AI Development Discussion

classic Classic list List threaded Threaded
1250 messages Options
1 ... 29303132333435 ... 63
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Black Elk
downloaded the latest jar and tried to play out a game a few different times, but alas, the issue seems persistent. The other day I was able to play well into the endgame, like 15 rounds or so to victory against the Ai, but now it seems to lock up about the 3rd or 4th round in, straight to black. Haven't been able to determine the cause. And then when I reboot, and reload the save instead of throwing off an error like before, it'll go for a few more turns, but then same deal again. Tried starting a new game each time just to make sure, though not sure exactly its going down. Seems determined to stall up my Allied advance tonight haha.
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

redrum
Administrator
@Black_Elk - Unfortunately, crashes are much harder to debug since I don't have any info on the reason. I'm guessing its memory related. Can you post your system specs? Also make sure you don't have anything else running on your computer when you are playing TripleA. I'd also recommend reinstalling TripleA if you keep having crashes to make sure there isn't any issues with your current install.
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Bungo
Great work with the hard AI! It's a whole different thing playing with it than the other AIs. I even care about my dice rolls, which is a great sign that the game is tense enough even if it obviously still has lots of room for improvement.

One thing I would like to see is that it places more value on keeping its own capital. I don't know if it understands the naval bases giving +1 movement to transports at all, but at least I have seen both that it leaves its capital very exposed to naval assault and that it has foregone taking another hard AIs capital by amphibious assault (both giving 100% chance of attacker victory without any significant sacrifices). I also once saw it give up a capital to a land attack, even though it could have saved by just moving every unit it had nearby home (a plan which would have lost every other territory nearby, but kept income and production capability for at least one more turn, and avoided sending off the treasury - at least to me it looked like a mistake, though I can't say 100% sure).
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Eschelon
In reply to this post by redrum
redrum wrote
@Eschelon - Interesting. You could also argue that the engine shouldn't even call the AI if the AI can't actually do anything. Either way I can definitely look to reduce purchase time for AIs that no longer can purchase/place anything. If you have a save game showing it then please upload so I can test with it.

It has methods to try to spend as many PUs as possible though there could be some edge cases that I haven't considered. If you have a save game showing it please upload and I'll take a look.

@Black Elk - I believe I just fixed that bug in the pre-release I uploaded today. Please try the latest and see if you still have the issue.
OK, if I remember right the game.xml is included with the save game (I made some changes to it).  Let me know if this works for you (I eliminated the sea zone capitals in the game.xml for this save).

aaafeb102015.tsvg

You'll want to look at the Seleucids, around turn 20 (note they have 116 unspent PU's, and the game is taking a bit to calculate the build phase for the Seleucids).  the Seleucids only had a couple of territories left prior to losing their capital, so they were only generating around 20 PU's per turn.


Doing the 'can't build' check on the Engine side probably makes a lot of sense, as that would apply to all AI's, not just the Hard AI.  With some maps, an Ally might liberate a capital, hence giving that AI player a new shot at life, so I'd think that said check would need to be made each turn.
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Black Elk
This post was updated on .
Yeah its bizarre. First time I've seen a crash this consistent when playing tripleA.

I uninstalled and reinstalled java Version 8 Update 31
Uninstalled and reinstalled TripleA _1_8_0_5  grabbed the latest Jar from the prerelease.

But now I'm lucky if I get out of the first round before it drops to black on me.

Its on older laptop by a few years, but usually runs tripleA just fine.

Intel(R) Core(TM) i5 CPU M 450 @ 2.40GHz
Video Card NVIDIA GeForce 310M
Memory 4.0 GB
Operating System Microsoft Windows 7 Home Premium Edition Service Pack 1 (build 7601), 64-bit

Only programs I'd ever have running in the background would be like MS security essentials, google drive, or Steam. But even when I kill those before launching the issues still seems to crop up. Any other details I could provide. Would love to get back into the fight against this cunning AI. I enjoy watching it perform, and teasing out its various personalities hehe.

