Accessing Database Carcasonne for Machine Learning

[penguin83]

Hi,
I am a machine learning engineer and I am new on this development site. Since Carcasonne is my favourite game, I would like to train a network that can play Carcasonne. To do so, I would like to access the data base of the arena from Carcasonne.

1) Is there an API to access the database? (Of course, I could simply scrape the games, but I would like to adhere to all data privacy policies that are in place).

2) On a more general note, is already any work being done to include AI players (I know that for Can't stop there is one)?

[Tisaac]

Hi,
I am a machine learning engineer and I am new on this development site. Since Carcasonne is my favourite game, I would like to train a network that can play Carcasonne. To do so, I would like to access the data base of the arena from Carcasonne.

1) Is there an API to access the database? (Of course, I could simply scrape the games, but I would like to adhere to all data privacy policies that are in place).

2) On a more general note, is already any work being done to include AI players (I know that for Can't stop there is one)?

1) no api, and scrapping data is against website policy
2) no because bga stance on that mater is "no bot" (can't stop is the only exception to that rules). BGA is here to make people play together, not to play against bot. Automa and solo mode are more of a grey area, so there are usually tolerated

[penguin83]

Thank you for your quick reply!

I see the point that it's made for connecting people. But it's a bit sad since there is no difference between an automa that could be integrated and an AI, they are essentially doing the same.

[Tisaac]

Thank you for your quick reply!

I see the point that it's made for connecting people. But it's a bit sad since there is no difference between an automa that could be integrated and an AI, they are essentially doing the same.

That's not completely true. Both allow you to play solo, and the goal is to provide a challenging experience, but most of the automa I know usually do that by allowing the automa to have a custom set of rules. This makes sense because otherwise that would mean that you could explain a very good strategy for the game in an algorithmic way (basically the automa rulebook), which might indicates that the game lacks complexity.
On the other hand, AI usually plays the game, but since the player is not the one making the moves for the AI, you can have a very complex program for computing the AI moves.

So, to me automas are great to play alone to learn the games, and then you can try to improve against them, but it might be very different than playing against "real" players whereas the goal of AI bot is to mimic a real player.

[penguin83]

What about implementations such as robots in Challengers? You can even choose the difficulty level here and there are warnings "WARNING: strong, not suggested for multiplayer". For me, the roles of AI/bots/automa/robots are the same, if a game needs a certain player count, if you are playing at odd hours and cannot find enough players, want to experience certain dynamics that happen at a specific player count - it does not matter so much whether the robot acts according to rules or another engine.

Anyway, my initial question regarding the database has been answered. If there will be a policy change, please ping me, I love board games and would be happy to contribute.

[ufm]

The role might be the same, but the process is completely different. And that's what matters.
Btw, Can't Stop 'was' the only game supported a bot. Can't Stop bot has been discontinued after its shift to premium.

[CitadelUK]

If no bots/AI is an official BoardGameArena policy, perhaps it should me mentioned in the wiki? https://en.doc.boardgamearena.com/Bots_ ... telligence

I took a little look into this but didn't get very far as I couldn't get the ML to learn (converge on a winning strategy or even a better than random one). I think it would be possible to create machine learning neural networks that players could play against. Based on what I've seen for other board games it's been done for, it doesn't look like it would need to scrape the servers. The ML bots could probably be trained independently of the boardgamearena servers. They seem to be lists/matrices of floating points that could be loaded and run on the PHP backend in response to an AJAX call. Potentially the training could be generalised to lots of games and done on BGA servers where the rules are already implemented.

As Tisaac says, playing against bots/automata/AI is a nice way to learn games without worrying that you are wasting the time of people who are already familiar with a particular game.

[ufm]

If no bots/AI is an official BoardGameArena policy, perhaps it should me mentioned in the wiki? https://en.doc.boardgamearena.com/Bots_ ... telligence

Probably. I think the contents of that page is about automa (not AIs or bots). Even the code says 'automa'.