Print

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

[prom_vrt]

So far the AIs in Goko Dominion tend to have different strategy.

Serf bot
Villager bot
Banker bot
Defender bot
Conquerer bot
Lord Bottington
Warlord bot

I can guess from the name, Banker bot is likely to employ big money and villager bot employs villages. Anyway I don't know the actual bot strategy programming. Anyone here knows this please provide me with the answer. Thanks

[eHalcyon]

You've posted this in the wrong forum.  This is more of a General Discussion thing, not an Article.

Nobody knows the actual programming, other than the developers.  I know that they had a person or a team working on AI, so it's quite possible that the fruits of their labor are proprietary, no matter the stupid mistakes that the bots sometimes make.  I think Conqueror Bot or Warlord Bot tend to favour attack cards?  Not sure.

As for Lord Bottington, you can get some insight on his strategies in this thread.

[ftl]

Lord Bottington is the default, programmed as best as they can.

Villager bot prefers actions. Banker bot prefers treasures. Defender bot prefers reaction. Warlord bot prefers attacks. I think conqueror bot prefers victory cards. Serf bot just plays randomly with no strategy.

[DStu]

You've posted this in the wrong forum.  This is more of a General Discussion thing, not an Article.

Nobody knows the actual programming, other than the developers.  I know that they had a person or a team working on AI, so it's quite possible that the fruits of their labor are proprietary, no matter the stupid mistakes that the bots sometimes make.  I think Conqueror Bot or Warlord Bot tend to favour attack cards?  Not sure.

As for Lord Bottington, you can get some insight on his strategies in this thread.

There is also a decent approximation of LordBottingham in dominiate:

Code: [Select]

{
  name: 'LordBottington'
  requires: ['Rats']
  gainPriority: (state, my) -> [
    "Province"
    "Duchy" if 0 < state.gainsToEndGame() <= 5
    "Estate" if 0 < state.gainsToEndGame() <= 2
    "Gold"
    "Rats" if my.countInDeck("Rats") < 2
    "Silver"
    "Copper" if state.gainsToEndGame() <= 3
  ]

  wantPlayRats: (state, my) ->
    100 * Math.random() < 90
}


[DG]

There are a lot of different facets to their programming but they typically pick the high cost terminals they want in their decks and buy a limited number of those terminals. Otherwise they buy the highest card possible for the coins they have available. They buy two potions in alchemy kingdoms even if they don't actually want to buy the alchemy cards when the opportunity comes. They always gain cards if they can even if they have no purpose. They buy copper if the deck is low on treasures.

The card play rules seem to have little flexibility and just play out action cards in priority (cost?) order. All actions will be played even if the results are detrimental.

[Tables]

I remember asking before. I think FTL is correct about all of them except Village bot, which likes increasing it's handsize (yeah, not playing villages... blame Goko for the name I guess).

[Lord Bottington]

There is also a decent approximation of LordBottingham in dominiate:

Code: [Select]

{
  name: 'LordBottington'
  requires: ['Rats']
  gainPriority: (state, my) -> [
    "Province"
    "Duchy" if 0 < state.gainsToEndGame() <= 5
    "Estate" if 0 < state.gainsToEndGame() <= 2
    "Gold"
    "Rats" if my.countInDeck("Rats") < 2
    "Silver"
    "Copper" if state.gainsToEndGame() <= 3
  ]

  wantPlayRats: (state, my) ->
    100 * Math.random() < 90
}


Excuse me, I happen to be rather good at reading code, and I object to the notion that I sometimes randomly won't play my Rats. I think you'll find that as a cantrip, it's a fantastic card to play EVERY TIME it's in hand.

[luser]

Also when I tried how fast I could do a 12 colony megaturn I found that best strategy versus serf bot is include stonemason and chapel in kingdom.

As serf bots know that early trashing is beneficial they decided that best opening is overpay stonemason to get extra two stonemasons and silver followed by buying lot of chapels.

[LibraryAdventurer]

This thread kinda makes me ashamed to have about a 50% win rate against the weaker bots (I mean villager bot and warlord bot, not serf bot).

