Print

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

[blueblimp]

I'm mostly just interested to see what sorts of crazy bots people sent in, so just do whatever involves least effort, in my opinion.

[WanderingWinder]

Ok, since nobody seems to care that much about things, and since I don't have a lot of time to sink into re-doing everything soon, I'm just going to post the results. I've attached a file with all the bots that were submitted. Check out the bottom of the post.
Results, in alphabetical order:
Bot                                                                                            Record                      Overall percentage
CQMBWO By DG                                                                            9-2                           73.87
Dstu-Chapel/Quarry/GM/Highway/Outpost                                      6-5 (5-3)                     52.08 (54.11)
Drunk Marine Students 5                                                              2-9 (2-6)                     28.88 (36.81)
Go SMoKL by DG                                                                           6-5                            49.02
Goons/Governor/Quarry                                                               1-10 (1-7)                  19.49 (24.05)
Ill-Gotten Gains/Gardens/Council Room                                           3-8 (2-6)                     28.14 (30.64)
Ironworks Rush                                                                            10-1 (7-1)                   78.24 (78.83)
KC-Bridge with Wharf*                                                                  5-5 (5-3)                   52.69 (56.62)
SaW by DG**                                                                                5-4                             60.87
Wharf/Hamlet/Bridge                                                                    1-10 (1-7)                   10.95 (12.60)
blueblimp: Governor/Goons/Quarry/Masquerade/Fishing Village*        9-1 (8-0)                     84.08 (88.07)
blueblimp: Ironworks/GH/Island/Lighthouse/Silk Road                       7-4 (5-3)                    65.29 (66.57)

The * represent how many of the matchups of that bot couldn't be finished. Basically, there's something with the SaW bot that for some reason can smash the simulator up sometimes. Not sure what the deal is exactly.
The parentheses represent the records of what happened if you take DG's bots out - they used starting conditions, so are technically 'illegal' - well, I don't know how much of an impact that has; on one of them, a lot, but overall, I don't know that if it's that much.
Anyway, you can see who's won already, but I'm going to talk a bit about the top few bots here.

CQMBWO: This bot is really really long. I actually had to get DG to resend the end of it, because it got cut off. Anyway, the cards it uses are quarry, worker's village, chapel, menagerie, and bridge. The idea is to build an engine capable of going mega pretty quickly, while looking to pile the game out if it can with a win at any point along the way. Interestingly, even though there are so many conditions, I'm guessing it's missing a LOT from how a human player would see it, in terms of piling out with a win. It's very hard to check everything, there are so many conditions. Of course, in the real matchups, your opponent is much more likely to pile out on you, too! This did use starting conditions, but given that it's 4/3, and that it wants a chapel pretty early anyway, I don't think this had a huge huge impact.

Ironworks Rush: A nice bot from the simulator-writer himself, Geronimoo. It's actually quite simple, quite possibly the simplest of all the bots I've seen in a long while for anything remotely competitive on the stupidly-powerful level. All it does is chapel down, gain lots of ironworks, then great halls, then islands, running piles super fast. I can't help but wonder if it would have been better if you could come up with a better 5th card than peddler. But anyway, it's extremely strong as is.

blueblimp: Governor/Goons/Quarry/Masquerade/Fishing Village: The winner, and well-deserved. Also fairly simple. The idea is to draw your whole deck with governor, play a goons, buy something, rinse and repeat, buying more FVs, Governors and Goons along the way, to achieve massive scores. Now, this deck benefits A LOT from being able to play goons into masquerade, which in the sim is going to mean stealing good cards, but them's the shakes - totally legal by the rules here.


I also want to mention Dstu's bot for a second - it is pretty cool, and of course quite strong, but it loses just over 20% of its games to everyone, and I don't just mean here. I mean, throw it up against a BM bot, it loses that much. What's the deal? Over-aggressive chapel trashing. I think this is fixable too, sadly? Someone want to explain how? Anyway, I don't think it's going to actually swing that many matchups in this tournament, but it's definitely something that made it look like it's a lot weaker than it is.



Also, only a handful of match-ups went to 100,000 games before getting a clear enough winner, and only one was close enough after that to call for more looks - Go SMoKL against KC-bridge with Wharf. Finally, these sims took tons more time than the stuff I normally simulate. All those buy rules were having me go over half a minute for several of the 10k sim matchups, whereas normally I'm in the ballpark of 10 seconds. You guys really know how to complicate things up

[DStu]

I also want to mention Dstu's bot for a second - it is pretty cool, and of course quite strong, but it loses just over 20% of its games to everyone, and I don't just mean here. I mean, throw it up against a BM bot, it loses that much. What's the deal? Over-aggressive chapel trashing. I think this is fixable too, sadly? Someone want to explain how? Anyway, I don't think it's going to actually swing that many matchups in this tournament, but it's definitely something that made it look like it's a lot weaker than it is.

