Dominion Strategy Forum

Please login or register.

Login with username, password and session length
Pages: [1]

Author Topic: Deck Scoring  (Read 12267 times)

0 Members and 1 Guest are viewing this topic.

UmYeah

  • Pawn
  • **
  • Offline Offline
  • Posts: 4
  • Respect: 0
    • View Profile
Deck Scoring
« on: June 30, 2011, 03:04:31 pm »
0

I am new to the world of Dominion, but have quickly fallen in love with the game and the mathmatical possibilities.

Is there an "commonly accepted" algorithm to score a deck of cards, or a games current state?  (Say between 0 and 1)

Here is where I am coming from: 

Taking the "game tree" approach to analyzing Tic-Tac-Toe you can fully populate the game tree such that every possible turn can be mapped out from the start to every possible finish.  Once you know the outcome of the end state you know to stay on a path through the game/decision tree that will lead to a win or a tie.  Tic-Tac-Toe is a very easy game and if you know the tree you will never lose.

Taking the game tree approach to analyzing Chess is a very difficult proposition.  There is yet to be a computer able to calculate and hold in memory all the potential decisions from the start to the many ends, and there may never be one that can becuase of the immense number of possibilities.  You can model the first n number of turns, and can model the more limited set end games but connecting them is impossible.  They way AI systems do this is to look at a particular board position and rank it between 0 and 1.  1 being a winning position, 0 being a losing position, 0.5 being a 50% chance of winning.  While you may not be sure of a particular decision you are best to go down the path that has the higest %win.

With that as a background, and assuming that it is similarly difficult to map the entire Dominion Game Tree, does anyone know of prior work to rank a particular deck or decks in a game on their %win potential?

If not, what would be a good ranking algorithm?  I.e. the more provinces and Gold get a higher score, the gardens/workshop combo gives it a boost, a particular engine strength, negativly coorelated with curses, etc...

Logged

WanderingWinder

  • Adventurer
  • ******
  • Offline Offline
  • Posts: 5275
  • ...doesn't really matter to me
  • Respect: +4384
    • View Profile
    • WanderingWinder YouTube Page
Re: Deck Scoring
« Reply #1 on: June 30, 2011, 03:20:46 pm »
0

Too complicated. Any single kingdom would be complex sorta like chess is, though probably a lot simpler, maybe simpler than checkers for most boards. If you want to solve all the kingdoms, it's going to be a lot more complex than chess is, i.e. it's never being solved.
To rate the winning chances, it's going to be really really difficult. Probably your best chance is some machine learning thing based on millions of high-caliber games, but even this is just not going to be very good/feasible.

hyku

  • Pearl Diver
  • **
  • Offline Offline
  • Posts: 11
  • Respect: 0
    • View Profile
Re: Deck Scoring
« Reply #2 on: June 30, 2011, 04:15:13 pm »
0

One standard to measure the strength of a deck in a game is to measure the average money per hand/card. So starting the game everyone starts with a hand value of $3.5 and a average card value of $0.7. Obviously it gets more difficult to find average card value when actions are involved but its still possible to get estimates.  Like smithy basically draws 2 extra cards so adding a smithy can be estimated by (total money)/(total cards - 2). So adding a smithy and silver on your first 2 buys gets (7+2)/(12-2) = $0.9 per card. Now if you wont be able to play all your actions the value is thrown off. So the more chance of terminal cards being drawn in the same hand the more the estimate suffers. Also it doesn't measure the ability of your deck to hurt your opponents with attacks, it just gives a snapshot of buying power for the moment.
Logged

randomdragoon

  • Salvager
  • ****
  • Offline Offline
  • Posts: 67
  • Respect: +15
    • View Profile
Re: Deck Scoring
« Reply #3 on: June 30, 2011, 07:01:07 pm »
0

Quote
Is there an "commonly accepted" algorithm to score a deck of cards, or a games current state?  (Say between 0 and 1)
Nope :) If you can think of good one, you can make a pretty good dominion-playing AI. As far as I can tell, there isn't a skilled AI Dominion player out there yet.
Logged