[jsh357]

This thread kinda makes me ashamed to have about a 50% win rate against the weaker bots (I mean villager bot and warlord bot, not serf bot).

I lose to the bots a lot too.  Don't feel bad; they are very good at effective formulaic strategies.  Sometimes if you're trying to be creative it just isn't as good as what they're doing and you lose.

[Donald X.]

Here's a question. They would kind of like a medium bot - better than serf bot, not as good as all the rest. The non-serf bots are all pretty close in skill, which wasn't so much intended.

So the question is, what behavior would be good for a medium bot? In what ways should it intentionally be worse, that will leave the game as enjoyable as possible for the people at that skill level? The idea being to make the bot worse without making it so much look stupid.

I'll get the ball rolling.

- Ignore PPR.
- Buy one more terminal than is reasonable.

[Stealth Tomato]

Here's a question. They would kind of like a medium bot - better than serf bot, not as good as all the rest. The non-serf bots are all pretty close in skill, which wasn't so much intended.

So the question is, what behavior would be good for a medium bot? In what ways should it intentionally be worse, that will leave the game as enjoyable as possible for the people at that skill level? The idea being to make the bot worse without making it so much look stupid.

I'll get the ball rolling.

- Ignore PPR.
- Buy one more terminal than is reasonable.

I like the extra terminal.
Ignoring PPR might be good, although I'd tend to prefer deficiencies that teach the player something via counterexample. PPR isn't something you can really do that with, since the counterexample is not doing it and thus not giving the player the idea that PPR is a thing.

Some other ideas that would tend to mirror low-to-mid-level play:

- Overvalue cantrips
- Undervalue Silver
- Overprioritize Village equivalents (buy too early and too many)

[KingZog3]

Here's a question. They would kind of like a medium bot - better than serf bot, not as good as all the rest. The non-serf bots are all pretty close in skill, which wasn't so much intended.

So the question is, what behavior would be good for a medium bot? In what ways should it intentionally be worse, that will leave the game as enjoyable as possible for the people at that skill level? The idea being to make the bot worse without making it so much look stupid.

I'll get the ball rolling.

- Ignore PPR.
- Buy one more terminal than is reasonable.

Some other ideas that would tend to mirror low-to-mid-level play:

- Overvalue cantrips
- Undervalue Silver
- Overprioritize Village equivalents (buy too early and too many)

With a good engine and only 1 Village source, the split can be important enough to make them a priority, locking the other player into either a weaker engine or weaker BM.

How about:
- Overvalue Alt VP (Great Halls, Gardens when situation is not ideal), or the opposite to undervalue them when they are strong
- Start them with 4 Estates Wait no

[DG]

So the question is, what behavior would be good for a medium bot? In what ways should it intentionally be worse, that will leave the game as enjoyable as possible for the people at that skill level?

I'll have a think about it but there are plenty of obvious fixes to create better bots above the existing ones, while it is less obvious how to degrade the existing ones. Example - improve card play order, don't buy cards that are never played (tacticians), don't play cards if they are completely detrimental (trading post trashing province), etc.

[DG]

If we look at the existing bots, the most successful ones are the banker and conqueror bots. The conqueror bot realizes the value of alternate vp so that logic should be denied to the medium bot. The banker plays simple treasure based strategies so also making the medium bot simple isn't actually going to make it worse. A medium bot also shouldn't be showcasing bad play to beginners in case they copy it, such as consistently buying vp at the wrong time.

The worse bots are the warlord bot and the villager bot. The warlord bot tends to self destruct by trashing and cycling the deck too much. It's so bad that I wouldn't want to see a new bot do the same. This leaves the villages bot and this might be the bot to follow. If the medium bot indulges in engine-type cards without building an engine then it will perform poorly. These could include cards like lookout, village, and oasis but this would involve deliberate complex decisions to buy badly instead of just taking silver. This does also carry a risk of the bot playing the action cards so badly that it looks stupid.

