19 replies to this topic

[ArkhanTheBlack]

Next and probably last one.

Changes:

- Added attack chains (AI's search another close enemy post as target if they finished an attack.)

- Improved harassing

- Fixed a heavy defend plan bug which sometimes led to suicide attacks

- More accurate analysis of attack situation

- Reduced allied support a bit (They aborted entire attacks too easy...)

- A few other slighter fixes...


This should be the last beta. If there aren't any biggies like crashes or sudden severe gameplay issues, then we're ready for our 1.8 release.

[thudo]

I'll send this up for the testers tonight.

[Quietdeath]

I would disable the squiggofant in the build order - in 5 games it always was stuck between buildings and no help for the AI - instead it is using vehicle points which could be used to buy other tanks...

[thudo]

I would disable the squiggofant in the build order - in 5 games it always was stuck between buildings and no help for the AI - instead it is using vehicle points which could be used to buy other tanks...

Uhmm not gonna happen - thats like saying: lets disable all ubers like LR or BBs because the map you played has a tight startpoint. Sorry folks but the AI cannot choose what vehicle building to build ubers from - there is no code for that or else we'd have ubers being produced on outer factories. This is all map dependant and AI code which doesn't exist for such large units to be built somewhere without being boxed in.

Btw.. 1.8b6 is going up in a few minutes for the testers!

[Finaldeath]

I'll have a go testing it - I've recently been much to busy to do any work on anything (the AI or anything else) and offline too doesn't help

Will see if the AI can actually beat me

[Slash]

Report #1 for v1.8b6

Match: ork(me) vs eldar
Map: Jungle Morning
A.I Difficulty: Hard
Resources Rate: Normal
Resources Sharing: Off
Winner: Me > AI

Comments:

- Since i was orks the AI could easily repel my forces using warp spiders, but instead it used dark reapers and later banshees (although shees are a good choice, but the AI should had brought them out at start). The AI should atleast stop using dark reapers after spoting what kind of units i have on the field.
-At the beginning of the game, i had already captured one point and half way to the next ,while the AI was still trying to capture its first. AI is really slow in capping points.
-IMO the AI as eldar shouldnt get 3 squads of gaurdiens, but instead 2. It should had used the extra squad for unit upgrade or buy a more usefull unit squad.
-AI uses too much time capturing points, since this is hard AI it should have attacked me instead. By doing that it could had damage my economy and kinda screw me up
-kept using dark reapers more than half way through the game
-the computer is still capping points while im attacking his base with a 2 sluggas and 1 shoota squads while with the hero >>>>>
-the computer was attaching its farseer to its dark reapers.......not very effective. I would recommend the farseer to attach it self to a shee squad
-inspite of all this, the ai did some really good dancing
-AI kept sending hero in by itself and sending it out of battle when being overpowered (used fleet of foot)
-AI doesnt regroup at a point and attack my forces, it just kept on attacking with scattered squads....which led to their deaths


Report #2 for v1.8b6

Match: ork(me) vs eldar
Map: Icy Wind Storm
A.I Difficulty: Hard
Resources Rate: Normal
Resources Sharing: Off
Winner: Me > AI

(i wanted to test out the ai in a custom map made by me)

Comments:

-AI did most mistakes it did in report number 1
-I like how the AI tries to rebuild it's HQ, but it builds it in the same spot as before....too predictable for the enemy to spot its newly constructed HQ
-I also like how the AI teleported it's builders out of danger, ive never seen this done before in the older builds


Well the AI has improved, but its still doenst do good when being rushed (2 sluggas and 1 shoota + hero) even though its a hard AI; it should try to be able to counter early attacks or do attacks it self. Also i would love if the AI can spot a cover and send it's units there, it would be awsome . Now im gonna test out a team game and see how the AI will do


-Slash

Edited by Slash, 28 June 2006 - 01:10 AM.

[LarkinVB]

I suggest doing a capper/builder/capper/builder initial squad build instead of builder/builder/capper/capper/ to speed up initial capturing.

I also suggest to reduce the number of capturing plans from 2 to 1 in mid/endgame or when fighting squads are desperatly needed.

[ArkhanTheBlack]

Thanks for the feedback! I made several final adjustments to get improvements to some of the issues:

- Dark Reaper ratings were indeed to high
- Ignoring enemies while capturing reduced
- Changed build orders to build scout squad before builder
- Better harassing
- Delayed critical capturing after harassing phase

I also suggest to reduce the number of capturing plans from 2 to 1 in mid/endgame or when fighting squads are desperatly needed.

Was already done!


I made some more test games and was satisfied with the changes so far. I fixed now the build and prepared the installer. Therefore we should be as good as finished now! Expect more from Thudo.

Thanks to all of you!!!

[thudo]

- Dark Reaper ratings were indeed to high

Hmm... Yeaa I figured the Eldar AI needs Banshees as well although at one time DRs were amazing. I guess we have to watch it here so DRs are not too excluded. But yes.. their rating was high.

I got the new build now.. will examine so we can get this mutha out the door before 1.5.

[Slash]

1.5 is out....will everything get screwed up now =/ ?

[thudo]

We're checking..

