Training katago to play 2 moves in a row on an 11×11 board about katago HOT 10 OPEN

I have checked it out, the turns of two-stone-per-move are not chosen by players but the foxwq server. It is not Go actually from my poor view, it's a fresh another game or so. And you can not learn much something from the lucky and the random. AI too.

from katago.

I have checked it out, the turns of two-stone-per-move are not chosen by players but the foxwq server. It is not Go actually from my poor view, it's a fresh another game or so. And you can not learn much something from the lucky and the random. AI too.

You are wrong if you really think so, there is a random component but strength of schedule has a big impact. You can see some people with high level, he can win many games in a row. This game has ascending and descending levels, divided into different levels, such as Blade God > Blade Immortal > Blade Saint

from katago.

我查过了，每招两石的回合不是玩家选择的，而是foxwq服务器选择的。从我糟糕的角度来看，这实际上不是围棋，这是一个新鲜的另一种游戏。而且你不能从幸运和随机身上学到很多东西。人工智能也是。

I know, I've played this game for years flying knife's, a lot of people play it, it's very addictive, know the rules will play. It's easy to play one or two dozen games at a time. It feels like you can gradually grasp the laws of this flying dagger - what you consider a completely random game of luck. Although the appearance of flying daggers is determined by Wildfox, he limited the rounds, is appeared in the 15th to the 50th round. So it's exciting for players to be ready for the arrival of both their own and their opponent's flying daggers, and everyone chooses to play more conservatively, with the opponent more likely to take aggressive moves when their own flying daggers are fewer in number than the opponent's. Also, when the flying daggers are not finished, both sides will try to make their pieces larger than 2 breaths, so as to avoid being eaten by themselves. Also, as you know from playing this game a lot, you should try to surround yourself as thickly as possible in the beginning, that is, you should adopt the strategy of fishing first and then shuffling, with the pieces connected as much as possible. Once the flying daggers come, using them to divide the opponent's pieces can easily cause the opponent's strength to be spread out over 2 or more pieces, thus causing the passive side to spend extra pieces to make up for it, and thus have no time to attack the other side. Once the splitting side's flying daggers come again, causing the death of another area without mending pieces

Maybe what you say makes sense to a certain extent, like flying chess, I don't know if you guys know this kind of chess except in China? It's just similar to some game play that has an element of luck in it, a taste of which is considered to be completely random. But it can't be denied that there are also some games with luck components that can be controlled by humans to a certain extent, and this kind of game is very exciting, and it would be a breakthrough for katago if it could be made. For this type of gameplay that contains luck, but not completely, but mostly depends on your understanding of luck, and requires the use of computation, katago how to train this type of gameplay under the game, is expected!

You haven't grasped the essence of this style of play because you've only read about it and haven't really played it or played it much. That's normal, a lot of people haven't played it and think it's similar to gambling. Including the tips mentioned above is only a small part of personal experience, and if it involves complexity, it's not something that can be summarized by these tips, depending on the situation. And there are more tips than that.

For a few people of equal strength, that is, with equal knowledge of the laws of this flying dagger and equal strength, it might be 55 to 1. But for those who will master the law, a weaker level of Go may also win against a professional player, so this is a game with a lot of variables, but taking time to gradually improve

I believe that flying daggers, although relatively simple, can reach a point where they can completely outperform humans if they are trained in flying dagger ai. Even if a human is lucky, it's hard to win 1 out of 1000 games. Because katago can learn these laws and relies on counting 2 steps

我知道，我玩过这个游戏好几年了飞刀的，很多人都在玩，这游戏很容易上瘾，懂规则会玩的。一次玩一二十盘都很轻松。感觉能逐渐把握到这个飞刀——你们认为的完全随机运气游戏的规律。虽然飞刀出现是由野狐决定，但是他限定了回合，是出现在第15到第50回合。所以在这个过程中，玩家很刺激，随时准备己方和对方飞刀的来临，大家都会选择比较保守的玩法，当己方飞刀数量少于对方时，对方更可能采取激进的招法。还有在飞刀未结束时，双方都会尽量让自己的棋子大于2口气，从而避免自己被吃掉。还有这个游戏玩多了就知道，一开始应该尽量围厚实，就是采取先捞后洗的策略，棋子尽量连起来。一旦飞刀来了，将飞刀用在分割对方棋子的地方，很容易造成对方2块以上棋的力量分散，从而导致被动的那一方需要花额外的棋子来补棋，从而无暇攻击对方。当分割方的飞刀一旦再次来临，导致另一块没有补棋的区域死亡