We can also consider some other things
- Always gaining cards even if they are no use - bots already do this
- Always buying the highest card available - bots already do this except the villager bot can split coins with +buys
- Reasonable but not optimal vp buying rules - bots already do this although they make no allowance for the number of players
- No defense against opponent's winning play - bots already only defend by buying points and they let piles get dangerously empty
- Missing three pile endings - the bots always win if they can but that's right surely?
- Random purchases - I'm guessing this is what the serf bot does already
- Card play - I think all the bots have the same card play rules. I don't see any extra room for deliberate stupidity
- Taking too many terminals and unproductive village cards - there is scope for this but needs complex, deliberate, varied failings
- Overpay, potions, coin tokens, vp tokens - bots already play these badly

[LibraryAdventurer]

I'm seeing that it could be pretty difficult to program a bot to play realistically mediocre. I'm thinking how could they make it play more like me?  Well, what do I tend to do badly?  hmmm.  I don't usually think ahead much when looking at the kingdom at the beginning of the game. I have a strong aversion to buying terminal draw without having plenty of villages (I'm still not really convinced that that's a bad thing...). It's hard to think of much that I do badly because if I knew what I was doing wrong, then I'd wouldn't be doing it (as much...).

[pst]

I suggest making small random adjustments for the preferences of the different cards at the beginning of the game, and then stick to those values for the whole game.

That leads to variation on what errors they make, making for different weaknesses to find how to beat for the player.

Some other ideas come from trying to replicate mistakes that many beginners make:

• not ending the game with a win (preferring another Province instead) -- typical mistake, and instructional
• ending the game, even though you are losing (but that is not as good/interesting to copy)
• increase your preference for cards you don't already have
• using your full buying power, never buying a Silver for $4 when there are cards out there that cost $4, etc

Instead of intentionally making the bot buy too many terminals I prefer my last item, which (among other things) can lead to the bot buying too many terminals just because it happened to get money that suggested to it to buy those cards.

[Holger]

Some other ideas that would tend to mirror low-to-mid-level play:

- Overvalue cantrips
[...]

The bots already do this, unfortunately; they love to piledrive Spies and Schemes, and Menageries on practically every board...

- Always gaining cards even if they are no use - bots already do this

They do always gain Ruins and some (all?) Action cards with extra buys, but  I don't think they usually gain Coppers (or Curses ) with extra buys. (Which is doubly absurd, since Coppers are usually better than Ruins...)

I suggest making small random adjustments for the preferences of the different cards at the beginning of the game, and then stick to those values for the whole game.

That leads to variation on what errors they make, making for different weaknesses to find how to beat for the player.

Some other ideas come from trying to replicate mistakes that many beginners make:

• not ending the game with a win (preferring another Province instead) -- typical mistake, and instructional
• ending the game, even though you are losing (but that is not as good/interesting to copy)
• increase your preference for cards you don't already have
• using your full buying power, never buying a Silver for $4 when there are cards out there that cost $4, etc

Instead of intentionally making the bot buy too many terminals I prefer my last item, which (among other things) can lead to the bot buying too many terminals just because it happened to get money that suggested to it to buy those cards.

These are all good suggestions IMO; the small random adjustments only work if the bot's buying decision is based on comparing numerical values, which may or may not be the case. Or you could randomly switch between the different (non-Serf) bots' strategies; changing plans mid-game tends to worsen your winning chances...

You might also work on improving Serf Bot instead - e.g. tell it not to always split money evenly on several buys, and not to buy Copper in the absence of Gardens or other Copper enablers.
But suggestions would be easier if we knew the different bots' existing strategies...

[hsiale]

You might also work on improving Serf Bot instead - e.g. tell it not to always split money evenly on several buys, and not to buy Copper in the absence of Gardens or other Copper enablers

I don't think it's good to improve Serf Bot. It is supposed to be weak. I'd prefer spreading the bots' skill more evenly - so that Serf Bot is still very weak, but there are bots with 1500ish Pro rating and also some with 4500ish (provided Goko can code a bot this good).

[Holger]

I don't think it's good to improve Serf Bot. It is supposed to be weak. I'd prefer spreading the bots' skill more evenly - so that Serf Bot is still very weak, but there are bots with 1500ish Pro rating and also some with 4500ish (provided Goko can code a bot this good).

