Print

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

[markusin]

I remember a game where I beat Cultist with PStone+Storeroom (which is known to be a strong combo, made even stronger by all those Ruins) - it was close though, and I have no idea what its win rate is like against it.

Even then, you had to pick a strategy that works well specifically against Cultist -> you didn't ignore Cultist.

Doesn't "ignoring" a card usually involve addressing it's power in some way. Ignoring Witch might mean you get an extra trasher.

And anyway, PS-Storeroom is good enough without Cultist, just like Workshop/Gardens can work without an oppoaing looter.

This argument holds for Counting House in Mountebank Games, where the CH strategy relies on your opponents Mountebank to even be decent.

[Awaclus]

I remember a game where I beat Cultist with PStone+Storeroom (which is known to be a strong combo, made even stronger by all those Ruins) - it was close though, and I have no idea what its win rate is like against it.

Even then, you had to pick a strategy that works well specifically against Cultist -> you didn't ignore Cultist.

Doesn't "ignoring" a card usually involve addressing it's power in some way. Ignoring Witch might mean you get an extra trasher.

And anyway, PS-Storeroom is good enough without Cultist, just like Workshop/Gardens can work without an oppoaing looter.

This argument holds for Counting House in Mountebank Games, where the CH strategy relies on your opponents Mountebank to even be decent.

I thought that ignoring a card involves treating the kingdom as if the card wasn't there. In Hermit/Market Square games you can probably ignore Cultist.

[zporiri]

reason why i love dominion #476: the weakest $5 card (counting house) is a great addition to a deck when the opponent has the strongest $5 card (mountebank)

[SCSN]

SCSN's article (or was it AI's article)

It was team-work, though he did the writing and ended up posting it.

[GendoIkari]

reason why i love dominion #476: the weakest $5 card (counting house) is a great addition to a deck when the opponent has the strongest $5 card (mountebank)

Stratego did it first.

[Holger]

But e.g. Rebuild-Militia beats Mountebank-Rebuild.

that comparison doesn't make sense. it's never mountebank or militia. you have to choose between militia and silver, because both are t1/2 options, and between rebuild and mount likewise. if all three are present, you open militia/silver, get one early mount and then go into rebuild

It's not that simple because Militia and Mountebank are both terminals which you don't want to collide (and of course there's 2/5 starts to consider). So you might not want to gain both even if either one helps Rebuild, or you may prefer a second Militia to sacrificing a $5 buy on a (potentially colliding) Mountebank.

yes, but the question is not "do i get militia or not", you're always going to get militia, it's just too important. the question is: "do i still get mountebank", and i'm pretty sure that you do. 2 terminals in a deck without draw or trashing isn't the end of the world, and if your opponent doesn't get a mount on his own, you can always change plans, ignore rebuild and play a BM strategy. my game sense tells me that BM-Mount > Rebuild >? Rebuild/Mount > BM-Mount, if all strategies use militia. mb i'm wrong here, but I'll have a very hard time believing that militia is enough of a reason not to get mount

second militia is a nombo, it's important early on. 5/2 is a different story, but I'm pretty sure you go mount there.

I've tried to simulate some of the combinations, and it seems that indeed you're right, Rebuild/1Mount/1Militia beats Rebuild/Militia. But (Rebuild+) 2 Militias is better than 1 in the absence of Mount., it's not  a nombo. (In BM games, Militia should hurt more in mid-game than in the early game, since the number of dead cards in hand decreases.)

my game sense tells me that BM-Mount > Rebuild >? Rebuild/Mount > BM-Mount, if all strategies use militia.

Huh? You previously said that "Rebuild < Rebuild/Mount", which seems to be true. And Rebuild-Militia beats Mount./Militia decisively, since Militia helps Rebuild more than Mount. due to being terminal. It now seems to me that the three-card combo beats all three two-card combos, which beat the respective 1-card strategies.

