Dominion Strategy Forum

Please login or register.

Login with username, password and session length

Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Messages - blueblimp

Filter to certain boards:

Pages: [1] 2 3 ... 89
General Discussion / Re: Random Stuff Part IV
« on: June 23, 2021, 02:17:54 am »
Oh hey if weíre posting chess puzzles here, this is a really cool one:

Is this it?

Uppercase = white
Lowercase = black

n = K
i = R
k = B
g = N
t = Q
h = P

Previous move was by black: bxc1=R+

I tried to only do provable deductions to get here, so it should be a unique solution, but my confidence in accuracy is low.

It would be nice to have a single, official implementation.
While ideally there would be one perfect implementation that makes everyone happy, in practice I couldn't disagree more with this. The healthiest state of computer Dominion so far was in the pre-Goko era, when there were a variety of unofficial implementations that had different advantages for different types of players. It's just too hard to have one implementation satisfy, for example, both the people who want to play casually against AI on their phone and the people who want to play at top competitive levels on their desktop. (I'm still salty that it hasn't ever been possible to play with a deck tracker in the post-isotropic era. I understand that not everyone likes deck trackers, and if there's a variety of ways to play, those people can be happy too!)

So the AI cheats? It plays a few turns ahead and makes decisions based on that?

Not really? I mean, it doesn't know exactly what shuffle it's going to get. It just simulates a bunch of potential shuffles based on what happens if it makes certain moves and then picks ones that made it win more in those simulations.

OK, that's fine for shuffles. But say it plays Wishing Well, or decides whether it's going to play Wishing Well. What then?

I guess I don't really understand what cheating you think could be happening. It plays Wishing Well, it runs some search, it guesses the card it thinks will give more win rate.
I think the confusion here is LastFootnote thinking you're referring only to future shuffles, while I assume you mean to also re-randomize the shuffles that already happened (which, practically, would mean re-shuffling the deck before each simulation).

IMO, though, properly re-randomizing unknown info is subtly difficult, both from an algorithmic perspective and to code it without bugs.

At first it seems simple enough to just randomize everything it doesn't know before each simulation: the order of its deck, the contents & order of its opponent's hand, deck, and discard (and possibly other things I'm not thinking of). But actually it can know something about these:
  • The top few cards of the decks may have been revealed for various reasons.
  • The opponent may have revealed their hand at some point.
  • etc.
Even trickier: it's possible to make inferences from the opponent's play about certain things that aren't actually known. e.g. If the opponent didn't play an action last turn, they probably didn't have their Smithy in hand; but this is just because we're assuming something about the opponent's strategy, because it would have been legal for them to not play it.

To do this all optimally is (I think) comparable to playing poker, and having it mixed in with all the other parts of Dominion makes it extra hard.

And it would be really easy, when programming this, to accidentally leak info that the AI is not supposed to have. The safest way might be a sort of clean-room approach: only provide the AI with public info, from which it would create a new game state. (The easier and more bug-prone approach is to copy the game state and try to scrub hidden info out of it.)

That said, if the objective is just to make a fun AI opponent for an app, none of this is too important. I mean, if it's just playing strategies like BM+Smithy, then whether it plays Wishing Well optimally or legally doesn't even matter.

And even if they're right and it's an amazing Dominion AI that can beat even seasoned players in most games, that sounds pretty bad for casual players!
They're not claiming that the AI beats expert players. The quoted player self-describes his experience as
Iíve been playing Dominion casually, in person, for years. Iíve probably played 100, 200, maybe more games.
So we're talking about an AI that can beat a player who has played 200 casual games of Dominion. Depending on the sets included, even BM+X might be able to do that.

"Instead of having a concept of a card in our neural network, we have a concept of each of the components of a card,Ē said Duringer. ďThe card has a cost. The card has a victory point value. The card might give you an extra buy or an extra action. If we can just understand these components, we donít need to know the Ďvalueí of a card. We can surmise it on the spot by understanding those components."

It's simultaneously hilarious and sad. The quote is from the Temple Gates CEO, so I can only hope this information is either misleading or outdated.
This bit struck me as odd too. Breaking down cards into components like this might work alright for vanilla cards, but most cards have unique text.

Plus it seemingly solves a non-problem. The AI doesn't need to adapt on-the-fly to new cards, because they can just re-run the AI training whenever they add new cards.