I agree; what I meant was adding an "improved Serf Bot" as an intermediate (e.g. 1500ish) skill level, while keeping the current Serf Bot for beginners. This might be easier to do than intentionally worsening the better bots (whose strategies we don't really know), as was suggested before.

[Polk5440]

"Degrading" bots:

 -- I really like not following PPR.
 -- Too many terminals is also good, but that presumes the bots are good at getting the "right" number in the first place.
 -- Waiting too long to buy green.
 -- Greening too soon.
 -- If they have generic "engine" code, force the bot to always go engine, if able. The bots are obviously better at big money + Smithy. Otherwise, have the bot go money + one copy of one kingdom card (not optimal big money + X).

Or, reduce the decisions the bots make:
 -- Always go for simple power strategies if they are on the board (e.g. Rebuild + X, Hunting Party + X, IGG into Duchies, Cultist only, Double Witch, buy all the Minions, Governor only, Hermit-Market Square, Tournament then Followers).
 -- Always buy power attacks if they are on the board and not listed as a power strategy (Sea Hag, Militia, Mountebank, Ghost Ship, Goons, Ambassador, etc.)
 -- Always buy some trashing if it is on the board (even if going for big money). I don't think the amusement of seeing Rats eat up Lord Bottington's deck should be denied to anyone.
 -- Never buy "weak" cards.

... wait, would this improve the bots

---

I actually would be mildly amused in seeing how TheExpressicist's psuedo-code version of "beginner's Dominion" would compare to the bots....

[Holger]

Or, reduce the decisions the bots make:
 -- Always go for simple power strategies if they are on the board (e.g. Rebuild + X, Hunting Party + X, IGG into Duchies, Cultist only, Double Witch, buy all the Minions, Governor only, Hermit-Market Square, Tournament then Followers).
 [...]
... wait, would this improve the bots

Unfortunately, it would, at least for Rebuild - the bots never go for Rebuild-BM, AFAIK...

Has someone investigated whether SerfBot even beats pure BM? Otherwise adding a pure BM bot would make sense to let every newbie get over the "silver test"   

[hsiale]

Has someone investigated whether SerfBot even beats pure BM? Otherwise adding a pure BM bot would make sense to let every newbie get over the "silver test"   

Tried 4 games. Won 3. The one I lost: http://www.gokosalvager.com/static/logprettifier.html?/20140320/log.50893139a2e67cff211cd7e5.1395358223710.txt - Colonies and Thief, Serf Bot buys Thieves turn 2, 4 and 10 stealing lots of money from me during the game. I lost by one point.

One of the boards I lost was also with Colonies, it took me 35 turns to get 8 Colonies, 1 Province and 1 Duchy with pure BM. The bot built some strange Golem/Beggar/Talisman/Spice Merchant engine which was quite good at gaining Coppers and Potions and probably trashing them.

[yed]

Or, reduce the decisions the bots make:
 -- Always go for simple power strategies if they are on the board (e.g. Rebuild + X, Hunting Party + X, IGG into Duchies, Cultist only, Double Witch, buy all the Minions, Governor only, Hermit-Market Square, Tournament then Followers).
 [...]
... wait, would this improve the bots

Unfortunately, it would, at least for Rebuild - the bots never go for Rebuild-BM, AFAIK...

Has someone investigated whether SerfBot even beats pure BM? Otherwise adding a pure BM bot would make sense to let every newbie get over the "silver test"   

I have tried that 3 times. Pure big money beats serf bot. So here you go pure big money bot.

[yed]

Challenge: find a board where pure big money is optimal strategy.

Ok it is easy: 10 villages kingdom

[amalloy]

One of the boards I lost was also with Colonies, it took me 35 turns to get 8 Colonies, 1 Province and 1 Duchy with pure BM. The bot built some strange Golem/Beggar/Talisman/Spice Merchant engine which was quite good at gaining Coppers and Potions and probably trashing them.