Curiously, if I allow the HardAI to just play itself, I'm able to able to generate a longer game without any crash till around round 8. Tried it just now with cpuid running to see if I was overheating, or thought maybe it might be this shoddy samsung battery, which no longer holds a charge, but I'm plugged in the whole time. Still cant figure it. Anyhow when its HardAI vs HardAI, things seem to truck along for a while. Just when I am taking over control of one side it craps out on me faster. Wonder if its something I am doing, or something with v5? Or just throw this sumsung out the window haha
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Irinam
@Black_Elk:
Have you tried different maps?
Perhaps try V2,NWO, LOTR ant 270BC with many HardAIs...

And have you tried playing with only one HardAI player?
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

captaincrunch
This post was updated on .
Alllright too fun. I tested your ai.jar update from the 12th. I can't believe that I kicked some butt and can't say if this ai.jar is better or worse because I'd haveto play the different updates many times to know for sure but here are my saves and hope they help you see how this current ai.jar performs. I played 2 matches and won both. The first game the AI placed planes fine but in the 5th round the German AI lost a battle for its capital in an 8 on 8 battle. You should check that out see how you think the AI should have spread around its pieces the round before. The 2nd match I picked off the AI's planes where they landed in round 9 and you asked to give saves to show how the AI is placing planes so check it out.





1st match: I was the Allies and the AI was the Axis and for fun I even forfeited my 1st turn with Russia and amazingly I played best ever game I've played and took Germany in round 5 and the game stretched out about 21 rounds when I finally took the Japan AI capital to end it but I had balanced all my country attacks and plus the German AI lost its capital in round 5 from its piece placement so was an interesting match to check out and see how the AI played. Game went about 2 and half hours. Not sure if I could repeat a match that well like I just did so don't tweak the AI too much.


latestaxisaidefeated10.tsvg





2nd match: I was the Axis and the AI was the Allies and I didnt even really take over Africa in the match but was still swift and took the Russian AI capital in round 6 and then I took the UK AI capital in round 15 with Germany and then game ended taking the USA AI capitol also in round 15 with Japan. Redrum or whoever you should see round 9 where I attack with a bunch of German planes against the UK AI that put 7 bombers on Gibraltar and they only defend at 1 so was an easy wipe of the UK AI's planes;


latestalliesaidefeated10.tsvg





OOk, was a blast those 2 matches and look forward to the next AI update! I still wonder if you developers will give a "final" AI patch for the main game download or will use the forum to constantly change the AI because I can't say what AI will end up being the best because it takes lots of repeated matches to know but still fun to test and keep updating!
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Black Elk
In reply to this post by Irinam
@Irinam

Haven't had a chance yet to try all that many maps. I fired it up to play on some older maps I made, Pact Of Steel, the Great War and Domination (Dom especially I had always hoped might be able to utilize an Ai). Checked them out briefly since I thought the AI might be novel there. But those maps all use the old style transports. Playing Classic and Revised the AI did quite well, since the unescorted transports still have a defense value there. But I'm a lot more interested in the smaller scale, newer boards. Specifically v5 and v6, but any of the world war II games post v3.

Haven't tried playing against a single AI opponent yet, since I'd have to take over one of them as a human and play against myself, which is less informative from a testing standpoint. Also haven't tried 2 humans vs AI yet either, though I'm interested in trying that as well.

I think for v5, Axis AI vs two or more Human Allies would probably be optimal. Since the need for Allied coordination and the challenges of it would likely balance against some of the AIs weaknesses or exploits like unescorted transports or bombers parked on the ground. Those are the big ones for me, since it's easy to nuke a lot of TUV if the AI goofs their non com movements.

Still crashing though, so for now kind of stuck waiting to play again haha. Look forward to getting back in there though
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Frostion
This post was updated on .
In reply to this post by redrum
I have also been play testing the latest Hard AI with my Dragon War map (Work in Progress). It’s great, but there are some minor issues:

• Does the AI value a unit that gives +/-1 A AND D. support to friend or enemy more than it values a mere +/-1 A OR D to friend or enemy? By the looks of the AI shopping on my map, it seems not to. However, it is hard to tell.