UmYeah

  • Pawn
  • **
  • Offline Offline
  • Posts: 4
  • Respect: 0
    • View Profile
Re: Deck Scoring
« Reply #4 on: June 30, 2011, 07:12:48 pm »
0

Too complicated. Any single kingdom would be complex sorta like chess is, though probably a lot simpler, maybe simpler than checkers for most boards. If you want to solve all the kingdoms, it's going to be a lot more complex than chess is, i.e. it's never being solved.
To rate the winning chances, it's going to be really really difficult. Probably your best chance is some machine learning thing based on millions of high-caliber games, but even this is just not going to be very good/feasible.

I agree that it would be very difficult and you may never solve the entire tree.  With around 100 card types and multiples it is very large.

I did see that you can get game data from isotropic, but high caliber...  I know that my games so far are not the high caliber ones.  ;)
« Last Edit: June 30, 2011, 07:23:51 pm by UmYeah »
Logged

UmYeah

  • Pawn
  • **
  • Offline Offline
  • Posts: 4
  • Respect: 0
    • View Profile
Re: Deck Scoring
« Reply #5 on: June 30, 2011, 07:15:22 pm »
0

One standard to measure the strength of a deck in a game is to measure the average money per hand/card. So starting the game everyone starts with a hand value of $3.5 and a average card value of $0.7. Obviously it gets more difficult to find average card value when actions are involved but its still possible to get estimates.  Like smithy basically draws 2 extra cards so adding a smithy can be estimated by (total money)/(total cards - 2). So adding a smithy and silver on your first 2 buys gets (7+2)/(12-2) = $0.9 per card. Now if you wont be able to play all your actions the value is thrown off. So the more chance of terminal cards being drawn in the same hand the more the estimate suffers. Also it doesn't measure the ability of your deck to hurt your opponents with attacks, it just gives a snapshot of buying power for the moment.

Average money per hand would be a good start.  It does get complicated to analyze how many $s it is worth to to hurt others or how how much you should lower your $/hand by how your opponents could harm you...

Logged

UmYeah

  • Pawn
  • **
  • Offline Offline
  • Posts: 4
  • Respect: 0
    • View Profile
Re: Deck Scoring
« Reply #6 on: June 30, 2011, 07:22:43 pm »
0

Quote
Is there an "commonly accepted" algorithm to score a deck of cards, or a games current state?  (Say between 0 and 1)
Nope :) If you can think of good one, you can make a pretty good dominion-playing AI. As far as I can tell, there isn't a skilled AI Dominion player out there yet.

Yeah, that I why I am interested...

I guess it would be easy to build a Big Money or Smithy-Big Money player... but that would be middle of the pack at best.

Are there any AI Dominion players out there right now?  If there are any idea what they are ranked?
Logged

DG

  • Governor
  • *****
  • Offline Offline
  • Posts: 4074
  • Respect: +2624
    • View Profile
Re: Deck Scoring
« Reply #7 on: June 30, 2011, 10:06:16 pm »
0

It's quite hard to value a deck. A deck could have very little treasure, some curses, a couple of tacticians, and some golems. That doesn't look like a winner. However if there's a vault in the kingdom supply then the deck might become a lot better, provided you buy the vault, but that could be far too difficult for any algorithm to work evaluate. Let's not even consider how to rate your opponent's deck once you're playing a vault every turn with golem.
Logged

rrenaud

  • Administrator
  • *****
  • Offline Offline
  • Posts: 991
  • Uncivilized Barbarian of Statistics
  • Respect: +1197
    • View Profile
    • CouncilRoom
Re: Deck Scoring
« Reply #8 on: July 06, 2011, 10:36:28 am »
0

I am working on a system to evaluate probability of winning given a game state.  rspeer's golem project does the same thing.  Hopefully the initial version will show up on councilroom soon.

The "formula" is however, a very complicated thing, learned from hundreds of thousands of isotropic games.

https://github.com/rspeer/golem

I am not sure if he has been playing ranked matches with golem.
Logged
Pages: [1]
 

Page created in 0.075 seconds with 20 queries.