Thank goodness that the mobile license is finally in the hands of a developer that has a track record of mobile boardgame adaptations. (I havenít tried this companyís apps, but they get good ratings in the app store.) Itís crazy to me that in the 8+ years since Iso was shuttered, this is the first time that has happened.

Adapting Dominion to a small screen sounds insanely hard, so I wonder how thatíll work. Iím also curious about the AI, which is also a hard problem.

General Discussion / Re: The Shipwreck Arcana and broken logic
« on: May 10, 2021, 06:15:00 am »
If you can agree ahead of time on a convention, then there are multiple conventions that can work, as silverspawn points out.

If you can't agree ahead of time on a convention but expect that your partner is going to play according to a convention, then you're trying to find a Schelling point: you and your partner want to choose the same convention, despite not being able to communicate. I don't know much about these, but they seem more about psychology than about logic or math. Jack Rudd's convention (of playing towards the edges) seems like a reasonable choice because it's the only symmetric convention. (Note: The reasoning in your original post resembles the level-n theory described in the Wikipedia article.)

If your partner doesn't play according to a convention (which is not an effective way to play, but maybe they're bad at the game), then there's probably not much you can do.

Other Games / Re: Celeste
« on: January 30, 2021, 06:32:18 am »
They've released a PICO-8 sequel to Celeste Classic: It's worth a play.

Other Games / Re: Hades
« on: December 31, 2020, 03:08:08 pm »
I feel like this game is a bit short on quality-of-life features. Like, there's no practice mode or level warp of any sort. There's a mod that does it (CodexMenu), which is useful but a bit janky (e.g. it seems like it hangs when I warp to Magaera while wielding specifically the sword... haven't figured out what's going on with that yet).

Edit: TBH I'm not sure this game is "for me" because I don't care about the narrative and meta-progression aspects and it seems like those are a big focus. But the arena combat is fun so I don't mind approaching it from that angle (vaguely like Hoplite, which also happens to be based on Greek mythology... coincidence?).

General Discussion / Re: Random Stuff Part IV
« on: December 17, 2020, 04:07:07 pm »
AFAIK, computers are still garbage at the strategic part of Starcraft.

I don't think so. AlphaStar had a decent series against Serral, and it had strictly limited APM.
I vaguely remember that being more driven by consistent micro (which is an advantage computers have even with limited APM) than by strategy. I think SC is kind of a bad testbed because being real-time makes it hard to get a meaningfully even comparison with humans.

General Discussion / Re: Random Stuff Part IV
« on: December 17, 2020, 05:00:56 am »
Randomness means search isn't going to help much, unlike in chess and go. Randomness isn't itself such a big problem; IIRC computers were playing superhuman backgammon before superhuman chess. But it does diminish the opportunity to out-calculate humans.

Doesn't that also work the other way round? Humans are notoriously garbage at calculating probability (especially at guesstimating it when the math gets too complicated to do on the fly), but for computers, it's trivial. Working out chess or go decision trees in your head is still impossibly hard to do even nearly as comprehensively as computers can because you just don't have the brain capacity to do it, but I don't think it's an inherently difficult skill for humans.
Could be. I'm not sure about this point.

I should also clarify that I'm imagining an AlphaZero-style AI that learns strategy for itself. It's probably more straightforward to make a superhuman Dominion AI by giving it a lot of per-card strategy hints, but that's a lot of work and not very interesting.

General Discussion / Re: Random Stuff Part IV
« on: December 17, 2020, 01:27:07 am »
How hard is dominion to learn for AI? Easier than chess? As hard as chess? As hard as Go? Harder?
Harder for sure.

First issue is how you deal with each card introducing its own rules. Arguably, a fully legit Dominion AI would be able to play with a new card without having to re-train the model. That's mega hard, so realistically you probably would just re-train the model.

Some strategies in Dominion need all the components in place or they are garbage, e.g. in some kingdom, if you try to play an engine without +buy it may be garbage, and with +buy the best strategy. You can't really just do random stuff and stumble into a decent strategy the way you can in chess and go. That's a problem because that's how self-play training works.

Dominion has hidden information, most notably your opponent's hand. It's like there's a little bit of poker embedded in the game. AIs can play superhuman poker now, but the techniques used for that are completely different from what's used for other games.

