AI Development Discussion

classic Classic list List threaded Threaded
1250 messages Options
1 ... 36373839404142 ... 63
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

redrum
Administrator
@captaincrunch - Nice games. The AI doesn't understand the 'Kamikaze Planes' option so doesn't defend properly against them. There are some games that allow it by default to some extent but most games its disabled by default. Something I may eventually add support for but low on the priority list.

@Black_Elk - Happy Memorial Day! Lots of interesting games! Looks like you aren't playing by the no strategic bombing AI house rule right now. Feedback in order of your games:

1. The big mistake I see is in round 5 when the Russians don't crush the Germans in West Russia. I tested that turn again and realized there was a bug in capital defense related to air only attacks. I fixed the logic and now the Russians pretty much always attack West Russia which leaves them in a much better position: Game1_Russia_Should_Attack.tsvg

2. Looks like this one boils down to the round 6 Russian attack on Karelia which is about an even TUV swing battle. The Russians just get extremely lucky (end up with like a 100+ TUV swing) which essentially ends the game.

3. Cheers to the the Axis AI! Well fought!

4. That's really a terrible bid. Especially the USA's bid to just throw its 2 tanks away round 1 in China. Makes me realize how bad the old bid logic is and how much it could be improved. Probably better doing a logical bid for the AI at this point. I don't think there were any huge mistakes besides the poor bid that I saw just some good plays by you. But yeah understanding how to strategic bomb would definitely level the playing field a bit.

5. There was a bug with the AI calculating enemy multi-move attacks when a battle had only planes remaining. In round 6, when the AI gets unlucky in the Archangel battle it didn't realize the tanks could then attack through it since the plane has to land somewhere else. I fixed the logic and now it properly defends Russia: Game5_Russia_Capital_Defense.tsvg

Both fixes will be in the next pre-release. I'd be interested to see you try not using strategic bombing and do a logical bid for the AI's. Then you'll probably see even more competitive games :)
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Black Elk
This post was updated on .
Killer! Thanks for looking em over.
Nailing down the fixes and moving forward. Glorious

Yeah haha, for whatever reason I felt like being merciless yesterday with the bombers! ;) I'd say the main difference between SBR games and no SBR games, is the amount of time it takes for game resolution. Bombing seems to shave off about a half dozen rounds in the endgame, one way or the other (Whether it works, or fails miserably, the game length is reduced a bit.) The trade in SBR is better in 1942.2 over say Classic or Revised, since the cost of bombers was dropped to 12, but it still backfires pretty regularly. Sometimes they crush, but they like to get shot down too just as often, or roll low numbers even if they get through on occasion. So I don't thinkAI bombing would necessarily improve the hardAI gameplay, it would just randomize the outcomes a bit more, as it does when playing against a human. Sometimes in the endgame though, when AI is stacking a ton of bombers altogether, a decisive raid would be cool to see, at least if no ready combat opportunities present themselves.

Letting the machine choose their bid was a bit of a novelty. Figured if a tripled the normal bid of 12, and gave it to each, then one of 3 Allies might use it effectively lol. Giving them a regular human-like bid would probably be better, though the machine doesn't always use the extra units in the way a human might. British subs to clear the med for example, or a crack on 37. I noticed a definite pattern however, when the machine chooses an Allied bid for itself. Russia is predictable, they use theirs for ground consistently. Either tanks, inf or artillery in varying combinations, and use these to make additional attacks. Finding a bid that persuades them to stack W. Russia has proved a challenge. They really like to go forward in Baltic or Belo. UK sometimes goes for extra naval units in sz 36. But more often they'they'll take extra troops in Egypt or Burma.  USA likewise either wants to go sz 11 or into China. Sz 11 seems wise, though sometimes I see them placing a transport here, which never works as G just smokes them before they can load. Sometimes UK does this in sz 36 with the extra transport, though there the transport at least has a better chance to survive.

I toyed with upping the Allied bid to 15 or 18 each, to see if it would change the dynamic, with factories or aircraft, but the current bid logic seems to stick to what it knows. Air or production bids might be fun, but I haven't seen them. Right now the artillery combo bids seem the most effective thing the AI bid has going for it. Ground with Russia/UK, and USA dropping a warship in sz 11. 3 bombers might be fun though. I might give that a try next.

