Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Tables]

This is a half-baked idea I came up with after seeing Donald X talk about 'medium bots' in another thread.

What if there was a bot which knew the entire game state, including cards on it's deck, card in your hand etc. and made decisions based on it? It would still only be an AI, so it wouldn't play perfectly, but it might be able to be tailored to use it's information in certain ways, e.g. knowing if you can block an attack, or if Pirate Ship will hit or not, or what to Wishing Well for and other things. Obviously it'd 'unfair', but could it be fun to play against?

[GeoLib]

I think that would actually be pretty cool. It would make up for a lot of the things that the bots do badly.

I think the biggest problem is that it would seriously mess with the balance between cards. Wishing well becomes lab for $3 and mystic becomes activated conspirator are obvious examples. Also Mountebank and Rebuild become even more powerful than they already are. Lookout is always safe. Buying Noble Brigands can be perfectly timed. Doctor is crazy. Journeyman is better.

I still think it would be fun to play against though. Probably not a high priority for Goko. It would be kind of cool to make these part of the adventures. One of the boss levels is against psychic bot. Depending on how the Goko bots are currently implemented it wouldn't necessarily even be that hard to do. If they've got some kind of probability calculations going on for things like WW, just have that part replaced with the exact answer.

[Awaclus]

I think the biggest problem is that it would seriously mess with the balance between cards. Wishing well becomes lab for $3 and mystic becomes activated conspirator are obvious examples. Also Mountebank and Rebuild become even more powerful than they already are. Lookout is always safe. Buying Noble Brigands can be perfectly timed. Doctor is crazy. Journeyman is better.

Would the bot "know" that it's psychic? In other words, would it still buy a Lab instead of two Wishing Wells with $6 and two buys?

[DG]

Giving bots extra information is useless unless their decision making is improved to make use of this information. Whilst there are some immediately achievable benefits, such as taking pirate ships for +coins if the attack would fail, a lot of decisions are more difficult. The bots seem to pick a terminal to play based on a fixed priority list so it would be a big advance in coding for them to compare the results of two actions, even knowing the decks, and evaluate the better outcome.

[Tables]

Giving bots extra information is useless unless their decision making is improved to make use of this information. Whilst there are some immediately achievable benefits, such as taking pirate ships for +coins if the attack would fail, a lot of decisions are more difficult. The bots seem to pick a terminal to play based on a fixed priority list so it would be a big advance in coding for them to compare the results of two actions, even knowing the decks, and evaluate the better outcome.

Yeah, this is what I meant when I mentioned that they'd still only be AIs. It might make some decisions differently, and would seem to usually get luckier with certain cards, but otherwise it'd probably just be like playing Lord Bottington.

[popsofctown]

I think the biggest problem is that it would seriously mess with the balance between cards. Wishing well becomes lab for $3 and mystic becomes activated conspirator are obvious examples. Also Mountebank and Rebuild become even more powerful than they already are. Lookout is always safe. Buying Noble Brigands can be perfectly timed. Doctor is crazy. Journeyman is better.

Would the bot "know" that it's psychic? In other words, would it still buy a Lab instead of two Wishing Wells with $6 and two buys?

As long as Trash for Benefit is absent Wishing Well is strictly better, because you can choose not to draw the second card.  Your example is excessive, such a bot would buy WW over Lab with 5$ and 1 buy.

[soulnet]

As long as Trash for Benefit is absent Wishing Well is strictly better, because you can choose not to draw the second card.  Your example is excessive, such a bot would buy WW over Lab with 5$ and 1 buy.

Actually, 3-piling considerations and how much the opponent would like to buy one is a concern as well. Also, variety for Cornucopia cards can make Lab better if you already have WW.

[popsofctown]

As long as Trash for Benefit is absent Wishing Well is strictly better, because you can choose not to draw the second card.  Your example is excessive, such a bot would buy WW over Lab with 5$ and 1 buy.

Actually, 3-piling considerations and how much the opponent would like to buy one is a concern as well. Also, variety for Cornucopia cards can make Lab better if you already have WW.

Well "This card is differently named and part of a different pile" is a natural given for every comparison.  That should fit into the exceptions to the term "strictly better", which is naturally an approximation.

Being able to draw only one card is a major difference though and I shudder to think that anyone always wishes for the top card of their deck if they know what it is.

[luser]

Giving bots extra information is useless unless their decision making is improved to make use of this information. Whilst there are some immediately achievable benefits, such as taking pirate ships for +coins if the attack would fail, a lot of decisions are more difficult. The bots seem to pick a terminal to play based on a fixed priority list so it would be a big advance in coding for them to compare the results of two actions, even knowing the decks, and evaluate the better outcome.

I had with similar idea, to create a bot that would handle endgame which would override existing bot when game will likely end in next three turn. It would generalize variants of PPR. Its relatively easy estimate odds of next two opponent hands when you know what is in his discard piles so I wondered how stronger it would be versus deck tracking of opponent.

If I extend this to psychic bot that can predict order after shuffle then it could on most boards play endgame perfectly as there are only few choices you could make. If both players are restricted to BM+X this could find outcome on turn 1.

For simplicity assume that no duration or topdecking. Then a turn could be described only by number of cards you draw during that turn and what cards you gain, your next turns will be same regardless how did you played. This is better explained on example, say that you have village/smithy engine and you know that your deck is ordered

s sm e e e s c sm g c c sm vi e e g c c c c c sm sm sm

Then optimum is play smithy first turn for duchy, on second turn play smithy but not village and on third turn buy duchy.  On third turn it does not matter if you played smithy in first or second turn as deck is in both cases following:

e e g c c c c c sm sm sm

and what you bought only comes after reshuffle.

So you could create memoization table where you note these pairs for both players along with number of greens remaining, run it in parallel with normal bot and if it finds that it can win within say 100ms then it overrides default bot.

This search could be improved if there is no trashing by replacing green cards with curses and giving appropriate vp tokens.

[amalloy]

As long as Trash for Benefit is absent Wishing Well is strictly better, because you can choose not to draw the second card.  Your example is excessive, such a bot would buy WW over Lab with 5$ and 1 buy.

WW isn't strictly better (and neither is Mystic - quite - an activated Conspirator), because information is revealed to the opponent by the correct guessing. If your opponent plays three Mystics successfully, naming Torturer each time, and then plays a village, you're in a better position than if Conspirator had drawn those Torturers in secret. Likewise he can reveal he's drawing enough money to buy the last Province, so that on his Governor play you know to remodel your 4-costs into Duchies, etc etc.