• I got some units on my map that support 10 allied units, not just 1 like artillery,  but 10. It doesn't look like those units act like they know that they support 10. They sometimes walk around and even attack with a single allied or alone. Do units currently act the same regardless of how many they are set to support?

• Bungo said: “I also once saw it give up a capital to a land attack, even though it could have saved by just moving every unit it had nearby home”.

I have noticed the same phenomena. Sometimes the AI players leave their capitals nearly unprotected. An enemy attacks via sea and captures it. In the same or next round, even if the player who just lost the capital has a fair chance to take it back, it does not seem to try. It just stacks up by the capitals borders. I think it would be great if the AI reacted more desperately/aggressively if it lost it’s capital, like going for an “all or nothing”, even suicidal, attempt to recapture its capital. But, I guess capitals are not of the same importance on all maps.

• I have (as you have seen, redrum) a Gaia-like player called “AI” on my map. It gets huge amount PUs throughout the game as it starts out owning a lot of land. All its units are like normal units/stats but costs x5 (aka 40-60 PUs). “AI” will also build these units from the “AI” factory units. No problem.

But the “AI”-player will never build new factories/villages, even if all its factories have been destroyed. I have tried to lower the cost down to 1 PU, but still no build. I have even seen it lost all factories, owning 3 PU territories, having 150 PUs in the bank, and still not build. On a side note, the other Hard AI players on the same map WILL build their factories. This is an example of one of “AI”s factories, and the corresponding unit that can be build there:
(Maybe it’s because only one type of unit can be build at each of the “AI” factories? Maybe because “AI” feels it is not safe / too risky to build? … even if it has no other factories?)
XML wrote
        <attatchment name="unitAttatchment" attatchTo="Barbarians" javaClass="games.strategy.triplea.attatchments.UnitAttachment" type="unitType">
            <option name="maxBuiltPerPlayer" value="10"/>
            <option name="attack" value="4"/>
            <option name="defense" value="3"/>
            <option name="movement" value="1"/>
            <option name="requiresUnits" value="Barbarian-Village"/>
        </attatchment>

        <attatchment name="unitAttatchment" attatchTo="Barbarian-Village" javaClass="games.strategy.triplea.attatchments.UnitAttachment" type="unitType">
            <option name="maxBuiltPerPlayer" value="3"/>
            <option name="isFactory" value="true"/>
            <option name="isInfrastructure" value="true"/>
            <option name="canBeDamaged" value="false"/>
            <option name="canProduceXUnits" value="1"/>
            <option name="destroyedWhenCapturedFrom" value="AI"/>
        </attatchment>
My newest xml (if needed):
Dragon_War_WIP.xml
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 Irinam
Hmmm curious. I decided to try another game vs AI in Revised to see whether it would crash. So far pretty far in without an issue. I wonder why? Running v5 it has been crashing on me much more consistently. Between the two version Revised to 1942.2 the main changes unit wise, are Factory repair, AAA guns as a combat unit, defenseless transports no hostile sz. Any chance a hang up with something like that my be crashing me?

Also, here's an interesting situation in Revised, that leads me to think Redrums earlier mention of greater AI coordination over the turns (or within the gameround) might be good to slate top priority.

I was lazily playing Axis at no bid, just how hard AI would gun at the center. And this interesting strategic decision arose. In this game, post Moscow/Center collapse to Axis (and especially when losing Moscow to Japan). I'm curious how Hard AI will approach the capital trade

I don't know, battle calc would tell you no way on a Berlin hit. But without taking the German capital right now, before Axis stack up yet another round behind Japanese fighter support the chance to do so is probably lost for a same round capital trade.

I like hard AIs endgame. Wish I could figure out the v5 thing though. I feel like I've done revised to death heheh. Though its nice to see the HardAi working quite well at it.