[silverspawn]

i mostly meant that, even if rebuild/militia beats rebuild/mount/militia in straight simulation, whenever you do go mount and your opponent didn't, you can still switch into a BM plan, because I do think mount/militia beats rebuild/militia (hence the "?")

i also don't really trust simulation... not because of the concept, but people seem to place way too much trust into simulations which do significant strategic mistakes

[SCSN]

i mostly meant that, even if rebuild/militia beats rebuild/mount/militia in straight simulation, whenever you do go mount and your opponent didn't, you can still switch into a BM plan, because I do think mount/militia beats rebuild/militia (hence the "?")

i also don't really trust simulation... not because of the concept, but people seem to place way too much trust into simulations which do significant strategic mistakes

I assume you mean tactical rather than strategic, as with scripts the strategies are given.

While I can say nothing about the Mountebanks, being the one who spent countless of hours optimizing the Rebuild scripts I can assure you they do not make any significant tactical mistakes. And if you believe to have found any, please let me know so I can eliminate them.

In fact, I believe the scripts play better than any human could, at least over the course of several consecutive games, during which lapses of attention inevitably result in suboptimal deck-tracking that would prevent you from making plays like this:

== RebuildMirror's turn 13 ==
RebuildMirror plays Rebuild.
...RebuildMirror names Estate.
...RebuildMirror reveals [Estate, Duchy].
...RebuildMirror discards [Estate].
...RebuildMirror trashes RebuildMirror's Duchy.
...RebuildMirror gains Province.
RebuildMirror plays Rebuild.
...RebuildMirror names Province.
(RebuildMirror shuffles.)
...RebuildMirror reveals [Copper, Copper, Copper, Province, Duchy].
...RebuildMirror discards [Copper, Copper, Copper, Province].
...RebuildMirror trashes RebuildMirror's Duchy.
...RebuildMirror gains Province.
Coins: 3, Potions: 0, Buys: 1
RebuildMirror buys Estate.
RebuildMirror draws 5 cards: [Copper, Copper, Silver, Silver, Silver]

[silverspawn]

i mostly meant that, even if rebuild/militia beats rebuild/mount/militia in straight simulation, whenever you do go mount and your opponent didn't, you can still switch into a BM plan, because I do think mount/militia beats rebuild/militia (hence the "?")

i also don't really trust simulation... not because of the concept, but people seem to place way too much trust into simulations which do significant strategic mistakes

I assume you mean tactical rather than strategic, as with scripts the strategies are given.

While I can say nothing about the Mountebanks, being the one who spent countless of hours optimizing the Rebuild scripts I can assure you they do not make any significant tactical mistakes. And if you believe to have found any, please let me know so I can eliminate them.

In fact, I believe the scripts play better than any human could, at least over the course of several consecutive games, during which lapses of attention inevitably result in suboptimal deck-tracking that would prevent you from making plays like this:

== RebuildMirror's turn 13 ==
RebuildMirror plays Rebuild.
...RebuildMirror names Estate.
...RebuildMirror reveals [Estate, Duchy].
...RebuildMirror discards [Estate].
...RebuildMirror trashes RebuildMirror's Duchy.
...RebuildMirror gains Province.
RebuildMirror plays Rebuild.
...RebuildMirror names Province.
(RebuildMirror shuffles.)
...RebuildMirror reveals [Copper, Copper, Copper, Province, Duchy].
...RebuildMirror discards [Copper, Copper, Copper, Province].
...RebuildMirror trashes RebuildMirror's Duchy.
...RebuildMirror gains Province.
Coins: 3, Potions: 0, Buys: 1
RebuildMirror buys Estate.
RebuildMirror draws 5 cards: [Copper, Copper, Silver, Silver, Silver]

I didn't want to make the claim that all simulations are bad. I said I don't trust simulations per se, because I've seen a lot of cases where they are just plain wrong. I remember someone saying that treasure map/masterpiece is worse than masterpiece because simulations show it, only to be proven wrong by someone else who made a few simple corrections in the code. I've also seen simulators having a <50% winrate with wharf/KC against simple BM, just to make two examples.

I want to take power away from the deux ex machina argument "simulations show that XX", because that on is own isn't worth much. But that doesn't mean that it's impossible to make good simulations, in fact I do think that it's theoretically possible to program a bot that plays better than any human ever could. It's really just a question of how good the ai is.