Eager to check it out again, and test. Playing against the HardAI is enjoyable. I dig it.
Fun stuff as always. Great work man!



Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Black Elk
This post was updated on .
Here was a fun game against hardAI allies. I gave them a  competent bid, spending 12 ipcs for each, in locations that seemed appropriate.

Russia with 4 infantry on the eastern front vs G.
UK with 2 subs, one in sz17 for the med and one in sz36 for the pacific.
USA with one cruiser in sz 11 for Atlantic transport defense.

So basically a bid of 36 ipcs, about 3 times what I'd give a human, to compensate for skill. HardAI Allies opened quite well! And they were going steady... Right up until a failed attack by the Soviets into Archangel totally unhinged their warplans, and left Moscow open to a German tank blitz! UK was forced out of India in the same round, (putting up an odd single fighter defense in the process), no match for Japan's East Indies focus. So Axis center crush in the 5th. But despite not planning for the worst case vs Karelian tanks, and going too light into Archangel (it happens, I've seen humans do the same), even still, I thought Allies did rather better than usual for an opener.

They blasted well with their extra Russian infantry, kept a portion of the royal navy alive with their extra UK subs, and made a nice push across North Africa and the med using the USA Atlantic transports they kept alive with their extra cruiser. Not too bad for HardAI Allies :)

hard_AI_Allies_vs_Elk,_12_ipc_each.tsvg

Also, check out this one. The bomber bonanza bid! haha...
I gave each of the Allies an extra bomber, one at Moscow, India, and Western US. I rather like how HardAI Allies used them. Axis had some tough rolls at the outset, so this one should be a nice uphill fight. Here is the situation on G2. Will give it a try tomorrow.
bomber_bonanza_G2.tsvg

Here is how the bomber bonanza resolved... on G9. HardAI Allies had a rather tough time managing the dark skies style approach from Germany. Axis bought a bomber a round with G, adopting the air armada strategy (no SBR, just combat bombers). HardAI Allies did their best to keep pace, but abandoned the center in the 8th round, allowing a walk-in on the Russian capital in the 9th.
bomber_bonanza_G9.tsvg

ps. one more vs hardAI Axis.
I think this game shows a fairly normal KGF attack plan by human Allies in early rounds (with limited goofs on movement or purchasing.) This is the sort of basic KGF play I'd hope to eventually see from an AlliedAI. Its interesting, right now I would say hardAI Germany does much better against the KGF, than hardAI Japan does against a KJF.  This is similar to human players who are first learning how to play, where a dedicated KJF, can paralyze and devastate the less experienced Japanese player, putting them into a position they aren't used to being in, and goading them into making poor decisions. This can lead to a rapid Axis demise, the sort of shock KJF aimed at knocking the Axis off balance early and then exploiting their goofs. Ultimately though, a KJF is much harder to execute against an experienced Japanese player, one who knows how to lure the Allies in and then play defense, while Germany goes monster. So I think this is another reason why KGF is optimal in 1942.2 and the one the AI should try to anticipate/emulate in most matches. Right now there is little downside to a KGF for a human allied player vs hardAI Axis, because Japan focuses so much energy into North America, you can basically contain them up in Alaska, while you press home against G. Takes a few rounds longer than a normal game should, but JapanAI allows you to drag things out without too much danger to the overall warplan against G. Here is such a game in the 8th round. Neither Axis capital is currently threatened, but with Russia holding at the center, it's just a matter of time before Germany gets squeezed out.
hard_AI_Axis_vs_Elk,_KGF_round_8.tsvg