Easiest fix would be to get Geronimoo's to change $-treshold for agressive trashing from 4 to 5 .

I spotted that problem, but this bots wins against its clone with standard trashing, so I took this one. One of the problems is that Quarry counts as $3 for this treshold, and as I only have actions cards costing $5 or more, and Quarry, you often get stucked with Highway/Chapel/Quarry. I didn't find a way out of this without losing the mirror, so I figured I should better try to win the other 80% as good as possible, as the rush is probably lost anyway once I get to this state.

It actually does better than 80% when attacked early, because it can't trash that agressively in this cases.

[WanderingWinder]

I also want to mention Dstu's bot for a second - it is pretty cool, and of course quite strong, but it loses just over 20% of its games to everyone, and I don't just mean here. I mean, throw it up against a BM bot, it loses that much. What's the deal? Over-aggressive chapel trashing. I think this is fixable too, sadly? Someone want to explain how? Anyway, I don't think it's going to actually swing that many matchups in this tournament, but it's definitely something that made it look like it's a lot weaker than it is.

Easiest fix would be to get Geronimoo's to change $-treshold for agressive trashing from 4 to 5 .

I spotted that problem, but this bots wins against its clone with standard trashing, so I took this one. One of the problems is that Quarry counts as $3 for this treshold, and as I only have actions cards costing $5 or more, and Quarry, you often get stucked with Highway/Chapel/Quarry. I didn't find a way out of this without losing the mirror, so I figured I should better try to win the other 80% as good as possible, as the rush is probably lost anyway once I get to this state.

It actually does better than 80% when attacked early, because it can't trash that agressively in this cases.

I think you can stipulate that you should buy copper if you have fewer than $3 in your deck, or fewer than $5 in your deck with at least one quarry, or fewer than $7 in your deck with 2 quarries. I *think* this stops it from trashing your copper down too far. I also want to say that one of the other competitors had this in their bot somewhere...

[DG]

I suspect my CQWMBO script might be better in play with an ironworks but the simulator trashes more with the quarry so I used that instead.

Geronimoo let me know that he'd found the fault with duration cards in play at end of game not always being cleared up between games. I specifically found this happening against swindlers, jesters, etc. Maybe we can have a look at these again after another patch.

[DStu]

Now that I had some time to look over the other bots, congrats to everyone. I'm really impressed how strongly and fast you can explode, given you have something like 9-10 turns or so. Especially bluellimps winning bot.

[DG]

Yep congratulations to Blueblimp. Goons + masquerade assembled quickly enough to beat the rush decks. The only way to stop it seemingly was to dismantle it entirely and put it all in the trash!

[Geronimoo]

Finally had a chance to look at these bots in detail. I was especially interested in DG's creations because he's always pushing the boundaries of what the simulator can do and he did not disappoint! Just go and watch a sample game of the Menagerie/Bridge bot, it's beautiful and I'm happy to see the simulator's play rules for Menagerie turned out working so well (although not with Ironworks... which would make the bot even more powerful). The Saboteur strategy is a lot of fun and the Golem Mountebank should cause a lot more mayhem once some fixes.

Dstu's Grand Market bot is extremely fast, but Chapel will trash the deck's economy completely at times (caused by Quarry which the simulator thinks = Gold). There's probably an easy fix for this.

Congrats to blueblimp of course for winning!!! When I was working on my bots I never even considered using Governor because I thought the simulator couldn't really play it... apparently I was wrong and the extremely simple play rules for Governor I implemented were good enough to create the winning bot. Nicely done!

And big thanks to WW for running this. It was a lot of fun creating these powerful bots.

I wonder if we can get some useful information out of this tournament for the general Dominion playing public? Or are 5-card combos too rare to matter?

[blueblimp]

Haven't gone through the bots yet, but I'm interested to see how Chapel was used effectively. I gave up on Chapel because of not being able to fine-tune the trashing. (My Ironworks bot originally bought a Chapel, but removing it actually improved the bot!) Masquerade is a nice way to get some trashing while causing problems for the opponent bot.

I got some luck that the simulator plays the Governor/Goons/Masquerade sequence very well. The play rules for Governor will almost always take +3 cards with this deck, which is exactly what I want. Also, it's good that Masquerade gets played after Goons.

A big strength of this bot is that once the engine is running, it essentially forces the other bot to play each turn with a 2-card hand (since games are so short that the bot doesn't run out of coppers and estates to send over). In principle this can be countered with Lighthouse, but LH takes up a valuable kingdom slot and is pretty awful if your opponent isn't attacking.

