Dominion Strategy Forum
Dominion => Simulation => Topic started by: ftl on May 11, 2012, 03:55:21 am
-
[edit 5/11: added baron, militia, goons, and a comment about bot optimization]
So, this is something I've thought about for a while, and slowly built up. I know a lot of popular BM+X options - Smithy, Envoy, DoubleJack, Wharf, etc. But I didn't really have a handle on how good they were relative to each other - if I have a board with Envoy or Vault, what do I pick up? What about Rabble or Margrave?
So, I tried to answer this question using the sims! The goal was to produce a linear ranking, worst to best, of common BM+X options. I included all of the above and some others. Didn't include cursers, since nearly all of the time those would necessitate some sort of response, leaving simulation results pretty useless. I didn't include any of the lab variants (Stables, Lab, Hunting Party) since they live or die by what terminal you add. I used Geronimoo's simulator. I mostly used existing optimized bots when I could find them, either ones labeled in the simulator as 'optimized' or ones from the 'Optimizing BM+X' thread. For some of the terminals I had to roll my own, I mostly used the same structure as the BM+Smithy bot but with the +cards card instead of Smithy and with the numbers tweaked.
I'm presenting a table, with the results ordered roughly in order from worst to best; I say roughly, because since I included some attacks or interactive cards, there are some rock-paper-scissors results, and some mis-ranked ones. Table attached as a spreadsheet! Each row/column box has what percentage of the games the row-card wins. Colored blue if the row-card won and orange if the column-card won. Linear rankings copy/pasted here:
Bureaucrat
Moat
Explorer
Militia
Baron
Library
Oracle
Council Room
Merchant Ship
Smithy
Envoy
Rabble
Embassy
Masquerade
Vault
Torturer
Noble Brigand
Margrave
Monument
Courtyard
Goons
Ghost Ship
Jack of all Trades
Wharf
Comments:
Bots weren't optimized against each other, so any BM+X card that is somewhat interactive should really be ranked slightly worse, since an opponent can respond. So Oracle, Council Room, Rabble, Vault, Torturer, Noble Brigand, Margrave, and Ghost ship are probably overrated to various degrees. Oracle is probably underrated, since proper Oracle play probably involves picking whether to draw or discard the top 2 based on what's left in your deck and what you want to buy.
One thing that surprised me is that Noble Brigand is, well, really good when playing against a pure or mostly pure BM+X. Really really good. This may be partly simulator error, perhaps because the sims don't re-buy silver and gold once they're greening, or for some other reason. But still. It gets crushed by the top-tier of Wharf and Ghost Ship, but beats even stuff like Masq or Courtyard. Not quite sure what the right way to incorporate NB into a BM deck is, but I bet that if done right it can give a huge edge.
Merchant Ship is actually a pretty good BM+X. On par with Smithy. It benefits a lot more from optimization than some of the ones around it, since the larger number of Ships and the duration effect allow for earlier greening.
Jack is heavily, heavily overrated because of how well it does against some of the discard-attack BM+X options - specifically, Goons and Ghost Ship.
Since I picked up optimized bots from various sources, they may be at different levels of optimization. None of them had PPR, but for different BM+X options, stuff like that may have a different effect. So, as with all sim results, treat this with the appropriate caution.
So, what do you guys think? Any more comments? Terminals I should add to the table? Ways to make this more interesting or useful?
-
I am shocked to see Masquerade this low. The simulators have to be missing something, playing it non-optimally, I don't know.
Embassy's another one I'd have expected to do better.
-
Perhaps part of Masq's power just isn't being used the way I've set it up here - the fact that it can plow right through cursing attacks, unlike nearly everything else.
-
Simulator comments:
-Oracle is not played well. This card is a lot stronger (if played by a good player tracking his deck and the opponent's).
-Embassy is really strong, but don't forget that giving free Silvers away is helping the opponent who is also going Big Money a lot!
-Masquerade is hurt severely by discard effects. The sim plays the card quite well, eg, not trashing stuff if it would stop you from buying a card and not passing Estates in the end game.