And yes, if you define strategy as what is played and tactic as how it's played, I mean tactical

[Mic Qsenoch]

i mostly meant that, even if rebuild/militia beats rebuild/mount/militia in straight simulation, whenever you do go mount and your opponent didn't, you can still switch into a BM plan, because I do think mount/militia beats rebuild/militia (hence the "?")

i also don't really trust simulation... not because of the concept, but people seem to place way too much trust into simulations which do significant strategic mistakes

I assume you mean tactical rather than strategic, as with scripts the strategies are given.

Can't the scripts have significant strategic mistakes? Why can't he be opposed to trusting simulation on the grounds that the programmed strategies aren't very good? Or maybe the simulator doesn't provide the flexibility in how strategies are defined to account for all the things a human might. I honestly don't know the answer to the last one, I've only played with Geronimoo's simulator a bit. Are there any obviously relevant pieces of game data that can't be accounted for easily while making strategies in the simulator? Ok maybe not obviously, are there any subtly relevant pieces of game data missing?

[SCSN]

I didn't want to make the claim that all simulations are bad. I said I don't trust simulations per se, because I've seen a lot of cases where they are just plain wrong. I remember someone saying that treasure map/masterpiece is worse than masterpiece because simulations show it, only to be proven wrong by someone else who made a few simple corrections in the code. I've also seen simulators having a <50% winrate with wharf/KC against simple BM, just to make two examples.

The results of a simulation are as good as the person who programmed it. When someone presents you simulation data and you respond by saying that you don't trust it, you accuse him of being either incompetent or sloppy, an accusation you are socially obliged to back up, e.g. by specifying what you think he did wrong. Saying that you have seen someone else mess up a simulation in the past is not enough.

[SCSN]

i mostly meant that, even if rebuild/militia beats rebuild/mount/militia in straight simulation, whenever you do go mount and your opponent didn't, you can still switch into a BM plan, because I do think mount/militia beats rebuild/militia (hence the "?")

i also don't really trust simulation... not because of the concept, but people seem to place way too much trust into simulations which do significant strategic mistakes

I assume you mean tactical rather than strategic, as with scripts the strategies are given.

Can't the scripts have significant strategic mistakes?

A script is a tactical implementation of a strategy chosen by its programmer. You can say that Advisor-BM is a terrible strategy, but you can't say that an Abvisor-BM script is making any strategic mistakes. You would be able to say that about an AI that analyzes a kingdom and ends up playing a terrible strategy.

Why can't he be opposed to trusting simulation on the grounds that the programmed strategies aren't very good?

He can, all I'm saying is that when someone provides specific data for specific strategies, any opposition should be equally specific.

Or maybe the simulator doesn't provide the flexibility in how strategies are defined to account for all the things a human might. I honestly don't know the answer to the last one, I've only played with Geronimoo's simulator a bit. Are there any obviously relevant pieces of game data that can't be accounted for easily while making strategies in the simulator? Ok maybe not obviously, are there any subtly relevant pieces of game data missing?

With Sparafucile's simulator, the one with which I've done most of my work, you can do everything you want. With it I'm confident that for over 90% of kingdoms I could produce a script within a day that plays better than I would on my first attempt.

Edit: 1111th post

[Mic Qsenoch]

i mostly meant that, even if rebuild/militia beats rebuild/mount/militia in straight simulation, whenever you do go mount and your opponent didn't, you can still switch into a BM plan, because I do think mount/militia beats rebuild/militia (hence the "?")

i also don't really trust simulation... not because of the concept, but people seem to place way too much trust into simulations which do significant strategic mistakes

I assume you mean tactical rather than strategic, as with scripts the strategies are given.

Can't the scripts have significant strategic mistakes?

A script is a tactical implementation of a strategy chosen by its programmer. You can say that Advisor-BM is a terrible strategy, but you can't say that an Abvisor-BM script is making any strategic mistakes. You would be able to say that about an AI that analyzes a kingdom and ends up playing a terrible strategy.

I tend to consider the so-called "tactical" decisions a subset of strategy. Dominion just doesn't have large enough range in the "scale/size" of decisions to meaningfully separate them for me. But I usually can understand what people mean when they say tactics, and I think I understand you.