I just had to see what this looked like. I think my favorite is turn 34:

Serf Bot   plays Golem
Serf Bot   reveals Beggar, Copper, Golem, Forager
Serf Bot   discards: Copper, Golem
Serf Bot   plays Beggar
Serf Bot   gains Copper
Serf Bot   gains Copper
Serf Bot   gains Copper
Serf Bot   plays Forager
Serf Bot   trashes Beggar
Serf Bot   plays Workshop
Serf Bot   gains Potion
Serf Bot   plays Talisman
Serf Bot   plays Talisman
Serf Bot   plays 3 Copper
Serf Bot   buys Potion
Serf Bot   gains Potion
Serf Bot   gains Potion
Serf Bot   gains Potion
Serf Bot   buys Beggar
Serf Bot   gains Beggar
Serf Bot   gains Beggar
Serf Bot   gains Beggar
Serf Bot   draws Potion, Talisman, Copper, Copper, Forager

Sometimes you just need four potions in a single turn, you know? And it's tough to decide whether you want fewer beggars or lots more beggars!

[SirPeebles]

Challenge: find a board where pure big money is optimal strategy.

Ok it is easy: 10 villages kingdom

At $5, I'd take Festival over Silver.  BM Festival?

[yed]

Challenge: find a board where pure big money is optimal strategy.

Ok it is easy: 10 villages kingdom

At $5, I'd take Festival over Silver.  BM Festival?

Village, walled, Border, Fortress, Wand Minst, Hamlet, native, workers, university, fishing

Ony useful is workers and fishing but silver is better.
Ok technically hamlet is better than nothing on $2.

[Awaclus]

If I'm correct, these are the cards that you can have in a kingdom without making purchasing anything other than basic cards a move worth considering (even if you would say no in almost every situation):

Village
Fortress
Walled Village
Scheme
Throne Room
Band of Misfits
Border Village
(Wandering Minstrel)
(University)

Wandering Minstrel and University aren't technically strictly worse than Silver, because one cycles at random and the other empties piles and both are effects that can be desirable, but in practice emptying piles with University is so slow that you won't ever get to do that in a BM game and the cycling helps you so little that it's not even worth mentioning. In order to complete the kingdom, you need one card that you actually have to consider getting instead of Silver. I'm not sure if Farming Village, Scout or Transmute is the worst possible alternative for that slot.

[Watno]

Border Village + Duchy is better than just Duchy for sure, becaquse you can choose not to play Border Village to avoid a reshuffle.

[JW]

If I'm correct, these are the cards that you can have in a kingdom without making purchasing anything other than basic cards a move worth considering (even if you would say no in almost every situation):

Village
Fortress
Walled Village
Scheme
Throne Room
Band of Misfits
Border Village
(Wandering Minstrel)
(University)

Why not Rats?

[Awaclus]

If I'm correct, these are the cards that you can have in a kingdom without making purchasing anything other than basic cards a move worth considering (even if you would say no in almost every situation):

Village
Fortress
Walled Village
Scheme
Throne Room
Band of Misfits
Border Village
(Wandering Minstrel)
(University)

Why not Rats?

Because it comboes with Fortress.

Border Village + Duchy is better than just Duchy for sure, becaquse you can choose not to play Border Village to avoid a reshuffle.

True. I guess you need two cards then.

[amalloy]

Why not Rats?

Because it comboes with Fortress.

Not much of a combo...it turns Rats from "cantrip: turn a good card into a bad card" to "cantrip: gain a bad card". Rats is strictly worse than nothing at all, let alone Silver.

[Awaclus]

Why not Rats?

Because it comboes with Fortress.

Not much of a combo...it turns Rats from "cantrip: turn a good card into a bad card" to "cantrip: gain a bad card". Rats is strictly worse than nothing at all, let alone Silver.

It empties a pile very fast.

[JW]

Not much of a combo...it turns Rats from "cantrip: turn a good card into a bad card" to "cantrip: gain a bad card". Rats is strictly worse than nothing at all, let alone Silver.

It empties a pile very fast.

The only reason I see to empty the Rats pile in this kingdom is that it will end the game with you ahead, and the only (non-Province) piles I see running out besides Rats are Duchy and Estate.  Additionally, Rats-Fortress only empties the Rats pile fast if you can draw your deck and I can't see how you draw your deck if Duchy/Estates are gone and you have a lead. So I don't see how Rats could be used as part of a strategy improving on Big Money even slightly.