Hard_AI_Revised_capital_trade_goof.tsvg

And then a few rounds later... in the aftermath of Moscow collapse. You can see even though US/UK HardAI is doing a pretty good job maintaining income and naval parity, in the capital trade game, they are fairly well locked off Berlin.

Hard_AI_Revised_capital_trade_goof_round_14.tsvg

Well at any rate. Hard AI is enjoyable with Revised. I'm going to try v5 again and see if it persists in crashing. I'm curious what the average gamelength in rounds would be when the AI plays itself, or how long its planning to prolong the war haha

Also, just tried running HardAI against itself in Revised to see how long I could keep it going. Got several rounds in, but then crashed out at this point.
autosave.tsvg
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Black Elk
This post was updated on .
Yanked the battery from my laptop and pressed on. Could be the crash was just a power issue and unrelated to tripleAI. So far so good! Another v5 trounce down. This time USA took Berlin at the close of the 12th round, on a KGF stack fest...

This time I wanted to see what the HardAI would do with their ships in sz19, with Axis in control of Panama.

I decided to let Japan take the canal, to see if they might it exploit it. Let them hold mexico for a round. They did immediately move to Panama, and I got excited. Let them hold Panama for a round also, but alas, they never moved their fleet to the Atlantic! Instead they just withdrew and kept stacking Alaska. Had they seized the chance, they could have launched a large fleet through the canal, taken Brazil and messed with USA's logistics, which might have been cool.

Instead the Americans take Berlin with a 20 stack hehe

HardAI_Axis_v5_round_12.tsvg

ps. Got in Another KJF v5 just now. Able to ice it in the 8th round. Demonstrating again the Japanese vulnerability to air attacks against their transports.
Hard_AI_Axis_v5_KJF_round_9.tsvg
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

redrum
Administrator
@Bungo - Thanks for the feedback. If you have any save games showing poor AI decisions please attach them to a post here so I can take a look.

@Eschelon - So I added some logic to try to use extra PUs for defenders if there is nothing else to safely purchase. I also added logic to exit purchase logic if the player doesn't have any valid place territories left.

@Black Elk - Glad its working for you again. There shouldn't be much difference in memory usage or processing power used for V2 vs V5. Hopefully it was just the power issue. If you get more crashes without error messages then you might want to try adjusting TripleA's memory.

@captaincrunch - Nice games. Looks like you played well and got some good rolls as well. I'm starting to work on the planes defenses so hopefully that shouldn't happen as frequently.

@Frostion - Responses:

• Does the AI value a unit that gives +/-1 A AND D. support to friend or enemy more than it values a mere +/-1 A OR D to friend or enemy? By the looks of the AI shopping on my map, it seems not to. However, it is hard to tell.

- So the AI doesn't give any value to enemy support but it does value both attack and defense support for allies. So if you create identical units but one gives attack and defense support to allies and one that only gives one or the other you should see it prefer the one that provides both support.

• I got some units on my map that support 10 allied units, not just 1 like artillery,  but 10. It doesn't look like those units act like they know that they support 10. They sometimes walk around and even attack with a single allied or alone. Do units currently act the same regardless of how many they are set to support?

- So for combat move, it calculates how many units will be able to receive support when choosing which units to attack with. For non-combat move, units are essentially all treated the same but should sort of move in the same direction towards what the AI identifies as the best target.

• Bungo said: “I also once saw it give up a capital to a land attack, even though it could have saved by just moving every unit it had nearby home”.

- I'd need to see a save game to determine why. In general, the AIs try to defend their capitals if they have any chance of holding it. Most likely either the AI thinks it has no chance to hold it or it didn't calculate that the enemy could attack it.

• I have (as you have seen, redrum) a Gaia-like player called “AI” on my map. It gets huge amount PUs throughout the game as it starts out owning a lot of land. All its units are like normal units/stats but costs x5 (aka 40-60 PUs). “AI” will also build these units from the “AI” factory units. No problem.