By contrast when HardAI Allies are at the helm, they invariably over commit to Africa, pushing giant stacks into French Equatorial Africa, bouncing around all over the place with transports especially. You can see this in the games vs hardAI allies above, especially on non-com. As a human, the rule of thumb is usually to take Africa with the minimal possible investment, either by locking down Egypt early, or by a quick blast south to cover sub Saharan Africa and then redirecting. A human Allied player will do this either with UK holding Egypt initially with transport/air support, or they will do it in round 3/4 with USA striking down the coast at the Equator. A human Axis player will usually do the same, trying to snatch Africa quickly, either by cracking the canal with G or J using light forces, but then focusing the majority of their effort on Russia/India. Invasions of North America by Japan are usually meant as distractions, or to stall an Atlantic Crossing by USA to Europe, but rarely a full commitment by Japan as this is always doomed to fail in A&A. So for their part, an India/Suez focus for Japan, or a straight to Moscow focus, would probably be more in line with the way humans play against each other. Although on this last point, I will say, it's pretty impressive how Japan manages the Alaska corridor, they definitely draw the game out and makes USA more fun to play, but it gives India/Moscow way too much breathing room, and makes it rather easier for the Allies to control Africa and the Med than would otherwise be the case.
Here is the savegame above vs hardAI allies after another 8 rounds of gameplay. Once W. Canada is stacked with enough ground or aircraft to deter Japanese landings in Alaska they will start to adopt a more standard style of play vs Russia/India. It takes a good deal longer to set up an adequate North American defense than it would against a human, due to the way JapanAI relentlessly attacks into it, but once the West Coast fortress is established an Atlantic crossing from E. Canada is fairly straightforward, and the road to Berlin becomes much more direct...

hard_AI_Axis_vs_Elk,_KGF_round_16.tsvg
Here Berlin fell to the double hit in the 15th.

Oh and one last thing I've noticed that has been puzzling me for a while. In the opening round Germany often moves their battleship to sz 17 off Egypt. They frequently do this on non-combat too, and I can't figure out the reason why the machine thinks its a good move?

If it is a combat move this is usually just the HardAI hitting the British destroyer, but not closing the canal, which just invites the British to attack into the med. Even if the canal is closed, sz 17 would still be a dangerous location for the German battleship, since UK can sink it with an all air strike and land the bomber in Egypt or Trans.
Sz 14 would be a slightly better location for the BB (with a destroyer purchase to block optimal)
sz 16 at least allows German a leg up against the Russians for one move before the Battleship dies.
sz 15 allows Germany to funnel troops to Libya, and gives a chance of Battleship survival if the British destroyer and cruiser are both sunk.
But sz 17 is almost certain death for the German battleship! haha ;)
Here is a save showing the goof...
German_battleship_to_sz_17.tsvg

I've seen a couple variations, but the battleship frequently ends up in this position, either after combat or (like here) on their non-combat.
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Lasse i Gatan
In reply to this post by redrum
Have played two games with the latest update. Very nice to see how the naval part is improving.
Some aspects I like to mention:
- Note how my allies fleets often are following my fleets. Wether it is to help me defend my fleet, or seek cover by my fleet, that is something Redrum knows(?).
- AI fleets are likely to "dance" around each other. Like boxers not knowing when to deliver a punch. This is actually quite realistic! Human players (well, me and my friend) tend to anxiously stay near a factory and try to outbuild the other. Before steaming away to deliver battle/invasion.
Overall: good naval unit compositions. Good combat and non-combat moves. Invasion fleets sometimes have difficulty knowing what to do/follow through. But they are not stuck for long.

Purchase: good distribution and composition of land, air and naval. There is however something with my hexagon map that makes Green AI overbuild transports. Now and then I have used the Editor to remove some transports and given back the PUs.

Improved map and two savegames:
HexagonMap8.zip
Hexagonmap8_save6.tsvg
Hexagonmap8_save7.tsvg
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

redrum
Administrator
@Black Elk - Some good games. Didn't see any huge mistakes by the AI. The Japan AI does need to better understand the value of going for India and Moscow instead of trying for the Western USA. It does better in revised because there are better factory options in eastern Asia.

I checked the german turn 1 and the AI does like to attack to SZ 17 with its battleship. The reason is that only the battleship and bomber can reach it. The AI tends to prefer attacking if possible and decides its worth it even though it'll probably lose its battleship then. Its not a great move but not terrible either.