Randomness means search isn't going to help much, unlike in chess and go. Randomness isn't itself such a big problem; IIRC computers were playing superhuman backgammon before superhuman chess. But it does diminish the opportunity to out-calculate humans.

Strategies in Dominion are highly diverse: what's good in an engine is totally different from what's good in BM+X. This is different from chess and go, where, despite some minor variation, you're doing essentially the same thing every game. I can't off-hand think of any game with diverse strategies where computers are superhuman. I don't know if that's because it's hard or just because it hasn't been seriously tried, but it means you're breaking new ground.

I'm not an AI researcher, but if I were trying to develop a Dominion AI, I'd start by making some massive simplifications. Singleplayer only (with the goal being something like buy 5 provinces as fast as possible) and limited subset of cards (maybe just first kingdom). This already sounds pretty hard. Try to get it to discover the first game engine, for example. (Although that has Militia, so maybe it's not even good singleplayer...)

Starcraft has an astronomically larger decision space than either of those games, and I don't think it was that much harder.

AFAIK, computers are still garbage at the strategic part of Starcraft.

How fast can a pure Menagerie kingdom buy out all the Provinces? I'm reasonably confident it's possible in 6 turns, maybe even in 5. I haven't worked out the details, but getting the key ideas is the fun part.

You're allowed kingdoms that Dominion Online allows (meaning up to 4 landscapes).

Here's my idea:

The optimistic plan:

Turn 1: Buy Stampede.
Turn 2: Play the Horses using Way of the Chameleon. Buy Populate. Because of Cavalry, clean-up triggers a reshuffle.
Turn 3: Buy 2 Populates.
Turn 4: Buy 4 Populates.
Turn 5: Buy all Provinces.

I'm not sure buying 2 Populates on turn 3 is actually possible, so maybe it takes 6 turns to do it. Shouldn't take longer than that though.

The 10 action cards (tentative):
Wayfarer, Livery, Displace, Barge, Hunting Lodge, Hostelry, Village Green, Animal Fair, Scrap, Cavalry

Include also Way of the Horse.

On Dominion Online, I used these action cards (except Cavalry was Supplies) to try exponential growth of Populate later in the game. So either this works as-is or should work with some tweaking.

General Discussion / Re: Most played board games of the past 30 years
« on: January 18, 2020, 08:36:28 pm »
That it's using accumulated plays (instead of rate of plays) makes the field seem more stagnant than it really is.
I wonder this particularly in the case of Codenames. Presented differently, its entry might actually look more impressive than the Dominion/RftG entry. It doesn't crack the top 3 in this visualization but that seems to be because the top 3 have so many accumulated plays.

General Discussion / Re: Most played board games of the past 30 years
« on: January 18, 2020, 08:03:28 pm »
That it's using accumulated plays (instead of rate of plays) makes the field seem more stagnant than it really is. But it also that makes the entry of Dominion and RftG that much more impressive.

I ended up watching in 2x though because 1x speed is way too slow.

General Discussion / Re: STAR WARS
« on: January 03, 2020, 03:29:54 pm »
There were so many great things about all the movies, but taken as a trilogy, they really feel like a joke to me.
They really do feel more like three unrelated movies than a coherent trilogy. That the 3rd movie is frantically retconning elements of the 2nd movie, when they knew _before making the 1st_ that they would be making a trilogy, is pretty unforgivable. Though, to be fair, the OT was a little guilty of this too. (AFAIK, Leia being Luke's sister wasn't decided until writing ROTJ, and that revelation creates some internal consistency problems.)

General Discussion / Re: STAR WARS
« on: January 02, 2020, 01:15:55 am »
I think I was saddened the most was the squandering of some great ideas. Rey having to choose between letting the Final Order kill the Resistance, or killing Palpatine and becoming a Sith... that was quite the conundrum. Then, of course, it doesnít matter, and Rey beats Palpatine by... using two lightsabers.
The way that scene resolved was so weird. In the first part, like you, I thought they were setting up a puzzle, where the rule is: if Rey kills Palpatine, she loses. The obvious way out is for Ben to somehow kill Palpatine instead. The writers could have found a clever way to do that, or could have found a way for someone else entirely to do it. But instead the setup is basically just forgotten once Ben shows up.

