Dominion Strategy Forum

Please login or register.

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

Author Topic: Cute little idea for game based on game logs  (Read 4688 times)

0 Members and 1 Guest are viewing this topic.

rrenaud

  • Administrator
  • *****
  • Offline Offline
  • Posts: 991
  • Uncivilized Barbarian of Statistics
  • Respect: +1197
    • View Profile
    • CouncilRoom
Cute little idea for game based on game logs
« on: April 21, 2014, 11:41:14 am »
+10

A game based on game logs?  We must go deeper.

Let's call it back seat driving for Dominion.  Or Dominion Offline.

Imagine you could find games between highly rated players in which the players strategies diverged meaningfully.  Further, imagine you could generate a terse summary of the strategy employed (say, Steward/Steward -> Minions or Tactician/Treasure Maps). 

You could show players the board and the strategies employed, and then the back seat driver picks who he thinks won.  Obviously, you want the guess correctly.  You could keep records of predictive accuracy, have a leaderboard, etc.

This sort of simultaneously selects for interesting boards (good players diverged) and focuses on the strategic part of the game rather than the tactical part.  It's totally plausible that this could actually be fun for hardcore F.DSers, and you could easily play it in 2 minute intervals, while on the subway, etc.
Logged

theblankman

  • Witch
  • *****
  • Offline Offline
  • Posts: 461
  • Respect: +383
    • View Profile
Re: Cute little idea for game based on game logs
« Reply #1 on: April 21, 2014, 12:42:07 pm »
+2

Neat concept, and finding games between highly rated players on log search shouldn't be all that hard.  "in which the players strategies diverged meaningfully" might be the rub... but I suppose you could do something like measure the deck-overlap at game end, i.e. how much of each player's final deck was the same, with a lower number or cards in common indicating divergence (could be done in total cards or uniquely named cards, not sure which is the better metric)? 

I could probably write a script that parses logs and calculates the number of cards in common in an afternoon (getting a free afternoon would be a feat though :) ), and I bet AI could add that kind of functionality to gokosalvager.com's log search if asked nicely enough (free afternoon caveat probably applies to him as well).  Then maybe if you want to "GM" the thing, you grab an appropriate game every so often, make a pick-the-winner poll on here with the game description you have in mind, and away we go. 

Don't know about the tracking and leaderboard part, it could be just for fun unless you really want to do the stat-keeping.  Either way it would be an interesting way to think about games and probably generate some good strategic discussion after the answer's revealed... I can envision long threads about how/why one strategy was just slightly better on that particular board. 
« Last Edit: April 21, 2014, 12:43:43 pm by theblankman »
Logged
it's a shame that full-random is the de facto standard

Awaclus

  • Adventurer
  • ******
  • Offline Offline
  • Posts: 11809
  • Shuffle iT Username: Awaclus
  • (´。• ω •。`)
  • Respect: +12849
    • View Profile
    • Birds of Necama
Re: Cute little idea for game based on game logs
« Reply #2 on: April 21, 2014, 12:50:34 pm »
0

Neat concept, and finding games between highly rated players on log search shouldn't be all that hard.  "in which the players strategies diverged meaningfully" might be the rub... but I suppose you could do something like measure the deck-overlap at game end, i.e. how much of each player's final deck was the same, with a lower number or cards in common indicating divergence (could be done in total cards or uniquely named cards, not sure which is the better metric)?
I think it would be a better idea to categorize the cards in some way - if player A buys Journeymen while player B buys Catacombs for their engines, it would recognize the strategies as similar.
Logged
Bomb, Cannon, and many of the Gunpowder cards can strongly effect gameplay, particularly in a destructive way

The YouTube channel where I make musicDownload my band's Creative Commons albums for free

rrenaud

  • Administrator
  • *****
  • Offline Offline
  • Posts: 991
  • Uncivilized Barbarian of Statistics
  • Respect: +1197
    • View Profile
    • CouncilRoom
Re: Cute little idea for game based on game logs
« Reply #3 on: April 21, 2014, 01:00:01 pm »
0

I wrote some code to do card similarity before (unfortunately, it depended on other data in councilroom that is not available for new cards).  All of the pictures have been lost to history, but some textual outputs are here that show it's sort of reasonable.

http://forum.dominionstrategy.com/index.php?topic=2996.msg51812#msg51812
Logged

qmech

  • Torturer
  • *****
  • Offline Offline
  • Posts: 1918
  • Shuffle iT Username: qmech
  • What year is it?
  • Respect: +2320
    • View Profile
Re: Cute little idea for game based on game logs
« Reply #4 on: April 21, 2014, 04:34:46 pm »
+10

Okay, let's give this a go.



Code: [Select]
Cellar, Courtyard, Masterpiece, Warehouse, Woodcutter, Gardens, Cache, Explorer, Minion, AltarNo Colonies, no Shelters

Masterpiece/Gardens (3/4) or Altar/Minion (3/4)?





Code: [Select]
Transmute, Duchess, Chancellor, Great Hall, Loan, Masquerade, Counting House, Harvest, Mint, RebuildColonies, no Shelters

Masquerade/Mint (3/4) or Rebuild (4/3)?





Code: [Select]
Chancellor, Great Hall, Wishing Well, Advisor, Silk Road, Treasure Map, Harvest, Pillage, Saboteur, HoardNo Colonies, no Shelters

Wishing Well/Treasure Map (4/3) or Harvest/Hoard (5/2)?