@Lasse i Gatan - Nice games. The AI in general tends to over build transport a little on most maps. It a difficult calculation to make and it tends to error on the side of a few too many transports especially later in the game.
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

feld
This post was updated on .
In reply to this post by redrum
Hello, I found TripleA a couple years back and have dabbled with it against the AI.  I'm very happy to see the upgrades to AI strength that are ongoing, and thought I'd test out your newest version on the NWO 1939 Lebowski map.  I played as the allies and the AI handled the Axis.  Hopefully I did it correctly.

The AI did seem to behave differently than it does in the current official release.  Smarter handling of the German fleet to start the game.  Willingness to build and move AA guns.  Built very few Elite units.  But there were weaknesses in the AI's game too (as well as mine - I definitely made many mistakes, forgot to do things and could have been more aggressive, but was still able to handle the AI with relative ease):

- The Germans take a turn longer to capture Paris than they really should.  Had I stacked it more I could have held them off for at least one more turn, simply due to them not being aggressive enough there.
- Germany gave up on supporting Finland much too easily.
- Germany (I think) never built any ships other than maybe in the first couple rounds, and never built a carrier.  They never did anything to seriously challenge the Allied fleets in the Atlantic, and made no moves all game to slow the influx of British and American troops into Spain via transport.  They never even bought many planes to threaten Allied ships.  They never bought a single Me262 all game.  This was a big weakness in their game.
- Italy also failed to do much with their Mediterranean fleet, never really augmented it and instead threw out single blockers that mostly just served to whittle down their fleet.  Did not buy any carriers, which would have served them well.
- Italy actually never seemed to accomplish, or even try to accomplish, much of anything.  They never seemed to have any sort of direction or focus.  Never challenged much for Africa.  Never did much against Russia.  Never pushed hard against Greece.  It left me wondering where they were all game.  They would really benefit from establishing some sustained effort in concert with Germany, beyond merely stacking troops in Madrid.
- Generally, the Axis tend to attack anytime they calculate that they are favored to win the battle.  But they don't seem to adequately consider the strength of a potential counter attack and the value of the units that could be lost to a counterattack following a projected victory.  This is what ultimately allowed Russia to break Germany.
- The AI has an extreme fetish for tankettes.  Buying some is useful, but often using them in place of infantry is a recipe for net losses in TUV, and the AI followed that recipe many times.
- The AI is very bad at handling mobile factories.  It attacks with them and leaves them in countries controlled by allies so they can't produce.

Hopefully some of this is useful info for you.  Cheers to you and great work on the improvements so far.

Lebowski_allies_end.tsvg
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

redrum
Administrator
@feld - Thanks for the feedback. I agree with pretty much all of your points and in general the AI plays much better as the Allies than the Axis on all NWO maps I've seen. Here is a simulation of the AI vs AI and the Allies roll pretty hard: test.tsvg

Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

captaincrunch
This post was updated on .
Alright this was another tough go at the latest ai.jar update. I hadto play 6 matches and won the 2nd and 6th match:




Match 1: I was the Allies and the ai was the Axis and I forfeited my 1st turn with Russia for fun and this is a hard scenario to win and the Axis ai rolled and the German ai held strong and took my Russian capitol with ease and I was overwhelmed with ai Navy and Airforce and I think I gave up around round 15.


Match 2: Rematch of Match 1 except I didn't forfeit my 1st turn with Russia for fun and I surprised myself and played well and beat the ai with ease taking the German ai capitol in round 4 and ended the match taking the Japan ai capitol in round 13;

latestaxisaidefeated20.tsvg





Matches 3-5: I was the Axis and the ai was the Allies and I used to think this was an easier scenario than when I would play as the Allies and beat the Axis ai but really trying to beat 3 ai countries is very hard and now I dread these matches against the Allies ai hehe. Mostly the biggest factor in winning is needing a good 1st battle against the Russian ai in Karelia in the 1st round with Germany. I had 1 battle where I lost every piece I used with Germany and the Russia ai also lost every piece in the Karelia battle but I found out that resulted in benefitting the Russia ai in the end. I had a hard time containing the UK ai and USA ai Navys! 1 Match I gave up maybe round 14 and another I gave up maybe round 18 and the last loss I think I went over 20 rounds.