也许你说的在一定程度很有道理，就像飞行棋，不知道除**以外，你们知道这种棋吗？就是类似于一些有运气成分的游戏玩法，尝尝被认为是完全随机。但是不能否认还有一部分包含运气成分的游戏玩法，可以被人类所在一定范围程度内控制，这种游戏非常刺激，如果能做出来也是对katago的一种突破。对于这一类包含运气成分，但是又不完全包含，而是大部分取决于你对这个运气的理解，需要用到计算的时候，katago如何去训练这类玩法下的游戏，令人期待

你因为只是看过，没有真正玩过，或玩的不多，所以你没有掌握这种玩法的精髓。这很正常，很多人没玩过，以为是类似于赌博。包括上面说的技巧也只是个人经验的一小部分，如果涉及到复杂的，不是这些技巧所能概括的，视情况而定。而且技巧不止这些

对于几个实力相当的，就是对这个飞刀规律认识相当，实力相当的人来说，可能是55开。但是对于那些会掌握规律的人来说，弱一点的围棋水平也可能赢职业棋手，所以这是一个变数很大，但是花时间可以逐渐提高的游戏

我相信飞刀虽然比较简单，但是如果训练飞刀ai，可以达到完全超越人类。即使人类运气再好，下1000盘也很难赢1盘。因为katago能习得这些规律，并且依赖于算2步的计算

from katago.

@awsjgy - If you're new to programming, as you mentioned in the other issue, I would not recommend you start with a project like this. But if you're willing to invest the effort, there are lots of resources online, and you can learn a lot about other things, depending on what you are interested in. Especially if there are other things you are interested in learning and developing skills for, beyond just the one single thing of training a bot to play specifically just this 11x11 Go variant.

Want to learn how to train neural nets on simple tasks? Search online for any number of tutorials that start with training a neural net to recognize digits, or other simple datasets, and try following them:
https://www.dataquest.io/blog/pytorch-for-beginners/
https://www.kaggle.com/code/amsharma7/mnist-pytorch-for-beginners-detailed-desc
https://pytorch.org/tutorials/beginner/basics/intro.html

Or if you want to understand how to implement game tree search and game playing, try a tic-tac-toe AI tutorial:
https://realpython.com/tic-tac-toe-ai-python/

or whatever other topic you like, etc.

I'm sure you can find plenty of resources in Chinese as well. You're most familiar with your own level - if some of these are too advanced already, then start with something even easier. If you need to, find and take a programming class, or follow any number of free online courses or course outlines on the basics of learning a programming language. There's a ton of all sorts of learning material on stuff like this online nowadays, as long as you're proactive in searching and finding material that works for your level of experience.

from katago.

• 如果你是编程新手，正如你在另一期中提到的，我不建议你从这样的项目开始。但是，如果您愿意投入精力，网上有很多资源，您可以了解很多其他知识，具体取决于您感兴趣的内容。特别是如果您有兴趣学习和发展技能的其他事情，而不仅仅是训练机器人专门玩这个 11x11 Go 变体的一件事。

想学习如何在简单的任务上训练神经网络吗？在线搜索任意数量的教程，这些教程从训练神经网络识别数字或其他简单数据集开始，并尝试遵循它们： https://www.dataquest.io/blog/pytorch-for-beginners/ https://www.kaggle.com/code/amsharma7/mnist-pytorch-for-beginners-detailed-desc https://pytorch.org/tutorials/beginner/basics/intro.html

或者，如果您想了解如何实现游戏树搜索和玩游戏，请尝试井字游戏 AI 教程：https://realpython.com/tic-tac-toe-ai-python/

或任何其他您喜欢的主题，等等。

我相信你也可以找到很多中文资源。你最熟悉你自己的水平 - 如果其中一些已经太高级了，那么从更容易的事情开始。如果需要，请查找并参加编程课程，或按照任意数量的免费在线课程或学习编程语言基础知识的课程大纲进行操作。现在网上有很多关于此类内容的学习材料，只要您积极主动地搜索和找到适合您经验水平的材料。

I just want to have fun, like most of the people who use katago, are they there to learn programming? They just enjoy the fun and intellectual development that katago brings to the game of Go. I'm not in the computer business, I'm interested, but it's hard to be like a professional.
我只想玩玩就行了，就像大部分使用katago的人，他们难道是为了学编程吗？而只是享受katago研究围棋游戏带来的乐趣和智力开发罢了.我不是干计算机这一行，虽然感兴趣，但是要想和专业人士一样，很难很难

from katago.

