起來有東西了：

It will be doing so by tapping into the same wave of puzzle-mania that helped simple games like Wordle find viral success and millions of players. Three early efforts are games called “Queens”, “Inference” and “Crossclimb.”

然後當時跟 LinkedIn 發言人確認也證實了這點：

“We’re playing with adding puzzle-based games within the LinkedIn experience to unlock a bit of fun, deepen relationships, and hopefully spark the opportunity for conversations,” the spokesperson said in a message to TechCrunch. “Stay tuned for more!”

試著走向更一般性的 social network？目前 LinkedIn 上面應該都是雞湯文與 clickbait，這個變化好像有點大...

Simon Tatham 的遊戲庫

在 Hacker News 上看到「Simon Tatham's Portable Puzzle Collection (greenend.org.uk)」這個討論，會讓我注意到的點是看到 greenend.org.uk 這個網址，印象中這就是 PuTTY 官網用的 domain，而點進去看以後發現就是 PuTTY 作者 Simon Tatham 另外的小玩具：「Simon Tatham's Portable Puzzle Collection」。

裡面都是單人遊戲，有提供 JavaScript 版可以在瀏覽器上直接玩，另外也有多個不同平台的執行檔可以用。是個墮落殺時間的區域...

Ruffle 對 crossdomain.xml 的處理方式

ruffle 是一個 open source 的 Flash 模擬器，用 JavaScript (以及 WebAssembly) 讓使用者可以在瀏覽器裡面直接執行。

不過想到 Flash 的 crossdomain.xml 在技術上應該是沒辦法繞過瀏覽器對 CORS 的限制，所以找了一下目前的情況... 在「Ideas for getting arounds CORS restrictions #8972」這邊有提到兩個方法，另外 comment 提到了「Add setting to spoof URLs in network requests #1486」這個。

看起來這個問題還在「有點子，但沒解決」的情況。

不過找了一下當年有玩的小遊戲 Zookeeper，在「Zookeeper - Flash Games Archive | FlashArch」這邊可以玩，看起來功能都還算正常，讓人懷念啊...

世紀帝國系列裡的 AoE 與 AoK 使用組合語言加速的情況

Age of Empires (1997) 與 Age of Empires II: The Age of Kings (1999) 是當年兩個很紅的遊戲，有傳言裡面有很大一部分是用組合語言寫的。

剛剛在 Hacker News 上看到「Original Age of Empires 2 dev talks about its usage of assembly code (reddit.com)」這篇，看起來是原作者跳出來說明這段傳言 (講的更細)，原文在 Reddit 的「AoE is written in Assembly - is this actually TRUE?! :O : aoe2」這邊。

32-bit 的 x86 組語部分有大約 13k 行，大多數 (11.5k 行) 是與繪圖有關的 (當年繪圖相關的功能都還是 CPU 在算)：

There were about ~13,000 lines of x86 32-bit assembly code written in total.

The vast majority, about ~11,500 lines worth, was in the 'drawing core', which drew SLP sprites in a variety of ways (mirrored, stippled, clipped). The code for this was in a separate .asm file, which was "compiled" by Microsoft Macro Assembler 6.1 into a .obj file.

最主要的目的當然是效能，這邊提到隔壁棚的星海爭霸預設還是 640x480 的時候，AoE 這邊預設可以上 800x600：

The use of assembly in the drawing core resulting in a ~10x sprite drawing speed improvement over the C++ reference implementations, and AoE's drawing core was notably faster than competitors like StarCraft, which is why the default resolution 'out of the box' for AoE was 800 by 600, when nearly all our competition was 640 x 480 resolution - we could scroll the screen and fill it with sprites as fast or faster even though we had twice as many pixels-ish in the game world area.

後來到 64-bit 年代就變成用 C++，主要是因為 compiler optimization 的改進，以及多 CPU 與 multi-threading 的技術，讓計算力提升很多：

I re-wrote the assembly functions into C++ for both Definitive Editions, as they are 64-bit programs, and inline assembly was never supported by the 64-bit C++ compiler, and the vastly improved register sets and compiler optimizations made it un-necessary. Additionally, sprite drawing in the definitive editions is multi-threaded, and will use up to 4 cores for that task alone.