Match 6: Rematch of the above 3 matches and I finally had a good 1st battle win with Germany in Karelia and it was a huge factor not losing any Planes and I made sure to keep a German Airforce and that kept the Allied ai Navys back and I constantly battled the USA ai's Navys and the UK ai's Airforce and the UK ai played so well that I actually hadto take the USA ai's capitol before the UK ai's capitol. I started out by taking the Russian ai's capitol in round 6 and then finally took the USA ai's capitol in round 25. I finished the Match taking the UK ai's capitol in round 27;

latestalliesaidefeated20.tsvg





@redrum, ya the Navys are very tough now and excellent variety of purchasing and the Fighters on Carriers makes it that much more harder to defeat. I only noticed something that I don't have saves for that I believe happened in the update before this and also in this update in a Match I lost ... twice when I battled an ai Navy ... the ai was left with 2 pieces in the battle which were 1 Fighter and 1 Carrier. Two times I would successfully hit the ai's pieces and the ai WOULD CHOOSE TO REMOVE THE CARRIER OVER THE FIGHTER. the 1st time it did it, I went and checked the Game History and the ai would find a place to land the Fighter BUT in a recent match when the ai had no place to land the Fighter well then basically my 1 hit killed TWO ai pieces because the ai should have removed the Fighter BEFORE it removed the Carrier. That mistake denied the ai from continuing 1 more round of fighting and getting 1 more Defense Attack Roll with the Carrier. I think both times the ai (USA ai or UK ai) had its own Fighter on a Carrier and not an Allied Fighter.


Thats just something to watch for is all I could think of. I don't think the game gives a "Military Victory" if you win against the Allied ai if you want to check that 2nd gamesave ...  oh and I will stop using Strategic Bombing unless the ai starts using it because I saw you mention it in the others gamesaves but anyways this ai is quite tough now and the Matches are taking longer and longer!!


Was fun and look forward to the next!

Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

redrum
Administrator
@all - Pre-release updated:
- Lots of carrier fixes including not moving allied planes with carriers
- Fixes around capital defense
- Fixes around planes not blocking enemy movement
- Fixes around transport movement
- Enhanced strategic territory value to value territories towards middle of map higher

@captaincrunch - Nice games. I haven't seen the remove carrier before fighter issue so if you have a good save game please upload it.
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

captaincrunch
partypartyparty

ok if I see it happen again I will save it or anything that may be fixable. Will get back to you when I beat the Axis and Allied AI update.

Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Black Elk
This post was updated on .
In reply to this post by redrum
Played an enjoyable game against Axis AI latest jar this afternoon. They fought hard and maneuvered pretty well despite a vigorous KGF triple team, right up until the dawn of the 12th round. The IJN was smoked in the Med with a cutthroat American air strike, just after Japan had completed the conquest of Africa on the ground. D-Day is pretty well set up on Berlin now, fun stuff. Great job Axis AI for keeping things engaging.

Axis_AI_round_12.tsvg


I've noticed that the best way to encourage HardAI Japan to fight more like a human player, is to stack W. Canada/Alaska early on. This seems to push them more towards a UK oriented campaign, which gives Japan considerably more income than the endless North American push does.

With the last two jars the AI is doing very well when it goes bomber heavy! Even if the AI nations don't currently engage in SBR with their bombers, they use them very effectively in combat. This holds for both sides :)

I think the bomber bids for the underdog might be optimal way to go for fun a time, human vs the machine. For example, if playing against AI Allies, I like giving each ally an extra bomber. 1 extra bomber in Moscow, 1 in Calcutta, and 1 in Honolulu for the AI Allied bid. Here is a pre-set game for the bid, with the extra bombers in place but before round 1 begins, in case anyone likes to give it a go vs the computer...