[JBird]

Beta: Build 6.
1)

Match: 2v2/ Me= SM Ally= IG Enemies: Orks + SM
Map: 4p_
A.I. Difficulty: Hard
Resources Rate: Standard
Resources Sharing: Off
Winner: Them
Win Trigger: Take & Hold @ 13 min 44 seconds

Comments:
Not to be obsessively vulgar, but my ally sucked ass. He got ran over within 5 minutes and I couldn’t save him because he didn’t even try to put up a fight. I was over-run in the resulting 2v1. Enemies did awesome… scary… Ally sucked.

2)
Starting Thoughts: This will find out whether IG is indeed crippled if I mow them over IG is pathetic. If they win, perhaps I’m just not a good IG player/ally.
Match: 2v2/ ME= Eldar Ally=SM Enemies: IG + IG
Map:
A.I. Difficulty: Hard
Resources Rate: Standard
Resources Sharing: Off
Winner: Them
Win Trigger: Take & Hold @ 22 min 53 seconds

Comments:
Well, I made the mistake of trying to rush attack them… My ally did not “catch-on” to my plan and did nothing to help me secure the T&H point in the middle. I had forced being built [Avatar and many tanks] but the clock was too short and I lost. Never mind about the IG, the ai handles them well.

Notes:
Enemies have gotten much better from 5 to 6, for skill-wise it’s a great thing! However, the allies seem to suck now, they don’t assist in your attacks, help defend you, or basically co-op work hardly at all… O well, I need more solo practice anyways.

[ArkhanTheBlack]

1.5 is out....will everything get screwed up now =/ ?

AI seemed to work with 1.5. I'll make adjustments today to react to the changes.

Not to be obsessively vulgar, but my ally sucked ass. He got ran over within 5 minutes and I couldn’t save him because he didn’t even try to put up a fight. I was over-run in the resulting 2v1. Enemies did awesome… scary… Ally sucked.

It's possible that your ally tried a vehicle fast tech. I personally noticed that the vehicle fast tech got extremely risky since I improved the harassing behaviour. I think I'll restrict it to 6+ players.

Allied cooperation in general is difficult since they can't communicate. Though they should assist if you're attacked. Maybe I nerfed the allied support a bit too much, since QuietDeath was very impressed by the allied support last time. I'll have a look...

Thanks for feedback!

[LarkinVB]

I find that fixed build programs are limiting the AI more than helping it.

The more Relic tries to balance the game the more "rush to certain unit type" will lead to disaster.

The AI should use a default balanced build program 80% of the time and only pick special ones if it knows the map/enemy/setup is right to do it.

Currently DoWpro AI is using only 1 build program and I doubt it will help (as in improve the AI performance) to add much more.

[ArkhanTheBlack]

I don't get it! The only difference to allround is that they build 2 - 3 fixed squad at the start. Even the 1.5 skirmish AIs and before had a fixed build order for the start which was even always the same. The selection of the programs are also modified by the opponent.
The unit selection of the standard build order doesn't seem to care about the opponents units race or units at all. Therefore at least at the start, the unit selection should be probably always the same. The consequence is that a player builds exactly the counter units to the AI's standard selection and I think it's pretty clear that this isn't really an advantage for the AI.
Besides that, at the start it's more important to effectively use a unit than to build one specific unit. The start units can't be too unbalanced since both players don't know the opponents units or even the race at this point. Nobody would play DOW if you can just give up a game only because the opponent built accidently unit xy and you yz and this results in an autoloss.

The vehicle fast tech is the only critical build, but it's only 6 players+ and you can even switch it off.

Edited by ArkhanTheBlack, 29 June 2006 - 04:46 PM.

[thudo]

And whats great about the buildprograms is that we can specify units to be queued so theoretically we could spam an all-around good unit all the time and own anyone BUT why? The buildprograms do their trick - the only thing I wish it could do is place Generators down when it really needs the power. Generally, though, I tweak its timings based on what consumes most power at what times in the buildorder. Buildprograms offer alot of flexibility and configurability.

[LarkinVB]

I beg to differ. I observed stormboy rushes uttely fail due to the AI inability to use them well. There are other examples. No problem if you think otherwise.
Stormboys seem to be boosted so who knows, it may work fine with 1.5.

[ArkhanTheBlack]

That's exactly what I said: It depends on the correct use of the units.
Besides that, pretty much every ork rush fails in WA 1.41, regardless what they build. It's by far the weakest race in 1.41. Stormboys are also extremly late for a tier 1 unit. That's exactly the reason why patch 1.5 reduced their pop requirement to 30 from the previous 40.
Before I joined the AI team, I've also done exactly what I said about the problem with one fixed build order. AI 1.3/1.4 had a fixed chaos marine build order for the start which was quiet hard at that time. But it could be easily countered by a full raptor/assault marine rush. I've done that with a friend on mountain trail and we won every game with that, just because the AI did the same start build all the time.

All the special builds have a 2 : 1 relationship of melee to shooter and vice verca to reduce the risk of unit massing. I also don't consider two units of the same type as serious 'massing'. Especially if you have three scout squads in addition.

Anyway you don't have to use several build programs for DOWPro.

[LarkinVB]

Anyway you don't have to use several build programs for DOWPro.

Don't get me wrong. I'm not against different build programs if program B is more effective than A under specific conditions.