But the “AI”-player will never build new factories/villages, even if all its factories have been destroyed. I have tried to lower the cost down to 1 PU, but still no build. I have even seen it lost all factories, owning 3 PU territories, having 150 PUs in the bank, and still not build. On a side note, the other Hard AI players on the same map WILL build their factories. This is an example of one of “AI”s factories, and the corresponding unit that can be build there:
(Maybe it’s because only one type of unit can be build at each of the “AI” factories? Maybe because “AI” feels it is not safe / too risky to build? … even if it has no other factories?)

- So the AI will only place factories if it deems territories are safe and won't be lost in the next few rounds. I would need to see a save game where you expect it to place factories and it isn't to determine if its a logic thing or if there is some issue with the factories.
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Black Elk
This post was updated on .
Woot! New battery in the mail, tripleA battle against the hard AI running steady.

Also Redrum, check out this thread on the AA.org boards....
http://www.axisandallies.org/forums/index.php?topic=35264.0

I was trying to give this guy a hand, since he has been playing against your HardAI in v5 and getting consistently rocked. In frustration he asked for some advice, so I did a play by play with savegames. Goes up through the mid-game 6th round. I tried to highlight some of the exploits against HardAI on a KJF.

Also...

Finally a game of v5 where Japan actually moves through the Panama canal!!!

I've played a lot of games but this is the only one where HardAI has successfully moved their massive Panama fleet from sz19 to the Atlantic. This time they brought a tank into Panama and then launched to nab Brazil, moving the fleet through on Non Combat. This was highly gratifying to see, as usually the HardAI just stacks ships in sz19 endlessly to no purpose. This time it took them 11 rounds to finally wise up. More behavior like this earlier on would be cool! ;)
Japan_Through_Panama_Canal_Finally.tsvg

The following round, Japan moves their fleet even further towards Gibralter! Success! Now, had Japan been doing this sort of thing from the get go HardAI might have come out on top with their crazy campaigns to mess with America hahah. This took some definite baiting on my part, but once the move to the Atlantic is made, the Japanese HardAI seems to behave as though it had a purpose once again...

Japan_Through_Panama_Canal_to_Gibraltar.tsvg

Japan_Through_Panama_Canal_to_North_Atlantic.tsvg

Check out this round 20 grinder... Probably the best showing I've seen yet from HardAI Japan in the endgame. Just a total stackfast haha. Clearly not the easiest route to KGF, but it does kind of demonstrate the power of the Karelia for the Red Turtle, and just how hard it is for Japan to mess with America. Here Japan has basically swept the entire globe. G stacked it magnificently, but failed to take Baltic in a battle they probably should have won, and then forced to retreat to Berlin for massive crunch on the 1-2 punch UK/US. I just wanted to see if there was any way Japan could be coaxed into moving Atlantic. This is the only time I've seen it happen in v5

And take a look at that Atlantic fleet blocking manuever Japan HardAI tried to pull off in round 18. Clever play. It came very close to working. Nice work HardAI!

KGF_stackfest_round_20.tsvg

What would be ideal I think for HardAI, since they gun so hard after N. America, is some logic that makes Moscow a priority, or encourages them to sail around S. America rather than trying to open the canal (since the latter is fairly easy for US to protect) unless the player just leaves it wide open like I did here. Germany got nailed on a hail mary drive. But who knows, Japan HardAI may yet prevail, even now!!! hehehe They have the line on Moscow and the strong TUV advantage!

Obviously its just masochism to stack this deep as Allies and focus so one dimensionally on the north, but this seemed the only way to goad them into transporting into central america. HardAI plays it very close when you get into the massive wall management on the Eastern Front. Just huge

KGF_stackfest_round_20_Japan_controls_Italy.tsvg

Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Black Elk
This post was updated on .
The thing I find most entertaining about HardAI Japan in v5 is how they don't ignore North America and USA the way most human players will. Its novel, but HardAI is going have a tough time beating humans with this approach, because it doesn't seem to appreciate the overall importance of controlling the center and destroying Russia for the endgame.