@awsjgy - If you're new to programming, as you mentioned in the other issue, I would not recommend you start with a project like this. But if you're willing to invest the effort, there are lots of resources online, and you can learn a lot about other things, depending on what you are interested in. Especially if there are other things you are interested in learning and developing skills for, beyond just the one single thing of training a bot to play specifically just this 11x11 Go variant.

Want to learn how to train neural nets on simple tasks? Search online for any number of tutorials that start with training a neural net to recognize digits, or other simple datasets, and try following them: https://www.dataquest.io/blog/pytorch-for-beginners/ https://www.kaggle.com/code/amsharma7/mnist-pytorch-for-beginners-detailed-desc https://pytorch.org/tutorials/beginner/basics/intro.html

Or if you want to understand how to implement game tree search and game playing, try a tic-tac-toe AI tutorial: https://realpython.com/tic-tac-toe-ai-python/

or whatever other topic you like, etc.

I'm sure you can find plenty of resources in Chinese as well. You're most familiar with your own level - if some of these are too advanced already, then start with something even easier. If you need to, find and take a programming class, or follow any number of free online courses or course outlines on the basics of learning a programming language. There's a ton of all sorts of learning material on stuff like this online nowadays, as long as you're proactive in searching and finding material that works for your level of experience.

Is there a way to do this by reducing the amount of calculations, for example if a computer is running 18b at 500v per second, it would be difficult to exhaust all the 2-handed combinations, more time consuming and configuration intensive. Then 1v per second is always possible, 1v is said to be able to reach amateur 6 level. This would allow us to find the 2-handed combination with the highest win rate in a short period of time, how would this be accomplished? For newbies
有没有一种方法，可以穷举，通过减少计算量的方法，比如电脑运行18b速度是500v每秒，要穷举所有2手组合的话很难，比较费时间和配置。那么每秒1v总可以了吧，1v据说也能达到业余6段水平。这样就能在短时间找出胜率最大的连续下2手的组合，这个要怎么实现呢？对于新手小白

from katago.

from katago.

@awsjgy请注意，GitHub是一个严肃的编程平台，它不是像Twitter或微信那样的社交媒体，我们不在这里聊天。您可以通过打开问题来请求新功能，例如每回合两块石头，但如果官方团队（作者和贡献者）不支持或同意，您最好先关闭问题并从其他地方获得帮助。这是GitHub的礼仪。至于每回合两块石头的功能，我可能会对GUI进行编程，@hzyhhzy可能会训练神经网络模型。就这样。 在 2023-10-17 09：47：13，awsjgy @.> 写道： @awsjgy - 如果你是编程新手，正如你在另一期中提到的，我不建议你从这样的项目开始。但是，如果您愿意投入精力，网上有很多资源，您可以了解很多其他知识，具体取决于您感兴趣的内容。特别是如果您有兴趣学习和发展技能的其他事情，而不仅仅是训练机器人专门玩这个 11x11 Go 变体的一件事。 想学习如何在简单的任务上训练神经网络吗？在线搜索任意数量的教程，这些教程从训练神经网络识别数字或其他简单数据集开始，并尝试遵循它们： https://www.dataquest.io/blog/pytorch-for-beginners/https://www.kaggle.com/code/amsharma7/mnist-pytorch-for-beginners-detailed-deschttps://pytorch.org/tutorials/beginner/basics/intro.html 或者，如果您想了解如何实现游戏树搜索和玩游戏，请尝试井字游戏 AI 教程： https://realpython.com/tic-tac-toe-ai-python/ 或您喜欢的任何其他主题，等等。 我相信你也可以找到很多中文资源。你最熟悉你自己的水平 - 如果其中一些已经太高级了，那么从更容易的事情开始。如果需要，请查找并参加编程课程，或按照任意数量的免费在线课程或学习编程语言基础知识的课程大纲进行操作。现在网上有很多关于此类内容的学习材料，只要您积极主动地搜索和找到适合您经验水平的材料。 有没有办法通过减少计算量来做到这一点，例如，如果一台计算机以每秒 18v 的速度运行 500b，则很难用尽所有 2 手组合，更耗时且配置密集。那么每秒1v总是可能的，据说1v能够达到业余6级。这将使我们能够在短时间内找到胜率最高的 2 手组合，这将如何实现？对于新手 有没有一种方法，可以穷举，通过减少计算量的方法，比如电脑运行18b速度是500v每秒，要穷举所有2手组合的话很难，比较费时间和配置。那么每秒1v总可以了吧，1v据说也能达到业余6段水平。这样就能在短时间找出胜率最大的连续下2手的组合，这个要怎么实现呢？对于新手小白 — 直接回复此电子邮件，在 GitHub 上查看或取消订阅。 您收到此消息是因为您发表了评论。消息 ID：@.>