-Torturer's play rules will cause the defender to ALWAYS discard 2 unless he's got a trasher in hand (like Masquerade). So it should probably be below Vault and Masquerade.
-Noble Brigand is just really really strong in Big Money. The bots will get stuck often in the greening stage where they aren't buying money anymore and the Brigand has stolen most of its economy, but trying to rebuild the economy is not going to help much if the Brigand just keeps stealing it while he's buying Duchies.
-Ghost Ship's play rules will always cause the defender to put back his worst cards which is clearly not optimal. So Ghost Ship is likely a little less strong (but I'd still rank it at n°2)
-Why is Ghost Ship included and not Militia?? Baron is also strong in Big Money (produces fast Golds).
I don't think people would have realized the power of Courtyard, Jack or Noble Brigand without the simulators. They look so harmless...
-
Militia is not included... I don't know, because I'm forgetful.
Baron? Never really considered that as something to put in BM+X... does it work if I just put in a single Baron to open with? Is that the right way to play that?
-
Well, double Baron certainly looks like overkill. You just use it to hit it with Estate and 2 or 3 Coppers for an early Gold.
-
A second Baron is actually fine in the early game. Also, you want to get Estates a little sooner.
-
Haggler?
Though the lack of 4 will hurt it a lot.
-
This is great work. You should think about writing an article for the blog when everything is done.
For me there are two astonishing things on this list:
Courtyard: Of course it is known that Courtyard is a great BM+ card. But i didn't think it is better then Envoy or Masquerade. Why do we see BM+Envoy and BM+Masq so often and never BM+Courtyard?
Bureaucrat: It is known to be a bad card. Since I've been watching WW's Dominion videos I already thought it is underrated, because he buys it more often than I would do. But looking on this list it seems to be even overrated. It's comparable with Moat and worse than everything else.
I think it would be interesting to compare all strategies depending on the starting hands. I would expect $5 cards to be stronger on 5/2 split and $4 cards on 4/3 split. If you have 5/2 it may be better to pick a usually inferior $5 card to start with.
-
Bureaucrat isn't intended to be a Big Money enabler. Rather, it's very good in an alternate VP rush like Silk Road, Gardens or Dukes. So it's neither underrated, nor overrated, it just needs the right context.
And I don't know about an article on the main blog. First, it's quite rare that you're going to play a pure Big Money strategy so this investigation only has value as a guideline. Second, people are extremely skeptical about simulator analysis (sometimes rightly so, sometimes not).
-
This is great work. You should think about writing an article for the blog when everything is done.
For me there are two astonishing things on this list:
Courtyard: Of course it is known that Courtyard is a great BM+ card. But i didn't think it is better then Envoy or Masquerade. Why do we see BM+Envoy and BM+Masq so often and never BM+Courtyard?
Bureaucrat: It is known to be a bad card. Since I've been watching WW's Dominion videos I already thought it is underrated, because he buys it more often than I would do. But looking on this list it seems to be even overrated. It's comparable with Moat and worse than everything else.
I think it would be interesting to compare all strategies depending on the starting hands. I would expect $5 cards to be stronger on 5/2 split and $4 cards on 4/3 split. If you have 5/2 it may be better to pick a usually inferior $5 card to start with.
Bureaucrat is best in alt-VP games: Duke, Silk Road, Gardens. (Ed: Ninja'd by Geronimoo.)
As for why we don't see BM-Courtyard more often, probably because the top-decking ability of Courtyard is temptation to go in a less strictly BM direction, since you can carry over extra actions to next turn.
-
We see masquerade played a lot with treasures since it nullifies a number of other strategies. In some kingdoms it might help a deck withstanding attacks. In other kingdoms it might disrupt a high quality deck. Jack-of-all-trades often changes a kingdom in the same way.
-
Could you give information about what optimized play method is used in these simulations? I assume, for instance, you're just buying one or possibly two Smithies, but all the Vaults you can get. But I'm honestly not sure for Courtyard. Even a short-hand description of the optimized play would be helpful.