Some kind of logic that tells the machine to gang up on single enemy nation and devote the majority of their resources in that direction to "double team" the chosen enemy nation would be helpful. Many human players will go 100% in one direction and then turn around, but that leads to kind of static one dimensional games. Maybe split on some kind of random percentage like 80/20 or 70/30 or 60/40, where the larger share of IPCs in the pile goes against a single enemy Nation, and the remainder left over is spread against the other opponents? I don't know that its necessary that they always try to team up on the weakest opponent necessarily, but that might be cool too.

Often times HardAI doesn't press the advantage when they have a nation on the ropes and close to the breaking point, but will ease off and send resources somewhere else. Perhaps an SBR focus could help too, since HardAI doesn't currently use their bombers beyond combat.

Here you can see Japan has 5 bombers at the ready, which should be enough to neutralize Russia, but alas they never bomb haha, and when Germany gets squeezed out, Russia is able to direct their entire purse against the Japanese...

HardAI_Japan_KGF_round_13.tsvg

And here is basically the reverse situation where G can't quite seize the initiative on the center, despite taking the lead in income and TUV. Again some SBR against Moscow would probably help.
HardAI_Germany_KJF_round_13.tsvg

Another different KGF showing how it sometimes makes sense (as a human) to sacrifice air to take a capital. Something HardAI doesn't really risk. Here UK makes the first hit, then USA goes in for the kill, with Russia to back them up just in case. Although here Stalin was disappointed as Ike got in there with 1 artillery unit, after dropping a whole bunch of bombers haha

HardAI_KGF_round_11.tsvg
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Black Elk
This post was updated on .
One other thing I just noticed. I don't think HardAI uses it's blitz units to take undefended territories and then return to safety when given the opportunity. A few times now I've seen them take a territory with armor, that could have been captured on the blitz without putting the armor at risk from counter attack.

Also I'm curious if HardAI is considering bombardments when it makes its odds calculations? Seems to me that it's not exploiting all the battleships and cruisers it buys to the fullest effect possible. Against W. USA for example, I often leave it light just to see if Japan will bring units down from Alaska to attempt the hit, but then never do.

So far pretty much all my energy has gone towards playing against HardAI as Axis, since I'm sure teaching it how to play Allies will be way harder, but even then HardAI USA (like Japan) will purchase a lot of battleships as well.
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

redrum
Administrator
@Black_Elk - Some nice games. So I posted a few things over on A&A forums to see if there is any interest in the TripleA AI development/testing. Thanks for the heads up.

Did you end up pulling off the stackfest game? You actually got pretty lucky on the dice with your UK/US attack on Germany as the odds weren't in your favor. Even after that it looks like Japan could possibly still win. I'm guessing you were mostly screwing around that game but it ended up in a very interesting position.

So the AI currently doesn't use SBR but this is on the list of things to implement. It appears SBR is more utilized in V5 compared to most other versions. You seem to use it extensively in most of the save games you posted. If say you didn't use SBR at all do you think you'd beat the Hard AI as easily?

The AI tends to just evaluate what is the best target each turn based on distance and value so after Japan clears East Asia it tends to think the USA is the best target. Japan plays better on maps that have better East Asia factory locations such as V2 since it can build land units and rush towards Russia then.

It will sometimes use blitz units to take territories but prefers to leave 1 unit in each territory so rarely leaves them completely undefended. It does use bombardments when attacking at least to some extent. If you have any saves that show places where it should have used them and didn't, please upload.