Pre-set Victory City Bomber Bid, save game for both sides:

AI_Allies_Bomber_Bonanza_Bid_-_Russia,_India,_Hawaii.tsvg


AI_Axis_Bomber_Bonanza_Bid_-_France,_Philippines.tsvg

This seems to set off a nice air-focused game, where the computer does a lot of fleet hunting and makes bolder attacks in general. ;)

Here was a great one I played just now, again with the new jar, where HardAI Allies bid the bombers. They rocked it well into the 12th round, with some cool pacific plays. AI Allies made several bold attacks early on, and Japan got lucky on UK2, Bushido dice to nail the pair of british bombers in a single engagement, destroyer defense. Wild! But that didn't deter AI Churchill, he bought yet more bombers and started creeping on Germany. USA fought a decent late game Pacific war and flexed against Europe several times for the duration, grinding away against Africa in overwhelming numbers. Allies stacked Moscow till the bitter end, and when all hope was lost, Stalin escaped the Russian capital in his big red Bomber... all the way to Borneo!!! haha

Fun showing by the AI!
AI_Allies_bomber_bid_round_12.tsvg


Here is another game just played tonight against AI Allies the triple bomber bid. Moscow collapsed much earlier here, when the Russian AI was brutally turned back in W. Russia, but again they were able to keep their airforce in tact and send the bomber and crazy missions for several more rounds. I played it to the 12th as well, just to see how things would look. Despite getting pushed into a corner, the Western Allies were fielding some impressive TUV by the time honorable victory was achieved...

AI_Allies_second_game_bomber_bid_round_12.tsvg
Another pretty decent run for AI Allies with the bombers added to their arsenal. Even when they get trounced down early, the HardAI can maintain pretty steadily now that they're buying carriers, doing stronger fighter transits, and air strike tactics on the water. I dig it :)

Here is a game vs AI Axis with the double bomber bid (Paris, Manila). It was a pretty glorious fight. HardAI Axis fought down a KJF and snatched up Moscow in the 10th round! Japan is still holding onto the home island and Germany is fighting for control of Asia by land out the captured Russian capital. Nice work AI Germany :)
AI_Axis_bomber_game_round_11.tsvg
AI Japan fought for 18 rounds into the deep endgame! before finally being cracked by USA on amphibious. Super Germany has been holding down the center vs Allied island production for some time now.
AI_Axis_bomber_game_round_18.tsvg
Nice to see AI Axis grind it out. Germany fought at the center for 26 rounds! Still holding Moscow when the curtain drew. Brilliant!
AI_Axis_bomber_game_round_26.tsvg

More than anything, lately I'm starting to think more long term, about how to design a custom 1942.2 style scenario specifically with the HardAI's playstyle in mind. The way HardAI Japan fights against North America could be turned to a real gameplay enjoyment advantage, creating the sort of pacific interest that I think A&A players have long sought after. A viable way for Japan to fight into America, when America ignores the Pacific entirely. I think perhaps if the ipc/production spread was a little different it would really make that type of game more interesting. I've long felt that Hawaii could use boost in ipc value, perhaps a starting Industrial Complex in Alaska or Hawaii for USA, or a 1 ipc boost in either of those territories. Things like that would allow for a more robust campaign Japan AI vs USA. As it stands they tend to get stuck up there in Alaska, and tied down in multi-round amphibious invasion loop. Basically I think we could make unit set up or map tweaks to a 1942.2 style gamemap, and cater the scenario towards the AIs strengths. This once the Hard AI is completed of course, but I think it would be a cool to make a custom World War 2 tripleA game, that would really showcase the newly developed AI, where we build a new game around the AI instead of the other way round haha. Might be fun
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

captaincrunch
@redrum, I got the ai to select the Carrier over the Fighter possible bug again!;

aitest21round11westspainseazoneselectionbug.tsvg

I played 3 matches so far and won the 2nd against the Axis AI but will post the wins when I finally beat the Allied AI which is being very tough right now.

Good luck with the next update!

Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

captaincrunch
OOk, I finished play testing the latest ai.jar update and played 5 matches and won match 2 and 5.