-
Thanks for making this list! You should probably also list the exceptions (i.e. next to a card, list the lower ranked cards that beat it) I think Ghost Ship loses to Jack and Oracle (not in sim because Oracle is played bad), and Jack and Torturer lose to Masq. Any others? Also, if you're including Discard attacks, you might want to include Goons, which should beat everything but Wharf and Ghost Ship.
-
There are plenty more terminals that you could add: mandarin, harvest, island, bridge, chancellor, mine, bishop, horse traders, nomad camp, watchtower, tribute, and even cards like ironworks. None of them would shatter your existing list though though and some of them need 'optimised' scripts in the simulator.
-
Chapel: would be nice to see if it can make it to a decent spot on the list or stew near the bottom, since it's such a bad BM-enabler.
-
There are plenty more terminals that you could add: mandarin, harvest, island, bridge, chancellor, mine, bishop, horse traders, nomad camp, watchtower, tribute, and even cards like ironworks. None of them would shatter your existing list though though and some of them need 'optimised' scripts in the simulator.
The problem is that to deal with these lesser BM cards, it probably takes a lot of optimization to get results that aren't really useful for anything.
For example, the current HT bot is about even with straight BM. I'm sure it can be made better by delaying the HT, since there is no reason to want to get to $5 on the first shuffle, and the +buy doesn't come into play until it's time to by Estates. If you add the condition that you need at least 9 treasures before buying HT, it gets about a percent and half better, and you can probably do more, but do you actually want to spend time optimizing that?
Bishop and Island might be nice though, to answer the question of how fast you have to be to ignore them.
-
Could you give information about what optimized play method is used in these simulations? I assume, for instance, you're just buying one or possibly two Smithies, but all the Vaults you can get. But I'm honestly not sure for Courtyard. Even a short-hand description of the optimized play would be helpful.
It's the optimized bots in the sim, which... he didn't design, by-and-large. I designed several. It's an involved process, which I might eventually do, to describe them all, but... it's hard to motivate myself for this. Especially since they play pretty badly in the endings - it's very hard to get all the variables right.
Further, most all of these aren't optimized, and some of them quite significantly not so. I will try to post my compendium of more optimized things in a single file in the next few days. I think most notably, embassy and wharf are underrated here. Yeah, Wharf is #1 and still underrated(!). Most all of the bots can be aided by adding a couple simple endgame things, like buying estates sooner on exactly 2, and sometimes with less than 5. PPR helps too. And switching 'has at least one gold' to 'at least X amount of money' in most cases. Goons should be included, it beats everything but Wharf, GS, Jack. Haggler should be included too, but... someone needs to optimize it. I thought I'd done that at some point, but apparently not...
-
Thanks for making this list! You should probably also list the exceptions (i.e. next to a card, list the lower ranked cards that beat it) I think Ghost Ship loses to Jack and Oracle (not in sim because Oracle is played bad), and Jack and Torturer lose to Masq. Any others? Also, if you're including Discard attacks, you might want to include Goons, which should beat everything but Wharf and Ghost Ship.
Ghost ship loses to oracle even in the sim.
-
Thanks for making this list! You should probably also list the exceptions (i.e. next to a card, list the lower ranked cards that beat it) I think Ghost Ship loses to Jack and Oracle (not in sim because Oracle is played bad), and Jack and Torturer lose to Masq. Any others? Also, if you're including Discard attacks, you might want to include Goons, which should beat everything but Wharf and Ghost Ship.
Ghost ship loses to oracle even in the sim.
Really! I must have a bad oracle bot, then, Oracle lost pretty badly to Ghost Ship when I tried it...
-
A second Baron is actually fine in the early game. Also, you want to get Estates a little sooner.
Thanks!
With those comments, Baron actually looks reasonable, pretty close to stuff like Library and Council Room.
Despite being, now, the only one to lose to the out-of-place Bureaucrat! Knew B-crat's attack was good for something... stifling Barons!