Can. I applied in my country to ask someone to do this katago for me for the flying knife rule, and many people can't do it, including programmers. They do not know where to modify the rule code, and how the code is modified. This need algorithm underlying logic seems to want. As for this flying dagger game, the 11 way board, trained to surpass human level, I don't know how much it will cost. Because the 11-way board is not as complex as the 19-way. This seems like someone will do it, it only takes about 10 days, if they know Go ai and algorithms. But to the average programmer, they don't really do it, if they haven't studied this kind of play. Here's what I've given programmers before in terms of requirements:
Flying Dagger Go Software Development Requirements Document
Project Objective
Create software based on the rules of "Flying Dagger Go", using deep learning to train AI to realize human-machine games, with a graphical user interface (GUI) and basic Go functions.

1. Go engine and interface
   Select and possibly modify the KataGo engine to adapt to the rules of Flying Dagger Go.
   Utilize or modify existing interfaces such as lizzieyzy, or others such as sabaki, gogui, or design your own if necessary.
2. Flying Daggers rules
   11 boards, 3 "flying daggers" per side on moves 15-50, 2 moves in a row on the flying dagger turn.
3. Interface Functions
   Play, setup, setup Flying Daggers (show number and available status), connect.
4. AI Training Requirements
   Adapt the training algorithm to the rules of Flying Daggers Go, using PyTorch and KataGo, the main task is parameter tuning and training to at least exceed the human level.
5. Personnel Composition
   Software designers: architecture design and planning.
   Programmers: code implementation and testing.
   AI algorithm researchers: algorithm design, tuning and optimization.
6. Project Flow and Interaction
   Design Flying Dagger Go rules, reward and punishment rules, model training, state reception, strategy generation, interface display results, AI backend engine interconnects with the application layer through JSON and other structures, converts into actions, and UI execution.
7. Evaluation indicators
   AI and human game winning rate, interface friendliness and response speed.
8. Other
   Project iteration optimization, open source game engine selection should have good documentation and community support, model training should have clear log records and performance evaluation.
9. Training dataset and training
   Download the games of Flying Dagger from the Wildfox Go platform as the training dataset.
   Conduct spectrum training to improve the performance of the AI.
   Technical Requirements
   Programming: Python, KataGo, GUI development (Qt, Tkinter or existing open source Go interfaces).
   AI and Deep Learning: PyTorch or other frameworks, training strategy design, model evaluation and optimization.
   Data processing and interface design: Familiar with JSON, XML and other data structures, interface design and integration.
   Testing and Debugging
   Functional testing, performance testing.
   Teamwork and Communication
   Project management, effective teamwork and communication.
   Project Deliverables
   Software source code, executable programs, user manuals.
   Considerations
   Code management (Git), documentation, ongoing communication to ensure successful completion of the project.