[silverspawn]

I didn't want to make the claim that all simulations are bad. I said I don't trust simulations per se, because I've seen a lot of cases where they are just plain wrong. I remember someone saying that treasure map/masterpiece is worse than masterpiece because simulations show it, only to be proven wrong by someone else who made a few simple corrections in the code. I've also seen simulators having a <50% winrate with wharf/KC against simple BM, just to make two examples.

The results of a simulation are as good as the person who programmed it. When someone presents you simulation data and you respond by saying that you don't trust it, you accuse him of being either incompetent or sloppy, an accusation you are socially obliged to back up, e.g. by specifying what you think he did wrong. Saying that you have seen someone else mess up a simulation in the past is not enough.

I disagree that by saying "I don't trust this simulation" I accuse the guy who created it of being sloppy or incompetent, or really of anything. For one, the substant matter may so so complex that even a really good simulation makes lots of tactical mistakes, but more importantly, "I don't trust XX" is just a fact which, by itself, doesn't imply anything. I happen to be a rationalist, and I also happen to study programming, but someone else may distrust simulations in general, or he might not be able to read code, or doesn't want to. According to your logic, he is then obligated to blindly trust every simulation someone presents to him.

Really, I think what you trust or don't trust on is entierly your choice. My reasoning can be summerized as this:
-> I have seen lots of simulations that I know are bad
-> i have heard good players saying that lots of simulations are bad
-> I have heard good players saying that simulation in general has very little significance on actual tactic/strategy
-> I disagree with the general approach to simulation I've seen in this forum, at least to some degree

=> If someone uses simulation as an argument without telling me anything about its quality, I'm not going to blindy swallow it.

Disagreeing with my reasoning is fine, of course. but if you say I accuse someone you're interpreting things that aren't there.

Generally, I do believe that the guy who makes an argument should back it up, not the one who is on the receiving end.

[SCSN]

I disagree that by saying "I don't trust this simulation" I accuse the guy who created it of being sloppy or incompetent, or really of anything.

If the results of a simulation are not to be trusted that must be due to either sloppiness or incompetence. If the matter is too complex to properly simulate that clearly falls under incompetence. By saying you accuse the programmer of neither you effectively claim to be making meaningless statements, which is of course fine.

I didn't want to make the claim that all simulations are bad. I said I don't trust simulations per se, because I've seen a lot of cases where they are just plain wrong. I remember someone saying that treasure map/masterpiece is worse than masterpiece because simulations show it, only to be proven wrong by someone else who made a few simple corrections in the code. I've also seen simulators having a <50% winrate with wharf/KC against simple BM, just to make two examples.

The results of a simulation are as good as the person who programmed it. When someone presents you simulation data and you respond by saying that you don't trust it, you accuse him of being either incompetent or sloppy, an accusation you are socially obliged to back up, e.g. by specifying what you think he did wrong. Saying that you have seen someone else mess up a simulation in the past is not enough.

I disagree that by saying "I don't trust this simulation" I accuse the guy who created it of being sloppy or incompetent, or really of anything. For one, the substant matter may so so complex that even a really good simulation makes lots of tactical mistakes, but more importantly, "I don't trust XX" is just a fact which, by itself, doesn't imply anything. I happen to be a rationalist, and I also happen to study programming, but someone else may distrust simulations in general, or he might not be able to read code, or doesn't want to. According to your logic, he is then obligated to blindly trust every simulation someone presents to him.

No one has to blindly trust anything, but anyone too lazy to specify their objections should just refrain from critizing people presenting the results of a non-trivial amount of work. I believe the work I have done on these simulations to be of very high quality (and from what I've seen from Holger he's equally competetent), so while it's totally fine for you to personally mistrust it, the moment you make this public I demand you to back it up.

[KingZog3]

I think he just meant that many people quote simulator results and it quickly becomes out of context to the actual power of the card. Why is everyone so quick to assume that others are ill intentioned?

[terminalCopper]

Why is everyone so quick to assume that others are ill intentioned?

That's a quick assumption.