Could you give information about what optimized play method is used in these simulations? I assume, for instance, you're just buying one or possibly two Smithies, but all the Vaults you can get. But I'm honestly not sure for Courtyard. Even a short-hand description of the optimized play would be helpful.
The bot opens silver/courtyard. It has the standard rules for greening (buy provinces once you have a gold, buy duchies/estates when there are a 4/2 provinces left, respectively, buy duchies over silver but not over gold when there are 6 provinces left.) It never buys courtyard over gold, but buys courtyard over silver when it has fewer than one courtyard for every eight treasures in the deck. I guess this would roughly mean a second courtyard sometime between turns 4 and 6, and probably a third one around turn 10-12.
I didn't write it, just summarizing.
This is great work. You should think about writing an article for the blog when everything is done.
Thanks! :)
Not sure it needs to be an article in its own right, for the reasons Geronimoo posted; but, I'm glad it draws attention to some underappreciated BM cards, that's why I made it.
For me there are two astonishing things on this list:
Courtyard: Of course it is known that Courtyard is a great BM+ card. But i didn't think it is better then Envoy or Masquerade. Why do we see BM+Envoy and BM+Masq so often and never BM+Courtyard?
Of the things the simulator tells us right, I think this is one of them.
Bureaucrat: It is known to be a bad card. Since I've been watching WW's Dominion videos I already thought it is underrated, because he buys it more often than I would do. But looking on this list it seems to be even overrated. It's comparable with Moat and worse than everything else.
Yeah, on putting it in this list, I realized it probably doesn't belong here at all; I guess you should never go Bureaucrat if you're aiming for a standard Silver->Gold->Province game. I didn't realize that until making the list, though, I didn't realize it would even lose to, like, Moat.
I think it would be interesting to compare all strategies depending on the starting hands. I would expect $5 cards to be stronger on 5/2 split and $4 cards on 4/3 split. If you have 5/2 it may be better to pick a usually inferior $5 card to start with.
Yeah, probably. Obviously, a big reason why the $5 cards lose to some of the $4 cards is that they spend a few turns buying Silver before getting the enabler.
I'm not sure I want to go back and reenter all that into the sims, though!
Thanks for making this list! You should probably also list the exceptions (i.e. next to a card, list the lower ranked cards that beat it) I think Ghost Ship loses to Jack and Oracle (not in sim because Oracle is played bad), and Jack and Torturer lose to Masq. Any others? Also, if you're including Discard attacks, you might want to include Goons, which should beat everything but Wharf and Ghost Ship.
Added goons. It loses to Wharf, Ghost Ship, and DoubleJack (?!). I'm not sure it really fits, though - when going for Goons, I'd probably open with one of the cheap BM options if available.
Also, I realized that the way I do the rankings, adding more cards changes the relative rankings - adding goons pushes Jack above Courtyard, because Jack beats Goons but Courtyard does not. Which then also pushes it above Goons and Ghost Ship, since it beats them both, which instantly makes Jack heavily overrated at #2...
Oh well. That's why I made a chart, originally I just had a linear ranking but that was susceptible to stuff like that, the chart keeps that info around.
There are plenty more terminals that you could add: mandarin, harvest, island, bridge, chancellor, mine, bishop, horse traders, nomad camp, watchtower, tribute, and even cards like ironworks. None of them would shatter your existing list though though and some of them need 'optimised' scripts in the simulator.
Yeah, I don't think I want to add all of those.
Tribute and Bishop are good ideas though.
Probably not Island... Island+BM isn't something that's relevant to think about, you'd potentially go Island into some other BM+X, or buy islands over estates at the end of a BM+X...
-
The bot opens silver/courtyard. It has the standard rules for greening (buy provinces once you have a gold, buy duchies/estates when there are a 4/2 provinces left, respectively, buy duchies over silver but not over gold when there are 6 provinces left.) It never buys courtyard over gold, but buys courtyard over silver when it has fewer than one courtyard for every eight treasures in the deck. I guess this would roughly mean a second courtyard sometime between turns 4 and 6, and probably a third one around turn 10-12.