Match 1: I was the Allies and AI was the Axis and I forfeited my 1st turn with Russia for fun and this is a hard scenario and I never was really in it and I got crushed and maybe quit at round 10

Match 2: I was the Allies and the AI was the Axis again but this time I didn't forfeit my 1st turn with Russia for fun and I played well and took the German AI capitol in round 7 and ended the match taking the Japan AI capitol in round 14;

latestaxisaidefeated21.tsvg





Match 3 & 4: I was the Axis and the AI was the Allies and this scenario is getting tougher and tougher and taking longer and longer. Match 3 I didn't play great and I'm pretty sure it all came down to my very 1st attack against the Russian AI in Karelia where I hadto retreat and I also didn't play Japan great and so eventually I was overwhelmed by the AI Navys and Transports and I gave up maybe round 15. The next match was a rematch and same result but I did a little better but I hadto give up maybe in round 25 ...


@redrum in Match 4 I spotted the AI choosing a Carrier over a Fighter again and thus resulted in me getting 2 kills in 1 shot so its the same possible bug and I'll submit it if it helps along with my other gamesave with the same bug which you are currently working on;

aitest21round14westspainseazoneselectionbug.tsvg




Match 5: This is a rematch of matches 3 and 4 and this time I got a good battle with Germany against the Russian AI in Karelia in round 1 and that set the tone in my win where I take the Russian AI capitol in round 6 then I finally take the UK AI capitol in round 15 and then end the match taking the USA AI capitol in round 21;

latestalliesaidefeated21.tsvg




Ok, was a blast and nothing stood out that I think was needed with this AI since I have faith in the devs working on it and look forward to the next update!
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

ZjelcoP
RedRum,

Great job on the AI. Regularly play with latest patch.
Carrier use is working great now, landing allied fighters and all.
Last game i noticed the AI was a bit overly fond of carriers, buying more than it could fill with fighters.

worldatwar.tsvg
File shows USA slightly overbuying. UK with 10 carriers, 8 fighters.

Cheers Zjelco
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Black Elk
This post was updated on .
In reply to this post by redrum
OK just got in another game v5 against the machine. This time instead of a bid, I gave hardAI Allies a bonus to income of 10 per round.

AI UK created a massive bomber armada, which it used to wipe the IJN, and send the entire Japanese fleet to the bottom of the sea.

Unfortunately for Churchill, the AI wasn't able to coordinate a decent defense for these bombers after they landed in Burma (this despite having a large ground stack in India and fighters that could have helped to prevent a Japanese counter attack.) Otherwise it would have been a spectacular trade!

Moscow_crush.tsvg

Here is a follow up game, this time vs HardAI Axis, with a flat rate income bonus of 10 ipcs.

HardAI sacked Moscow with Japan, and achieved Honorable Victory in the 18th round. Well done HardAI!
HardAI_Axis_10_income_flat_rate_round_18.tsvg

Another game against HardAI Axis, this time with 25% bonus to income for the AI. I think I enjoy the percentage bonus a bit more than the flat rate one.
HardAI_Axis_25_percent_income_bonus.tsvg


ps. this game vs HardAI Allies with the 25% income bonus has been pretty exciting. USA sprung a nice little invasion on the Med seizing Rome, while Germany was busy establishing a new colony in Australia hehe...

HardAI_Allies_25_percent_income_bonus.tsvg

In this game HardAI Britain allowed for a pretty devastating hit on their main transport group in the 7th round, leaving 5 transports naked, exposed to a merciless airstrike by Germany. Despite UK's goof, the Americans are doing rather well with the bonus, and Russia has been fighting like a big red Bear!

pps. The game concluded after 18 rounds, with HardAI Allies putting up a pretty tough fight. They faced down invasion USA from Japan, and even saw UK expand production in E. Canada. Somehow Allies managed to stay in it, and deny the Axis the necessary VCs to steal a technical win for some time. Not sure if that is in the logic yet, but it will be interesting to see how the AI performs if those sorts of objectives are factored in. Here Axis achieve the honorable victory, with Japan all the way to the Rockies and holding San Francisco for a whole round.

HardAI_Allies_25_percent_income_bonus_round_18.tsvg


Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Lasse i Gatan
In reply to this post by redrum
Have played using the latest update.
Would like to report on a one thing:
- In round 8, AI Yellow moves to 2 bomber and 2 fighters, to attack two defenseless transports. Only one unit was necessary for the attack.

Apologize for uploading yet another version of the map. I am tweaking to reach the right balance and fun.