Installment Payment
可以。我在国内申请请人帮我做这个飞刀规则的katago，很多人做不了，包括程序员。他们不知道在哪里修改规则代码，并且代码怎么修改。这个需要算法底层逻辑好像要。至于这个飞刀游戏，11路的棋盘，训练到超越人类水平，不知道需要多少成本。因为11路棋盘没有19路那么复杂。这个好像有人会做，只要10天左右，如果是懂围棋ai和算法的。但是给一般程序员，他们并不会做，如果没有研究过这种玩法的。我之前给程序员的需求是这样：
飞刀围棋软件开发需求文档
项目目标
创建基于"飞刀围棋"规则的软件，利用深度学习训练AI实现人机对弈，附带图形用户界面(GUI)和基础围棋功能。

1. 围棋引擎与界面
   选用并可能修改KataGo引擎适应飞刀围棋规则。
   利用或修改现有界面如lizzieyzy，或其他如sabaki, gogui，必要时自行设计界面。
2. 飞刀围棋规则
   11路棋盘，第15-50手每方3把“飞刀”，飞刀回合连下2手。
3. 界面功能
   下棋、摆棋、设置飞刀(显示数量和可用状态)、连线。
4. AI训练需求
   调整训练算法适应飞刀围棋规则，使用PyTorch和KataGo，主任务是参数调整和训练，至少超过人类水平。
5. 人员组成
   软件设计师: 架构设计与规划。
   程序员: 代码实现与测试。
   AI算法研究员: 算法设计、调整和优化。
6. 项目流程与交互
   设计飞刀围棋规则，奖惩规则，模型训练，状态接收，策略生成，界面显示结果，AI后台引擎通过JSON等结构与应用层互连，转换成动作，UI执行。
7. 评估指标
   AI与人对弈胜率，界面友好度和反应速度。
8. 其他
   项目迭代优化，开源游戏引擎选择应具良好文档和社区支持，模型训练应有明确的日志记录和性能评估。
9. 训练数据集与训练
   从野狐围棋平台下载飞刀游戏的棋谱作为训练数据集。
   进行跑谱训练以改进AI的性能。
   技术要求
   编程: Python, KataGo, GUI开发(Qt, Tkinter或现有开源围棋界面)。
   AI与深度学习: PyTorch或其他框架，训练策略设计，模型评估与优化。
   数据处理与接口设计: 熟悉JSON, XML等数据结构，接口设计与集成。
   测试与调试
   功能测试，性能测试。
   团队协作与沟通
   项目管理，有效的团队协作和沟通。
   项目交付物
   软件源代码，可执行程序，用户手册。
   注意事项
   代码管理(Git)，文档编写，持续沟通以确保项目成功完成。

分期付款

from katago.

@awsjgy Please be careful, the GitHub is a serious programming platform, and it is not a social media like twitter or wechat, we don't chat here. You could request a new feature such as two-stone-per-turn via opening a issue, but if it is not supported or agreed by the official team (the author and the contributors), you'd better close the issue first and get help from somewhere else. It is the etiquette of GitHub. As for two-stone-per-turn feature, I perhaps program the GUI and @hzyhhzy may train a neural network model. That's all. 在 2023-10-17 09:47:13，awsjgy @.> 写道： @awsjgy - If you're new to programming, as you mentioned in the other issue, I would not recommend you start with a project like this. But if you're willing to invest the effort, there are lots of resources online, and you can learn a lot about other things, depending on what you are interested in. Especially if there are other things you are interested in learning and developing skills for, beyond just the one single thing of training a bot to play specifically just this 11x11 Go variant. Want to learn how to train neural nets on simple tasks? Search online for any number of tutorials that start with training a neural net to recognize digits, or other simple datasets, and try following them: https://www.dataquest.io/blog/pytorch-for-beginners/https://www.kaggle.com/code/amsharma7/mnist-pytorch-for-beginners-detailed-deschttps://pytorch.org/tutorials/beginner/basics/intro.html Or if you want to understand how to implement game tree search and game playing, try a tic-tac-toe AI tutorial: https://realpython.com/tic-tac-toe-ai-python/ or whatever other topic you like, etc. I'm sure you can find plenty of resources in Chinese as well. You're most familiar with your own level - if some of these are too advanced already, then start with something even easier. If you need to, find and take a programming class, or follow any number of free online courses or course outlines on the basics of learning a programming language. There's a ton of all sorts of learning material on stuff like this online nowadays, as long as you're proactive in searching and finding material that works for your level of experience. Is there a way to do this by reducing the amount of calculations, for example if a computer is running 18b at 500v per second, it would be difficult to exhaust all the 2-handed combinations, more time consuming and configuration intensive. Then 1v per second is always possible, 1v is said to be able to reach amateur 6 level. This would allow us to find the 2-handed combination with the highest win rate in a short period of time, how would this be accomplished? For newbies 有没有一种方法，可以穷举，通过减少计算量的方法，比如电脑运行18b速度是500v每秒，要穷举所有2手组合的话很难，比较费时间和配置。那么每秒1v总可以了吧，1v据说也能达到业余6段水平。这样就能在短时间找出胜率最大的连续下2手的组合，这个要怎么实现呢？对于新手小白 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