I originally designed this bot to not buy any green at all, but that was a bit too slow against Ironworks. Luckily the engine is strong enough to support Colony buys pretty early.

Geronimoo let me know that he'd found the fault with duration cards in play at end of game not always being cleared up between games. I specifically found this happening against swindlers, jesters, etc. Maybe we can have a look at these again after another patch.

Does this mean that Fishing Villages in play at the end of game could contribute money to the first turn of the next game?

[DG]

I was especially interested in DG's creations because he's always pushing the boundaries of what the simulator can do and he did not disappoint! Just go and watch a sample game of the Menagerie/Bridge bot, it's beautiful and I'm happy to see the simulator's play rules for Menagerie turned out working so well (although not with Ironworks... which would make the bot even more powerful).

Um, well actually against Blueblimp's bot it typically either discarded the menagerie when goons was played or passed it on the masquerade so it almost worked! Perhaps the discard is worth a coding change. On the whole though the play rules are good. To return the compliment I thought Geronimoo would have some green card script that would be faster than my CQWMBO script and indeed that was true, just. The average game length between them was 10.0 turns, always emptying 3 piles.

Congrats to blueblimp of course for winning!!! When I was working on my bots I never even considered using Governor because I thought the simulator couldn't really play it... apparently I was wrong and the extremely simple play rules for Governor I implemented were good enough to create the winning bot.

Same here, although I did know that governors were a problem for the chapel decks. They can donate silvers that are never trashed and will clog up a deck, especially a menagerie/chapel deck. The chapel engines generally can generally draw through the deck too so the extra cards are worthless. In that respect I'm not surprised that two of my scripts can at least give the Blueblimp script a fight since they can survive as a big mess.

If anyone wants some fun they can try my mountebank deck but with swindlers instead. In one way this works much better since the simulator always plays the swindler after the golem but unfortunately it swindles so much that the game goes completely out of control. You can also replace the saboteurs in my third deck with bridges, mountebanks, coppersmiths, whatever.

[Schneau]

I'd love to see a table of the round-robin matchups between all the different bots. I submitted three of the worst bots (Drunk Marine Students 5, Goons/Governor/Quarry, and Ill-Gotten Gains/Gardens/Council Room). I didn't have much time to optimize these, but it thought it would still be fun to see how they would do, and they all managed to win at least one matchup. Even beyond my bots, it would be interesting to see which strategies beat which others.

[Geronimoo]

Does this mean that Fishing Villages in play at the end of game could contribute money to the first turn of the next game?

The problem is with King's Court+Duration cards. The rules of the game say that King's Court will be set apart with the duration it tripled. I coded it so the duration is linked to the King's Court, but sometimes this link will not be cleaned up at the end of the game, causing null-pointer exceptions in the next game (pretty much crashing the simulator).

[Geronimoo]

I'd love to see a table of the round-robin matchups between all the different bots. I submitted three of the worst bots (Drunk Marine Students 5, Goons/Governor/Quarry, and Ill-Gotten Gains/Gardens/Council Room). I didn't have much time to optimize these, but it thought it would still be fun to see how they would do, and they all managed to win at least one matchup. Even beyond my bots, it would be interesting to see which strategies beat which others.

You can easily run these yourself. Just "load players" with the xml WW added as an attachment.

[Schneau]

I'd love to see a table of the round-robin matchups between all the different bots. I submitted three of the worst bots (Drunk Marine Students 5, Goons/Governor/Quarry, and Ill-Gotten Gains/Gardens/Council Room). I didn't have much time to optimize these, but it thought it would still be fun to see how they would do, and they all managed to win at least one matchup. Even beyond my bots, it would be interesting to see which strategies beat which others.

You can easily run these yourself. Just "load players" with the xml WW added as an attachment.

Sure, but if WW has a table with the results, its a lot easier and faster than me running 12*11=132 simulations by hand. Or is there a way to play a round robin tournament with the simulator that I'm not seeing?

[DG]

Load them up. It's fun to see how these scripts win and lose as well.

[Schneau]

It looks like the Wharf/Hamlet/Bridge bot is missing from the xml file.

[WanderingWinder]

I'd love to see a table of the round-robin matchups between all the different bots. I submitted three of the worst bots (Drunk Marine Students 5, Goons/Governor/Quarry, and Ill-Gotten Gains/Gardens/Council Room). I didn't have much time to optimize these, but it thought it would still be fun to see how they would do, and they all managed to win at least one matchup. Even beyond my bots, it would be interesting to see which strategies beat which others.

You can easily run these yourself. Just "load players" with the xml WW added as an attachment.

Sure, but if WW has a table with the results, its a lot easier and faster than me running 12*11=132 simulations by hand. Or is there a way to play a round robin tournament with the simulator that I'm not seeing?

I have a table. It's on a piece of paper.