Hexagonmap8_save9.tsvg
HexagonMap8.zip
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Black Elk
This post was updated on .
I've noticed something similar, where the machine sends more units than would be necessary to kill a defenseless unit.

Had another chance this afternoon to play against HardAI Allies. This time I gave them a 33% income bonus, and went hard on them, with a no mercy Moscow crush strategy. G cracked the Soviet capital in the 8th round, but not before Western Allies set up some interesting endgame options of their own.

I think this income boost at around 1/3rd of the AI nation's total, is at the level of a genuinely "hard" experience.

For percentages ~10% seems about normal, i.e. the computer will play with the difficulty level of middling human overall. At ~25% the machine starts to present some definite challenges even to the experienced player, gaming more like a seasoned vet, who just goofs on occasion.

At ~33% bonus, you're right on the edge of where the income advantage will start overcome whatever disadvantages the computer might have in terms of it's current play logic. It's pretty hard to beat the machine on this setting, but the AI will still make mistakes that can be exploited. Right now these are primarily in leaving their defenseless transports exposed, or when moving their air defenses, getting tripped up by the turn order sequence. As was the case in the game below when Moscow fell...

HardAI_Allies_33_percent_income_bonus_round_8.tsvg 


Here is the same game a few rounds later. The Western Allies are managing pretty well into the deep endgame, with a massive naval advantage. HardAI Britain has wisely expanded their production, and built a factory in Egypt!

Slick Plan HardAI! ;)
HardAI_Allies_33_percent_income_bonus_round_12.tsvg







Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

captaincrunch
@Black Elk hey when you say Income Bonus do you mean just to start OR for every round they get +33 more Income??

33% Income Bonus to an AI Country to start would help them but ya a 33% Income Bonus to an AI Country every round would be a super hard challenge to beat I agree hehe.
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Black Elk
This post was updated on .
The bonus applies every round, based on whatever the AI's total income is at the begining of their turn. You can select this in the game options tab. Same place where you would enter a bid.

The section dealing with AI bonuses is on the bottom right.

I find I like percentage better than flate rate lately. It gives a little more incentive not to let the AI grow too powerful and start to get away from you. Flat rate at above 10 ipcs was pretty hard. I think Percentage is the way to go. It seems a bit less distorting than a bid, and lets the AI build out its force on their own terms haha.

I think it would actually be pretty interesting and useful if this "income bonus" feature could be added to any individual nation/player separately. Instead of as just an AI feature.

It would be helpful in many A&A style games to have such a feature (this can be achieved via edit mode, but would be easier as a game option), where for example: Russia, or Anzac, or UK Pacific might benefit from a recurring bonus, even among two humans playing each other. Or if as the AI, you could choose different flat/percentage bonuses for different nations.

ps. Here is that game from above, still going in the 14th round. Axis are finally starting to recover position on the water, mainly via airstrikes from the Eurasian stronghold. Gaming against the AI has convinced me of the overall power of fighters and bombers vs Naval in A&A. Basically the only way to keep fleets off your rear, is to build bombers and then build fighter fodder to cover those bombers ;)
HardAI_Allies_33_percent_income_bonus_round_14.tsvg

pps. Check out this crazy game just started vs hard AI Axis, using the same bonus at 33%. Given their stronger starting position, AI Axis are particularly challenging with the bonus at this level. I decided to see how the AI would respond to all out slam against Japan in the first two rounds. Everything, plus the kitchen sink! Just to see how red we could make Asia. But even after losing the majority of their empire, Japanese Bombers and battleships are still murdering in the Pacific!

G even snatched India on a sneaky amphibious maneuver while Allies were distracted with their nutso Pacific plan haha. Fun stuff!
HardAI_Axis_33_percent_income_bonus_Crazy_KJF_round_6.tsvg
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

captaincrunch
@Black Elk ook well I find the default AI hard enough hehe. I get your point about it though and how you can get a feel for its identity. I just wondered exactly how you meant the bonus but I see you are having some interesting matchs. I do like the Airforce strategy and its cool how Redrum and the devs are fixing up each part of the AI's game piece by piece and the interesting development of this game's logic.


1 ... 36373839404142 ... 63