Thanks for all the feedback! Looking to upload the next pre-release soon which will be the first take on improving safely landing air units.
Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Black Elk
This post was updated on .
Yeah right now I am enjoying teasing out reckless strategies against HardAI Japan, just to see what he does haha. v5 is kind of a peculiar successor map to Revised, in that it eliminated many of the best shucks for US/UK, so it becomes fairly tricky to overcome the German stack machine without steady bombing I find. Right now the easiest path to victory seems to be either the fighter wedge at the center, or gutting Japan early and then redirecting. Its also a bit more challenging to run a pure KGF against G, since Japan HardAi is fairly active against Alaska, and pushes their fleet vs W. US if you don't do anything about it.

Also trying some KJF scenarios where Russia just reckless throws everything east to see how much force it takes to deter Japan from buying the Manchuria factory. Most players will wait until J2 to see what US does before committing to the factory buy (just in case), but to persuade HardAI to hold off, Russia has to send in armor. Flying both Ruskie fighters will back Japan HardAI off Bury which is interesting. But for the most part Japan will focus a lot of energy on North America.

The usual response to this kind of play since Classic is the build out of W. US, then shift the ground north to W. Canada, then East to E. Canada for the launch on Europe. So basically a 3 round set up, with Canada coverage to get the ground in position safely and deter Japanese invasions to stall it. In v5 its always a USA launch with the transports, since the shuck out of E. Canada is removed, meaning US has to buy a lot more transports than in earlier versions. UK fleet Atlantic fleet also starts in a fairly poor position, so most times allies drop south to Brazil to cover Africa in the opening round, then decide where to go.

But where HardAI really shines is the Infantry push mechanic, consolidating stacks, and especially the turtle up. It does this brilliantly, which makes any game concluding before the 10th round very challenging. Quite entertaining. Right now to overcome it, I think the air attack vs transports or carrier based fighters is really the best thing I've got going as human vs this thing. Because otherwise its very adept at keeping its transports out of range of bombers and air.

Here next I'll run a stack game with no SBR to show how deep Germany is willing to grind it, if you don't cut down its factories. He shoots for the long game, probably like round 15 optimal.





Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

Black Elk
This post was updated on .
Ps. OK so basically a 13 round KGF slog, no SBR.
Steady tanking instead! with an India armor redirect for the clincher.

KGF_Tank_Fest_no_SBR.tsvg

Most of the effort went into turning Karelia early. D-Day was in round 10.

Still bought bombers with USA but no SBR this time to see how long it would take, just flying them around for defense pips half the time if no combat was optimal. A bit more cut throat against the AI with less fooling around hehe. Air vs transport exploits helped a bit to set up the early med situation, picking them off sooner rather than later seems best.

I find that generally if I'm not going to bomb heavy in v5 as Allies, then I gotta gamble with the Russian opening. If I can get a halfway decent strafe on Ukraine and retreat the Armor then I'll go for a tank wall strategy. Then I prefer to push armor out of India for as long as possible with UK, jump them up into southern Russia at the last second, to be directed on KGF. I think I'd probably go the same way against another player in a similar situation.

Tanks are more high stakes, but I like how they can put heat on Germany to turtle up. Its basically this or try for some sort of Belo Blast or just stack W. Russia and buy a shit ton of fighters. Whether Ukraine strafe or Ukraine take, or whatever the plan, I just think buying 8 infantry with Russia is a bit weak sauce on this board. Its hard for the western Allies to do much in time, if Stalin isn't prepared to go a little nuts, so I say its the best way to get ahead against the AI, same as a human. Here Germans hit back fairly well, retreated when I got down to just the tanks. They dropped I think 4 guys in W. Russia.

Starting to get a better feel for how to pull the HardAI around a bit, just to threaten amphibious without actually launching them seems to work well. You can hold G to the west, but it takes a lot of ground. I think it really pays to stage US ground in England on this board, wait another couple round and just buy more defenseless transports constantly. Fun stuff though. You can see HardAI Japan wasn't going to take it lying down! Fun stuff

edit pps.

Here is another sort of game...