Code: [Select]
Haven, Fishing Village, Shanty Town, Advisor, Feodum, Ironmonger, Counting House, Rebuild, Trading Post, ForgeColonies, no Shelters

Forge and buy Colonies (4/3) or Rebuild (4/3)?



As well as finding a compact description of the strategy, you probably want to know other things like the opening splits.  There's also no guarantee that the games aren't full of blunders.
« Last Edit: April 21, 2014, 05:07:11 pm by qmech »
Logged

LastFootnote

  • Adventurer
  • ******
  • Offline Offline
  • Posts: 7495
  • Shuffle iT Username: LastFootnote
  • Respect: +10721
    • View Profile
Re: Cute little idea for game based on game logs
« Reply #5 on: April 21, 2014, 04:36:44 pm »
+3

Do those games have Colonies? How about Shelters?
Logged

qmech

  • Torturer
  • *****
  • Offline Offline
  • Posts: 1918
  • Shuffle iT Username: qmech
  • What year is it?
  • Respect: +2320
    • View Profile
Re: Cute little idea for game based on game logs
« Reply #6 on: April 21, 2014, 04:44:00 pm »
+3

Do those games have Colonies? How about Shelters?

I'll edit that and the starting splits in.

Andrew, could the Colonies and Shelters information be added to the Visualizer?
« Last Edit: April 21, 2014, 04:52:37 pm by qmech »
Logged

Mic Qsenoch

  • 2015 DS Champion
  • *
  • Offline Offline
  • Posts: 1709
  • Respect: +4329
    • View Profile
Re: Cute little idea for game based on game logs
« Reply #7 on: April 21, 2014, 05:01:09 pm »
+3

Do those games have Colonies? How about Shelters?

I'll edit that and the starting splits in.

The second game does have Colonies.
Logged

qmech

  • Torturer
  • *****
  • Offline Offline
  • Posts: 1918
  • Shuffle iT Username: qmech
  • What year is it?
  • Respect: +2320
    • View Profile
Re: Cute little idea for game based on game logs
« Reply #8 on: April 21, 2014, 05:08:53 pm »
0

Do those games have Colonies? How about Shelters?

I'll edit that and the starting splits in.

The second game does have Colonies.

:-[

This is why it should be automatic.   
Logged

AdamH

  • Margrave
  • *****
  • Offline Offline
  • Posts: 2833
  • Shuffle iT Username: Adam Horton
  • You make your own shuffle luck
  • Respect: +3879
    • View Profile
    • My Dominion Videos
Re: Cute little idea for game based on game logs
« Reply #9 on: April 22, 2014, 09:22:02 am »
0

I question whether you can get any useful data out of this without simulation (or even with it).

Different draws have to factor into this quite a bit; even not taking into account the cases where your draws might lead you into one strategy over the other, but I think you'd need a really large sample size of actual games played that meet a specific criteria to make any kind of comparison. At least 100, I'd say. Are any of these comparisons giving that many games to look at?

As much as we say "depends on the kingdom" around here, I think that especially applies here. There could be cards that support one strategy more than the other in certain games, and that could affect the win rate. Automatically categorizing this type of thing seems very difficult, so I would think any conclusions you'd draw from doing something like this would have to include a statement like "Strategy A's higher win rate could have something to do with it being easier to find support cards for it"

It seems like if you wanted to answer more theoretical questions like "is strategy A better than strategy B?" then using simulators will give you better results. This kind of analysis is probably geared more towards practical applications if I'm understanding correctly, right?
Logged
Visit my blog for links to a whole bunch of Dominion content I've made.

rrenaud

  • Administrator
  • *****
  • Offline Offline
  • Posts: 991
  • Uncivilized Barbarian of Statistics
  • Respect: +1197
    • View Profile
    • CouncilRoom
Re: Cute little idea for game based on game logs
« Reply #10 on: April 22, 2014, 10:40:08 am »
0

I am confused.  You get to see the kingdom.

I agree that there is a lot of noise/randomness in the which strategy will work better, since the edge between different strategies chosen by good players will probably tend to be small (good players tend to not execute very poor strategies).  You could pick the strategy that will win 60% of the time on a given board, and by definition, you'd be wrong 40% of the time.

But if you consistently pick the strategy with the edge, as executed by very good but imperfect players on both sides, your winning percentage will show it, even if each kingdom shows up only once in the logs.

I agree that this would be very hard to actually aggregate up to find strategy A is best, or strategy A is better than strategy B, since even accurately classifying strategies is difficult.

But on the flip side, the problem with simulators in practice is that they execute complex strategies poorly. 
Logged

AdamH

  • Margrave
  • *****
  • Offline Offline
  • Posts: 2833
  • Shuffle iT Username: Adam Horton
  • You make your own shuffle luck
  • Respect: +3879
    • View Profile
    • My Dominion Videos
Re: Cute little idea for game based on game logs
« Reply #11 on: April 22, 2014, 10:52:31 am »
0

I'm just wondering if there's any attempt to factor in support cards that appear in the kingdoms you see, since I assume each game you look at has a different kingdom. Looks like we're just abstracting that away, though.
Logged
Visit my blog for links to a whole bunch of Dominion content I've made.

blueblimp

  • Margrave
  • *****
  • Offline Offline
  • Posts: 2849
  • Respect: +1559
    • View Profile
Re: Cute little idea for game based on game logs
« Reply #12 on: April 22, 2014, 02:24:37 pm »
+4

Thanks for putting those together, qmech. They were a lot of fun!
Logged
Pages: [1]
 

Page created in 0.138 seconds with 20 queries.