算是時代的演進，意外的留下一些記錄... 而更後面的時代則是把這些計算 offload 到顯卡上面了。

Twitch 宣佈退出韓國市場

Twitch 宣佈 2024/02/27 (星期二) 退出韓國市場：「An Update on Twitch in Korea」。日期不知道是怎麼選的，可能跟某些合約有關？

Twitch 目前的公告會有繁體中文，也可以看這份：「Twitch 韓國現況更新」。

另外今天早上找了一下，Hacker News 也有討論了：「An update on Twitch in Korea (twitch.tv)」。

目前官方給出來的理由是虧本，而且找不到方法克服虧本的問題：

Ultimately, the cost to operate Twitch in Korea is prohibitively expensive and we have spent significant effort working to reduce these costs so that we could find a way for the Twitch business to remain in Korea.

這邊提到的包括了 p2p model 以及降到 720p，但即使如此網路費用 (應該就是頻寬費用) 是其他區域的十倍以上：

First, we experimented with a peer-to-peer model for source quality. Then, we adjusted source quality to a maximum of 720p. While we have lowered costs from these efforts, our network fees in Korea are still 10 times more expensive than in most other countries. Twitch has been operating in Korea at a significant loss, and unfortunately there is no pathway forward for our business to run more sustainably in that country.

在 Cloudflare 這邊，2016 年還叫做 CloudFlare 的時候也有抱怨過：「CloudFlare 對 HiNet 成本的抱怨 (還有其他 ISP...)」。

當年是這樣寫 HiNet 與 KT，成本大約是歐美區的 15 倍：

Two Asian locations stand out as being especially expensive: Seoul and Taipei. In these markets, with powerful incumbents (Korea Telecom and HiNet), transit costs 15x as much as in Europe or North America, or 150 units.

而尤其是韓國的部分，政府介入讓降價的速度比全世界慢，所以時間拉長後成本相較於其他地區就貴很多：

South Korea is perhaps the only country in the world where bandwidth costs are going up. This may be driven by new regulations from the Ministry of Science, ICT and Future Planning, which mandate the commercial terms of domestic interconnection, based on predetermined “Tiers” of participating networks. This is contrary to the model in most parts of the world, where networks self-regulate, and often peer without settlement. The government even prescribes the rate at which prices should decrease per year (-7.5%), which is significantly slower than the annual drop in unit bandwidth costs elsewhere in the world. We are only able to peer 2% of our traffic in South Korea.

不過不確定現在的情況，2016 年的 CloudFlare 跟 2023 年的 Cloudflare 已經差了七年了...

KataGo 1.12.0 與 UEC 杯用的 model：b18c384nbt-uec.bin.gz

剛剛看到 KataGo 出了 1.12.0，同時也放出了在 2022 年十一月 UEC 比賽時用的 model：「New Neural Net Architecture!」。

1.12.0 比較特別的新的類神經網路架構：

This version of KataGo adds support for a new and improved neural net architecture!

這個新的架構以及其他的改善讓訓練的速度改善：

The new neural nets use a new nested residual bottleneck structure, along with other major improvements in training. They train faster than KataGo's old nets and learn more effectively.

另外一個是他把 UEC 比賽時用的 model 放出來了，很特別的是採用 b18c384，而 KataGo Distributed Training 這邊目前主要是 b40c256 與 b60c320，看起來是為了比賽而一次性訓練出來的。

依照他的說法這個 b18c384 版本跟目前訓練網站上的 b60c320 有差不多強度，但計算速度會比 b60c320 快不少，甚至在一些機器上會跟 b40c256 差不多快：

Attached to this release is a one-off net b18c384nbt-uec.bin.gz that was trained for a tournament in 2022, which should be of similar strength to the 60-block nets on http://katagotraining.org/, but on many machines will run much faster, on some machines between 40-block and 60-block speed, but on some machines even as fast as or faster than 40-block.

