Custom AI decks UPDATED 2013/08/27

[wololo]

It would be nice though if targeting, attacking and blocking decisions can be made better, since the logic involved is really not that complicated

Actually, the logic involved *is* complicated, but as Human beings we don't realize this. The AI doesn't know in advance what kind of effect a card is going to have. It has very basic understanding if a card has a "positive" effect or a "negative "effect, and that's it. For example, the AI doesn't know that a "shock" is not going to kill a 3/3 creature. But it is true that some small improvements could do lots of good things. I Just need to find the time to code them.
I am extremely open to very specific suggestions regarding the AI. You can look for Psyringe's posts on this thread: viewtopic.php?f=6&t=416&start=30

I also have a noobish question regarding the avatars. How do I install them properly? I dragged all of the posted avatar pictures and put them into the ai folders, but when I get to the play screen, the default avatar is always the one up. Thanks in advance

they go in the ai/baka/avatars/ folder

[Psyringe]

Hello dvdira, welcome to Wagic.

the AI is not targeting properly (...) The AI also does not block and attack properly

Yes, these are known problems of the current AI. If you can provide an improved algorithm for any of this tasks, then please do, but keep in mind that (a) the algorithm would have to remain pretty generic, since a long list of special conditions is bound to cause problems in a game with 10,000 different cards and tons of rules exceptions, and (b) the algorithm would have to perform decently on weak hardware (the PSP isn't the fastest machine on the planet ).

There's a thread specifically about AI Improvement, suggestions are very welcome!

It would be nice though if targeting, attacking and blocking decisions can be made better, since the logic involved is really not that complicated.

If the involved logic is easy for you, then you're probably the perfect person to write down a better algorithm. Pseudo-code would be quite okay. I've tried my hand at it myself, but I have to admit that with all the different evasion abilities involved (flying, shadow, fear, intimidation, landwalk), several counters against these (e.g. reach and reachshadow), other important abilities like first strike / double strike, trample, rampage, or deathtouch, cards that can't block or can't attack, etc., the logic for choosing attackers/blockers does indeed seem rather complicated to me. Nothing that can't be solved, but I estimate that I'd need at least 5 days of continuous work to come up with a good result (which means 5 days of doing nothing else but eating, sleeping, and working on this, and the result would be an algorithm which still would need to be implemented by someone), and so far I've focused on other things. If you can do it quicker, then by all means do, I'd be thrilled to have a better algorithm for these tasks (imho targeting and assignment of attackers/blockers are the two main reasons for the weakness of the current AI).

I also have a noobish question regarding the avatars. How do I install them properly?

They need to be in the folder Res/ai/baka/avatars and have to be named avatar1.jpg, avatar2.jpg, etc.

[jeremy]

Seeing as how its nearly halloween, heres a enchanting deck to trick or treat with.
4 Birds of Paradise -- accelerates mana n flies
4 Fist of Ironwood -- gives trample and 2 more tokens
3 Gaea's Anthem -- +1/+1 to all your guys in addition to whatever bonus from enchantress or mask
4 Predatory Hunger -- rewards you for every creature the enemy puts down
4 Primal Rage -- gives trample
2 Retaliation -- +1/+1 should the enemy choose to block your stuff
4 Web -- +2 toughness and lets you block fliers (would be Seal of Primordium if the AI wasn't so trigger happy sacrificing stuff)
4 Wall of Blossom --early blocker n nets you a card (would be Seal of Strength if the AI wasn't so trigger happy sacrificing stuff)
4 Yavimaya Enchantress -- the star of the deck, grows ever bigger with each additional cheap enchantment
3 Ancestral Mask -- Yavimaya Enchantress that you can target on your 0/1 Bird
4 Argothian Enchantress -- Cheap untargetable card advantage (altho the AI sometimes tries to attack with it =/)
21 Forest --simple forest base, since everything is 1-3 mana cost, this should be fine.

[The extension rar has been deactivated and can no longer be displayed.]

[Psyringe]

That looks like a nifty deck.

One question: I've been toying with an Enchantress deck myself (Yavimaya / Verduran / Argothian Enchantress plus Lorescale Coatl), but the deck sometimes wouldn't start well because it has many aura enchantments, but only 12 creatures to cast them on (since the Argothian Enchantress has shroud). In my tests, I had an uncomfortably high chance to end up with a hand full of auras and no creature to cast them on. Is that a oncern for your deck too, or have you found a solution for this? (I haven't tested your deck yet).

I like the decks you've posted recently. Would you be interested in including them in my AI deck comparisons?