[silverspawn]

No one has to blindly trust anything, but anyone too lazy to specify their objections should just refrain from critizing people presenting the results of a non-trivial amount of work

The amount of work that has gone into a statement is not relevant if it's not reflected by how the statement is made. I had no reason to assume that whoever has done the simulations Holger was referring to has spent more than 5 minutes on them, because he hasn't told me anything about them. For me, just saying "because simulation", which is what he did, doesn't have any more weight than saying "because I believe it". If Holger had said the latter, you wouldn't accuse me of anything for not trusting it, so fact is that you do assign weight to the mere mention of simultation. that's precicely where we disagree.

[KingZog3]

Why is everyone so quick to assume that others are ill intentioned?

That's a quick assumption.

That's a quick assumption that I quickly assumed that... yeah it was. But it wasn't a quick assumption and people being ill intentioned. It was a quick assumption about people thinking other are ill intentioned. I don't think they are.

[SCSN]

I had no reason to assume that whoever has done the simulations Holger was referring to has spent more than 5 minutes on them, because he hasn't told me anything about them. For me, just saying "because simulation", which is what he did

He specifically told you he did the simulations himself. If you're at all familiar with the simulation work that's being done you know that Holger generally knows his stuff, so that his words deserve to be taken seriously and any objections be expressed with the appropriate level of detail. And if you're unfamiliar with the simulation work you should inform yourself before you criticise.

[silverspawn]

It's debatable whether he told me that he did the simulations himself. I'd argue that he didn't. What he said was this:

I've tried to simulate some of the combinations

Does that mean that he wrote the code? He could have just used someone else's simulator.

that's not essencial though. I am not familiar, and I didn't criticise anything, which is my whole point; all i said was that I don't trust the mere mention of simulation without further evidence

why don't we just let him decide whether or not he thought I was criticising him?

[SCSN]

why don't we just let him decide whether or not he thought I was criticising him?

Because what you're criticising is mostly my work, which Holger builds upon. And I certainly feel you're doing it injustice.

[c4master]

Well, showing the code so that anyone can investigate its correctness, would have helped here.

silverspawn doesn't like this "black box" magic where someone puts some (non-public) script data into a black box and then claims that either of the strategies represented by these scripts is superior to the other.

SheCan'tSayNo points out that the person who has done this simulation should be trusted to do this correctly. There might be evidence for this, for example previously done work by that person.

I really like to trust simulations, but only to a certain limit. I still believe, the AI does major tactical mistakes if the programmer doesn't take care of this. I've seen this by myself - an AI discarding a Monument to a Cellar, where Monuments are its only terminal. Obviously this strategy will not work correctly until some programmer tells the AI that playing such a card is an aim of strategy on its own. There might be other results like this in almost any other strategy, e.g. triggering an unwanted reshuffle etc. So unless the programmer tells me, what he has implemented (which features), I would distrust pure simulation if the results are closer together than, say, 60-40.

[pacovf]

Either SCSN or Holger could start a thread (if it hasn't already been done) discussing what is going into their simulator, the kind of choices they make, how it chooses a strategy, etc. SCSN being a remarkably good player, his claim that the simulator could potentially win in 90% of possible kingdoms after a day (which means a week of coding and another of debugging in devspeak) of serious scripting is very interesting.

I would very much like to read the thinking process that would get the simulator there. Would you use hard coded strategies? Evaluation algorithms (the way chess programs are mostly coded) to pick and choose which cards work best in a kingdom? And then there's more specific problems, like understanding first player advantage or multiplayer games, which I guess are not your prime concern, but I would still like to read about it should you eventually get down to work them into the simulator.

Personally, I distrust statements like "BM gets 4 provinces in 16 turns on average" more (who cares about variance anyway), because then people try a strategy that gets the fourth province on turn 18 and conclude that it was worse that straight BM.

[awildnoobappeared]

I remember a game where I beat Cultist with PStone+Storeroom (which is known to be a strong combo, made even stronger by all those Ruins) - it was close though, and I have no idea what its win rate is like against it.

Even then, you had to pick a strategy that works well specifically against Cultist -> you didn't ignore Cultist.

True, I guess that "Harvest" and "don't buy Cultist because I think something else is better on this board" are not the same thing.