另外一個大改變是他把訓練工具從 TensowFlow 跳槽到 PyTorch：

The training code has been all rewritten to use pytorch instead of tensorflow.

在 release note 裡沒有提到原因，但這個頗讓人好奇的...

對西洋棋選手 Hans Niemann 的調查

前幾天在「用情趣用品在西洋棋比賽裡面傳遞摩斯電碼作弊」這篇裡面有提到目前對 Hans Niemann 的調查都還在進行，現在 Chess.com 先丟出調查報告了：「Hans Niemann Report」，另外 Hacker News 上也有討論：「The Full Chess Cheating Report of Hans Neiman (chess.com)」。

PDF 的 summary 開頭就提到分析了線上的對奕記錄作弊的比率比起 Hans Niemann 所承認的高很多：

We present evidence in this report that Hans likely cheated online much more than his public statements suggest. However, while Hans has had a record-setting and remarkable rise in rating and strength, in our view there is a lack of concrete statistical evidence that he cheated in his game with Magnus or in any other over-the-board (“OTB”)—i.e., in-person—games. We are presenting our findings here and will cooperate with FIDE on any further investigation.

除了棋局上的記錄，在後面「How Cheat Detection Works」這個章節也提到了很多額外的 event log，像是視窗切換之類的資訊。

然後 Chess.com 也宣佈撤回賽事邀請，並且停權：

We uninvited Hans from our upcoming major online event and revoked his access to our site based on our experience with him in the past, growing suspicions among top players and our team about his rapid rise of play, the strange circumstances and explanations of his win over Magnus, as well as Magnus’ unprecedented withdrawal. In order to have more time to investigate the OTB situation and our own internal concerns, we uninvited Hans from our event and prevented his access to Chess.com. We are open to continuing a dialogue with Hans to discuss his status on Chess.com.

先有一個組織跳出來下刀了...

用情趣用品在西洋棋比賽裡面傳遞摩斯電碼作弊

標題資訊量有點大... 先講一下最近西洋棋界的新聞，九月的時候 Magnus Carlsen 指控 Hans Niemann 作弊，這件事情到目前還沒有完全落幕，可以翻維基百科的條目看看目前的進展：「Carlsen–Niemann controversy」。

而「ButtFish – Transmit Morse Code of chess moves to your butt (github.com/ronsijm)」這篇，則是示範怎麼用肛塞 (我這邊就拿 PChome 24h 的搜尋頁，有「興趣」的可以自己翻)，加上很多情趣用品是可以透過 API 操作的 (作者提到了 Buttplug 這個專案)，於是就造就了這個新的專案：「ButtFish」，專案名稱看起來是「引用」自西洋棋領域裡很有名的 Stockfish 引擎。

另外這張 Three circles model 圖也很好笑：

不過這個方法應該過不了金屬探測器，現在的競技比賽應該都會有才對？所以這專案只是很靠背而已...

專案頁面上最後面的自嘲也很好笑：

用 Poetry 的相依性演算法解數獨 (Sudoku)

在 Daily Lobsters 上看到「Solving Sudoku with Poetry's dependency resolver」這篇完全是惡搞 Python 下 Poetry 套件 XDDD

作者搞出來的方法是這樣，指定 81 個版號來表示題目，然後跑 Poetry 找可以的版本組合：

[tool.poetry.dependencies]
python = "^3.6"
sudoku-cell11 = "*"
sudoku-cell12 = "2.0.0"
sudoku-cell13 = "*"
sudoku-cell14 = "8.0.0"
sudoku-cell15 = "*"
sudoku-cell16 = "9.0.0"
sudoku-cell17 = "*"
sudoku-cell18 = "*"
sudoku-cell19 = "*"
sudoku-cell21 = "3.0.0"
sudoku-cell22 = "7.0.0"
sudoku-cell23 = "*"
sudoku-cell24 = "6.0.0"
...

另外作者有提到，本來是打算用 Yarn 來解，但看起來各種嘗試都會搞爆 Yarn，才換到 Python 上面玩 XD