[Holger]

Why not Rats?

Because it comboes with Fortress.

Not much of a combo...it turns Rats from "cantrip: turn a good card into a bad card" to "cantrip: gain a bad card". Rats is strictly worse than nothing at all, let alone Silver.

It empties a pile very fast.

Not really - it takes >4 shuffles to empty the Rats pile after you bought one (if you don't buy more than one), even if you play it every time you draw it.

[Stealth Tomato]

Wand Minst

This confused me substantially, I was wondering what the hell sort of wand a wand minst was.

[Drab Emordnilap]

Don't forget King's Court.

[Awaclus]

Don't forget King's Court.

King's Courting a village gives you one extra card.

[Holger]

Has someone investigated whether SerfBot even beats pure BM? Otherwise adding a pure BM bot would make sense to let every newbie get over the "silver test"   

Tried 4 games. Won 3. The one I lost: http://www.gokosalvager.com/static/logprettifier.html?/20140320/log.50893139a2e67cff211cd7e5.1395358223710.txt - Colonies and Thief, Serf Bot buys Thieves turn 2, 4 and 10 stealing lots of money from me during the game. I lost by one point.

One of the boards I lost was also with Colonies, it took me 35 turns to get 8 Colonies, 1 Province and 1 Duchy with pure BM. The bot built some strange Golem/Beggar/Talisman/Spice Merchant engine which was quite good at gaining Coppers and Potions and probably trashing them.

Has someone investigated whether SerfBot even beats pure BM? Otherwise adding a pure BM bot would make sense to let every newbie get over the "silver test"   

I have tried that 3 times. Pure big money beats serf bot. So here you go pure big money bot.

Thanks for testing this. It's ... surprising that they intentionally (?) made SerfBot worse than the simplest possible strategy. How much rating difference does a 75%+ win rate correspond to, I wonder? Serf Bot is level -37 (!) on Isotropish, after all...

[Holger]

Lord Bottington is the default, programmed as best as they can.

FWIW, Lord Bottington is now the second-worst of the non-serf bots on Isotropish (at level 6; Warlord is level 1; the best is Conqueror at 15), though the order changes quite often...

[SirD]

Did they already change the quality of the bots? Boy is Conqueror playing badly. I always thought he'd play quite decently:

http://www.gokosalvager.com/static/logprettifier.html?/20140421/log.516d6636e4b082c74d7bdf6f.1398121207859.txt

[amalloy]

Did they already change the quality of the bots? Boy is Conqueror playing badly. I always thought he'd play quite decently:

http://www.gokosalvager.com/static/logprettifier.html?/20140421/log.516d6636e4b082c74d7bdf6f.1398121207859.txt

Looks pretty normal. I was expecting worse when I saw Warehouse in the kingdom, really: some of the bots love that card, and will pile Warehouses then transition into a "play five Warehouses per turn" strategy. But it looks like the Remake/Poor House interaction stymied him: he kept gaining cards he didn't expect with it!

[SirD]

@amalloy Right. I see you have got some experience with the stupid side of the bots.

I mean much good stuff is already said in this thread about how to decrease the bots' skills. I would suggest to take the best bot you build so far (and maybe even improve it!). Then you can additionally implement the following rules:

- Buy too many action cards
- Buy no, or not enough, or too late cards with plus buy
- Underestimate the value of drawing cards (buy not enough or too late)
- Green to late(!)
- Only go for provinces (dont buy duchies, continue building)
- Nearly never buy estates or alt VP

You should be able to adjust the skill level of the bots by the percentage they follow the bad rules instead of the good ones.

But hey, I would really love to see some improved bots also

[florrat]

This bot tried out a new strategy, Golem Big Money:

Banker Bot cards: 13 Silver , 1 Golem , 7 Copper , 3 Gold , 2 Masterpiece , 1 Potion , 3 Estate , 3 Province , 2 Duchy
Banker Bot total victory points: 27
Banker Bot turns: 17

[SCSN]