I didn't write it, just summarizing.
I thought it was supposed to be duchies at 5. Or is it 4 with CY, 5 in general? Or am I just mixing things up completely?
-
I thought it was supposed to be duchies at 5. Or is it 4 with CY, 5 in general? Or am I just mixing things up completely?
Not sure. The courtyard bot has 4, the ghost ship one has 5, the BMU one has 4. I thought 4 was the general answer - but at the end of the day, it's not that big of a deal... ideally, you'd also take into account whether you're about to reshuffle and what your opponent is doing...
-
Bureaucrat isn't intended to be a Big Money enabler. Rather, it's very good in an alternate VP rush like Silk Road, Gardens or Dukes. So it's neither underrated, nor overrated, it just needs the right context.
Point taken. I'll keep that in mind the next time I see someone playing Bureaucrat.
As for why we don't see BM-Courtyard more often, probably because the top-decking ability of Courtyard is temptation to go in a less strictly BM direction, since you can carry over extra actions to next turn.
I think you're right. But the data suggests, it would be better to overcome that temptation. (In decks where BM is the dominant strategy.)
On the other hand I'm almost sure I've more than one time bought a Smithy to go for a BM strategy and ignored the Courtyard in the same kingdom. I won't do that again. :D
We see masquerade played a lot with treasures since it nullifies a number of other strategies. In some kingdoms it might help a deck withstanding attacks. In other kingdoms it might disrupt a high quality deck. Jack-of-all-trades often changes a kingdom in the same way.
True. That's why i didn't mention Jack. Masq has this effect in some situations, too. So it was a bad example by me.
-
Probably not Island... Island+BM isn't something that's relevant to think about, you'd potentially go Island into some other BM+X, or buy islands over estates at the end of a BM+X...
True but islands on their own are surprisingly good.
-
Is there a way to get a separate ranking for non-attacks when they compete with non-attacks?
-
Is there a way to get a separate ranking for non-attacks when they compete with non-attacks?
Why do you want a separate ranking?
-
I attempted to replicate some of what I thought were the more interesting results, and my results differ from yours in a few cases, so I was hoping you could post the bots you used.
I have Margrave beating Library (~49-42), and I have Vault beating Noble Brigand (~54-42).
I didn't put any effort into optimizing VP card buys, so that might be the difference (I could see Library maybe beating Margrave via early VP buys, counting on the filtering from Margrave to get them out of the way). I just took the default bots with Margrave buying 2, Library 3, Vault unlimited (but 1 above Gold), and Noble Brigand 4. I also tried messing around with limiting number of NBs based on amount of treasure, but couldn't really find something that beat Vault.
-
Hm, I think you're right, I'll go back and check.
-
This might help you out:
For those who want it, here's an XML of all the latest single-card optimizations I have.
Edit: Click on the link above the quote to get to my post. The file is attached there.
-
So the Noble BrigandWW bot beats the current Vault bot, but loses if the Vault player buys 1 Vault ahead of Gold.
-
You can just about toss the torturer results out the window, because the simulator plays TERRIBLY against it. Oracle is underrated, rather than overrated, because it plays the gain part worse than the attack. And I don't think noble brigand is much overrated at all, in these strict matchups. Obviously, in real games, if there's some good virtual money around, or if you can pick up venture or harem or something... but in the head-to-head, you're dependent on that silver/gold.
-
So now we have Beggar, Catacombs, Count, Cultist, Hunting Grounds, Pillage, Scavenger, and Storeroom.
There's also Death Cart, which might not count since you also get Ruins? We've also got Rogue and Knights, but those seem like they'd be a little weird to simulate.
-
Counterfeit is worth taking a look at too.
-
I don't really think Counterfeit is a good BM+X card. It's somewhere between a Moneylender and a Salvager that can't trash non-Treasure cards, and neither are particularly great in BM+X.