[jeremy]

Hey, whatever you'd like boss.
Been busy lately due to mooching around in a new workplace, 10am-asia.
But sure, theres actually more but I had a few problems adapting due to sometimes accidentally including cards from other expansions like Defamation n FF7.
but I'll upload a full user deck that I was toying around with (with a few only having a few cards in as I haven't time to go explore, and some variations as I was trying to adapt the deck to "official" sets.
Could you pm me your email so I can sling it across to you with yousendit.com?
2 that I was working on were G/R goblins that focused more on control then aggro(blightning, wort,etc) and a pinger direct damage deck(which I could have made mono-red, but I like the splashing of blue, which I might have to revisit n increase the blue cards in)
As for the shroud problem, I included a couple other targets for them like wall of blossom and bird of paradise, not just the enchantress.
And most of the enchantments were "world" enchantments that doesn't need targets to function, I tried to make them all that way but they just were either too high up the mana costing or able to be sacrificed.
Some suggestions for replacements of 4 web and 4 Fist of ironwood would be ground seal(world enchant n helps block reanimators), Blair Patch(-1/-1 on enemy attackers), or maybe more mana acceleration in the form of elves. The legendary shrines I saw from the kamigawa block seemed too heavy on the mana cost, I was hoping the AI would be able to swing with a 10/10 enchantress by turn 4.
turn 1 -- forst, Bird
turn 2 -- forst, cast Yavimaya Enchantress
turn 3 --forst,cast double fist of ironwood, swing with 4/4 enchantress (4 dmg)
turn 4 --forst, cast ancestral mask on enchantress and ground seal,swing with 11/11 trampling enchantress and 4 1/1 tokens(19 dmg if unblocked) or you could cast the mask on one of the tokens have have them make a hard decision.
Oh well I'll be able to finish the deck with more testing.

[jeremy]

Oh and Psyringe, I'm currently trying to piece together a token generating deck centered around Nath of the Gilt-Leaf, Geth's Grimoire and discarding effects. Imagine them targeting you with a whole bunch of tokens out from the previous discards and having an overflowing hand from howling mines then casting a gaea's cradle powered mind twist. They might not be smart enough to ground seal but I'm sure with the amount of cards thats getting drawn any threat you put on the table gets terrored or the like to smithereens. Then comboed with Megrim to deal damage when your tossing cards or setting down an imperious perfect down to do a mass pump n walk all over you with the mountain of 2/2s with a freshly pumped Tarmogoyf leading the charge?
Food for thought?
rgds Jer
oh yeah and bonus, with enough elves more often then not Gilt-Leaf Palace will be a free dual land that fits in this theme.

[Psyringe]

Niegen, I added "Millage" and "Bad Dreams" to the SVN now. I could've added "Magnivore" as well, but after running it through my AI deck comparison, it seems that your suspicions about this deck were well-founded. It could only win 22% of its games. Personally I would have added it nevertheless (the Magnivore is a cool creature / feature), but if I have the right impression of your very quality-oriented approach to deck building, then you'll probably prefer to further improve the deck before spreading it in an official release. Just give me a shout if I'm wrong.

I also noticed two incorrect set infos in the decks "Spectral Rack" and "Bad Dreams":
Spectral Rack: "Volrath's Stronghold (TMP)" should be "Volrath's Stronghold (STH)"
Bad Dreams: "Font of Mythos (ALA)" should be "Font of Mythos (CFX)"

I already corrected these in the SVN.

Finally, I thought you might be interested in some of my recent endeavours:

Finished first round of the AI deck comparison, with several of your decks showing a very strong performance (Elfball nearly came out as the champion, but Treefolk managed to win one single game more ... out of 116 ...).

Cheat mode - a special option to aid testing. The "Complete Collection" and "Deck integrity" functions might be handy for deck creators.

Match, Gauntlet, and AI test modes - another set of tools to aid AI decl development. This one is currently still buggy (the features work, but it breaks other things, like the Test Suite), and the code has grown pretty ugly, so I wouldn't recommend using it unless you have a backup exe. I'm using the "Gauntlet" feature for my AI comparisons.

Have fun.

[Psyringe]

Could you pm me your email so I can sling it across to you with yousendit.com?

I'm having trouble with my regular email at the moment, can you upload it to rapidshare / mediafire etc. and send me the link in a forum PM?

As for the shroud problem, I included a couple other targets for them like wall of blossom and bird of paradise, not just the enchantress.

For me, having 12 alternative targets was not enough, but I may have had bad luck in testing. It's good if it works.