Instead of allowing the Axis HardAI to make its usual opening moves, I played out the first round for them, using a somewhat more optimal approach to Axis. Here as Germany, I make the Allies pay in Egypt for not sending a Russian fighter to back it up, and positioning the Air in the way that produces a greater threat to Atlantic naval builds from UK. Japan rather than purchasing a factory in Manchuria, purchases their Factory in East Indies! And buys 2 transports, save 1 ipc, light hit on Pearl. This is basically a version of an Axis opener I've seen many times from human players. Now I want to see what HardAI will do from this position as Axis.

I think this is fairly optimal Japan play to model for a HardAI opener. For the Allies first round, I went with one of my standards, when I have to deal with a light hit on Pearl...

Here's one with a German ukraine bait, Egypt hit, Pearl hit, Japanese East Indies Factory
HardAI_v5_Axis_opener_East_Indies_factory.tsvg

Here's another similar one, but with a German med Carrier play and more conservative position on the eastern front, again with a Pearl hit light and a Japanese East Indies factory.
HardAI_v5_Axis_opener_German_Med_Carrier_round.tsvg

I'm going to try a few games to see how HardAI performs from these positions in the second round, and will post later on what HardAI is able to achieve, if they are able to exploit the factory in East Indies.



Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

panguitch
@redrum

So I've been unable to reproduce the "loops not allowed in" error on my Greyhawk map with the Hard AI, or even with the Easy AI. So it seems isolated to the Medium AI. And I'm guessing that, while the Medium AI isn't exactly deprecated, it's not receiving further development?

In any case, putting the Hard AI through it's paces on Greyhawk more thoroughly has produced the results I've already mentioned: (1) slow performance (perhaps because of the greater range of all units?), (2) confusion between the hybrid offensive support/factory role of leaders (which you've already explained), and (3) some smart moves by the Hard AI that make me wish my game design was more compatible.

What's new and the point of my post is an error that I've seen several times now and seems unique to the Hard AI in Greyhawk (which is just one map, of course, but the error may be of interest to you). While I get the error in about half the Hard AI games I run, for whatever reason it doesn't persist--I mean that if I save a game right at the error, and then load that game and run it, the error often doesn't recur. It seems to have something to do with factories and/or terrain effects. But it's over my head.

Here's a saved game: hardAIerrorR7.tsvg
And here's the error log: hardAIerrorR7.txt

Greyhawk Wars
TripleA in the original Dungeons & Dragons world

Reply | Threaded
Open this post in threaded view
|

Re: AI Development Discussion

redrum
Administrator
@all - Just committed the latest changes and updated the pre-release jar:
- First pass on improving landing air units safely
- Added check to abort purchase calcs if the AI no longer has any valid place territories
- Added logic to allow the AI to attack with lower odds if the reward is high (capitals, factories, large TUV, etc)
- Fixed a few transport bugs that would cause them to move to undefended locations
- Added logic to handle move restrictions based on territory effects
- Added logic to reduce the number of low value amphib attacks

@Black Elk - Interesting idea to give the AIs a standard opening and see if it makes much of a difference. You should see some improvements to many of the things you commented on. The one that I still haven't tackled though is naval units getting stuck in certain locations (panama canal) as that's not an easy change. I've added it to the TODO list though.

@panguitch - So you have a lot of warnings going on which isn't good. If the AI logs a warning it means something essentially failed but it handled it gracefully. Most of them appear to be around territory effects and factory issues. I think I resolved the territory effects issues so you should see less warnings now. The actual error you get appears to occur because the AI can't determine what territory a certain unit is located in (this should never happen). I'd suggest testing with the latest pre-release and uploading a save along with warnings/errors and feedback. Then I'll take a look. I probably won't really consider looking at performance with as many other issues exist with the AI on your map as right now I'd consider it a non-AI map. I'm willing to try to resolve issues as long as they are relatively small fixes.

PS - Yeah, the other AIs are essentially deprecated and I won't be making any changes to them unless there is an error on one of the standard maps. I believe Veqryn has the same stance so I doubt anyone will ever make fixes to the medium AI for other maps.
1 ... 29303132333435 ... 63