You have a point, to make an algorithm for a flying dagger rule is harder to implement for many programmers in China. You could start by making a Go rule that plays 2 moves in a row and then go get trained. an 11 way board shouldn't take much training to get to a level beyond human, if not 19 way.
你说的有道理，要做一个飞刀规则的算法，对**很多程序员来说比较难实现。你可以先做一个连下2步的围棋规则，然后去找人训练。11路的棋盘应该不用多长时间训练就能达到超越人类水平，如果不是19路。

from katago.

@awsjgy Please be careful, the GitHub is a serious programming platform, and it is not a social media like twitter or wechat, we don't chat here. You could request a new feature such as two-stone-per-turn via opening a issue, but if it is not supported or agreed by the official team (the author and the contributors), you'd better close the issue first and get help from somewhere else. It is the etiquette of GitHub. As for two-stone-per-turn feature, I perhaps program the GUI and @hzyhhzy may train a neural network model. That's all. 在 2023-10-17 09:47:13，awsjgy @.> 写道： @awsjgy - If you're new to programming, as you mentioned in the other issue, I would not recommend you start with a project like this. But if you're willing to invest the effort, there are lots of resources online, and you can learn a lot about other things, depending on what you are interested in. Especially if there are other things you are interested in learning and developing skills for, beyond just the one single thing of training a bot to play specifically just this 11x11 Go variant. Want to learn how to train neural nets on simple tasks? Search online for any number of tutorials that start with training a neural net to recognize digits, or other simple datasets, and try following them: https://www.dataquest.io/blog/pytorch-for-beginners/https://www.kaggle.com/code/amsharma7/mnist-pytorch-for-beginners-detailed-deschttps://pytorch.org/tutorials/beginner/basics/intro.html Or if you want to understand how to implement game tree search and game playing, try a tic-tac-toe AI tutorial: https://realpython.com/tic-tac-toe-ai-python/ or whatever other topic you like, etc. I'm sure you can find plenty of resources in Chinese as well. You're most familiar with your own level - if some of these are too advanced already, then start with something even easier. If you need to, find and take a programming class, or follow any number of free online courses or course outlines on the basics of learning a programming language. There's a ton of all sorts of learning material on stuff like this online nowadays, as long as you're proactive in searching and finding material that works for your level of experience. Is there a way to do this by reducing the amount of calculations, for example if a computer is running 18b at 500v per second, it would be difficult to exhaust all the 2-handed combinations, more time consuming and configuration intensive. Then 1v per second is always possible, 1v is said to be able to reach amateur 6 level. This would allow us to find the 2-handed combination with the highest win rate in a short period of time, how would this be accomplished? For newbies 有没有一种方法，可以穷举，通过减少计算量的方法，比如电脑运行18b速度是500v每秒，要穷举所有2手组合的话很难，比较费时间和配置。那么每秒1v总可以了吧，1v据说也能达到业余6段水平。这样就能在短时间找出胜率最大的连续下2手的组合，这个要怎么实现呢？对于新手小白 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you commented.Message ID: @.>

That's fine too, the hard work starts with changing the gui and katago to the algorithm and code for the next 2 steps in a row. The rest of the work will be left to training. The ai trained this way should be able to play flying dagger, it won't be very strong or powerful but it helps. When do you expect to make it, and can you let me know first when it's done? Do you know @hzyhhzy, I don't know him well.
也行吧，辛苦你先将gui和katago改成连续下2步的算法和代码。剩下的工作就交给训练。这样训练出的ai应该玩飞刀也可以的，虽然不会很强很强，但是有帮助。你估计什么时候做出来呢，到时候做出来了能第一时间告诉我吗？你认识@hzyhhzy吗，我和他不熟

from katago.