Some suggestions for replacements of 4 web and 4 Fist of ironwood would be ground seal(world enchant n helps block reanimators), Blair Patch(-1/-1 on enemy attackers), or maybe more mana acceleration in the form of elves. The legendary shrines I saw from the kamigawa block seemed too heavy on the mana cost

I agree with your analysis. I don't know good replacements for Web / Fists of Ironwood - I've been using Ground Seal in my own Enchantress deck, it's not that useful on its own, but as you said, there aren't that many good alternatives. Briar Patch might be useful ... I didn't put it into my deck because I was using using the Enchantresses as fast card drawers to pump up the Lorescale Coatl, so I wanted cheap enchantments, and Briar Patch was 1 mana too expensive for that. Perhaps I should rethink that, since Briar Patch would also counter the AI's stupidity of sacrificing its creatures as blockers instead of waiting until they get pumped up enough so that they won't die.

I agree that the shrines are too expensive for this kind of deck. I wouldn't add elves because I have a feeling that this would dilute the focus of the deck, but that's just a feeling, I may be wrong.

I'm currently trying to piece together a token generating deck centered around Nath of the Gilt-Leaf, Geth's Grimoire and discarding effects. Imagine (...)

Sounds interesting. Try it out. It might be too much for the current AI, but it's worth a shot.

[jeremy]

I took out all the pictures, n compressed the whole thing so its pretty small, also added 4 of each card for easy deck building to the play set. Pming you after I finish uploading.
There is the other option of splashing white for other non targeted enchantments, would kinda dilute the decks focus tho, and slow it down even tho there are some dual lands that are viable. Additions of Aura Shards would be brutal to artifact based decks.
Or maybe kitchen Finks as more viable blocking targets as well as the alternative is to take out Agoth and put in Verduran Enchantress instead that does the same thing except lacks shroud.
As for more viable world enchantments,
Concordant Crossroads gives everything haste n cost G,it could hurt it more then help against unfavorable match ups like sligh
Compost could be splashed for black match ups instead of ground seal.
Crosswinds splashed to deal with flyers.
Lifegift for some additional stall tactics.
Primal rage is a 1/G trample provider.
Retaliation helps in its stupid creature trade in the sense that any blocked attacker gets +1/+1 for 2/G enchantment
Storm front G n GG to tap fliers
Thoughtleech for GG n pretty much helps against every blue deck.
There are quite a few non targeted, non sacrificed enchantments, but they lack the omph to really stand out from being just a cheap pumper to your enchantress as they have pretty specialized situational abilities.
This could be adapted to player based decks with the addition of seals for giant growth in enchantment form and maybe Wall of Mulch to sacrifice itself n blossom to.
Nothing spectacular till they add more viable cards tho.
rgds Jer

[Niegen]

Niegen, I added "Millage" and "Bad Dreams" to the SVN now. I could've added "Magnivore" as well, but after running it through my AI deck comparison, it seems that your suspicions about this deck were well-founded. It could only win 22% of its games. Personally I would have added it nevertheless (the Magnivore is a cool creature / feature), but if I have the right impression of your very quality-oriented approach to deck building, then you'll probably prefer to further improve the deck before spreading it in an official release. Just give me a shout if I'm wrong.

Well, I don't mind Magnivore still being into the SVN, however, where are we with the 64 AI decks limit? If the limit is still there, by all mean, keep it out of the SVN for now.

I also noticed two incorrect set infos in the decks "Spectral Rack" and "Bad Dreams":
Spectral Rack: "Volrath's Stronghold (TMP)" should be "Volrath's Stronghold (STH)"
Bad Dreams: "Font of Mythos (ALA)" should be "Font of Mythos (CFX)"

I already corrected these in the SVN.

Good thing the deck parser was still able to load them!

Finally, I thought you might be interested in some of my recent endeavours:

Finished first round of the AI deck comparison, with several of your decks showing a very strong performance (Elfball nearly came out as the champion, but Treefolk managed to win one single game more ... out of 116 ...).

A shame! But what was the result of the Elfball Vs Treefolk match?

Cheat mode - a special option to aid testing. The "Complete Collection" and "Deck integrity" functions might be handy for deck creators.Match, Gauntlet, and AI test modes - another set of tools to aid AI decl development. This one is currently still buggy (the features work, but it breaks other things, like the Test Suite), and the code has grown pretty ugly, so I wouldn't recommend using it unless you have a backup exe. I'm using the "Gauntlet" feature for my AI comparisons.

Have fun.

Cool, I'll try those!