General Discussion / Re: STAR WARS
« on: December 31, 2019, 03:45:20 pm »
It's just a sequence of plot points the author wanted to get to connected by meaningless action scenes, and the resulting story looks like swiss cheese.
One of the many examples of this was

The scene from the teaser trailer where Rey fights a TIE fighter. It's a good teaser, because it makes you wonder how she got into that situation. Then, in the movie, it turns out that it happens for no reason and has no impact on the plot. Argh.

General Discussion / Re: STAR WARS
« on: December 31, 2019, 02:26:43 pm »
TROS is for me perhaps the worst of the new trilogy because it's the only one where I at times felt bored while watching it in the theater. The combination of nonsensical plot plus relentless action is pretty dull. TLJ was for me the most enjoyable of the three, but it was also a mess that made the badness of the third movie almost inevitable. I guess I'd rank TLJ>TFA>TROS, but really they all played a part in convincing me that I'm done with Star Wars forever.

General Discussion / Re: Maths thread.
« on: December 05, 2019, 03:38:32 am »
I wrote the Putnam the maximum 4 times. It was fun. I enjoyed it more than the other math contests I've tried. I'd suggest practicing specifically past Putnam questions to get used to its particular style if you haven't, though it's a little late to do that now.

Tip-wise, scoring on the Putnam is very bimodal, it's basically 10/9/8 if you get it or 0/1 if you don't get it.

I've been told this too, but somehow, my first time writing, I managed to get a 6 (I think it was).

Hearthstone / Re: Semi-interesting Hearthstone moments
« on: October 02, 2019, 11:29:06 pm »
Even just giving a random golden epic at the end of every wing on top of what we already get would have likely been enough to get me to buy them.

If they think I will pay a premium to get access to extra single player content, they are mistaken.
To be honest, I don't understand this mentality. Golden cards in general are useless bling, and a random golden epic is unplayable useless bling. You'll pay for that and not for hours of high-quality PvE content? I mean, Hearthstone PvE is better than many standalone singleplayer games.

Hearthstone / Re: Semi-interesting Hearthstone moments
« on: October 02, 2019, 06:10:38 am »
after a few turns of fatigue, he gets a new deck. wth?
The plague lords get a new deck whenever they change phase.

So, 4th wing is out, plus the final boss. The 4th plague lord was the easiest by far, for me. I brought a Hearthstone with me and didn't need to use it, and my deck wasn't that great (though it did have quite a bit of value).

The final boss is somewhat similar to the Monster Hunt final boss, which I enjoyed a lot, but it feels meh here. I think that's because the hero powers aren't very interesting. Monster Hunt had some very complex hero powers.

Overall, the adventure was alright but the format feels stale. The biggest change, the 300-HP plague lords, I don't like much. If the next adventure is like this one then I might not buy it.

Hearthstone / Re: Semi-interesting Hearthstone moments
« on: September 25, 2019, 09:37:39 pm »
The third plague lord is my favorite so far, because you have a lot more control over what's on your board compared to the others. The only thing I found dumb about the encounter is the insane amount of self-healing he runs. In particular, when he plays an Obsidian Statue in phase 2, the amount of healing he gets from it is massive.

I managed to take him down in one run (on Normal) but only because I had an OP deck and a Hearthstone to do it in two parts. I had the passive that makes your spells summon a minion from your deck, lots of huge minions (4x Kel'Thuzad), lots of cheap spells (6x Raven Idol, 2x Worthy Expedition), and also the passive that triplicates your discovers with a cost discount. My best-ever turn 1 (on an earlier boss, not the plague lord) was something like Worthy Expedition into triple Power of the Wild, which put a bunch of stuff on the board, including two Kel'Thuzads.

I'm currently thinking that Hearthstone is by far the best treasure, if you're trying to take out the plague lord in one run.

General Discussion / Re: roguelike games
« on: September 25, 2019, 09:30:16 pm »
Has anyone tried Noita yet? It sounds interesting, I can't justify buying it yet but it's on my radar.
I'm intrigued by the per-pixel simulation. Reminds me a bit of games like Liquid War, Lemmings, Worms, etc.

The reviews make it sound like it's in a pretty early state though. So I might wait a while.

BTW I thought they might use the GPU to do the simulation, but the system requirements make it seem more like they're doing it on the CPU.

Pages: [1] 2 3 ... 89

Page created in 0.088 seconds with 19 queries.