Jump to content


Photo

Skirmish AI 3.0 Beta 4 - Post Comments In Thread!


48 replies to this topic

#1 ArkhanTheBlack

ArkhanTheBlack

    title available

  • Members
  • 814 posts

Posted 29 March 2008 - 10:34 PM

Changes:

- AI researches should be more efficient now

- Improved T2 behaviour for all races

- Fixed force tech bug (especially DE)

- Lot's of changes and improvements for DE and Sisters

- Some more improvements to harassing code

- 1 x capturing squad forced in T1

- Other changes I can't remember now...


Remark:

AI crash check is activated. That means that in every game a samll file AI_Log.txt is written into the DOW SS directory. If the game crashes, just have a look at this file. If it ends with 'End AI Think' the crash was not (directly) caused by the AI. If it's missing the crash was caused by the AI. In this case just post the file here. Try to remember the races which were involved. A line in the AI files starts with the player number 1000, 1001, etc. . That's first player, second player, etc. Check this number and add the race of this player to the report.
IMPORTANT: The AI_log.txt file is overwritten each game. Therefore make a copy of the file when you had a crash or post it immediately. If you start another game, the info is lost.

#2 LarkinVB

LarkinVB

    title available

  • Members
  • 1,488 posts

Posted 30 March 2008 - 12:10 AM

Thanks.

Quick observation :

1. One DE builder is not enough in T1. He is often wandering to front to build LP or slave chambers, slowing down building of more important stuff at base. DE therefore had lots of req (700+) at T1. Slave chambers shouldn't be build half across the map.
2. Harassing archon retreated as soon as under fire from a 4 men eldar guardian squad. Guess a bit fighting weaker squads won't hurt while harassing.

#3 ArkhanTheBlack

ArkhanTheBlack

    title available

  • Members
  • 814 posts

Posted 30 March 2008 - 01:00 AM

1. One DE builder is not enough in T1. He is often wandering to front to build LP or slave chambers, slowing down building of more important stuff at base.

Yes, I guess you're right. I've thought about it too, but I wasn't sure if it's better to save the resources for the builder and accept a bit slower building or invest the resources for the second builder.


Slave chambers shouldn't be build half across the map.

Hmm, so it's a general problem. I guess I set it back to HQ or military. It makes no sense to set it on front if the placement algo for the building is broken on some maps.

#4 Zenoth

Zenoth

    title available

  • Members
  • 469 posts

Posted 30 March 2008 - 01:23 AM

I see that my suggestion for a single DE Slave in Tier 1 back fires on me :p

Well you know Arkhan, as I suggested, it has to be linked with the tactical decisions at the game start. If the DE team in question decides to harass early, he'll need the 75 Req for another Mandrake Squad as soon as possible, instead of a second Slave. But if the decision is to concentrate on faster building/teching to reach Tier 2 faster overall, then why not having a second Slave? I guess it just depends on how you want to play it, how the A.I wants to start the game. I really don't mind much if you guys decide to force two Slaves again really, I mean it was just a suggestion, I'm no balancing expert, but from my own experiences with Beta 3 I thought that a single Slave allowed for good harassing early in Tier 1.

Now as Larkin points to, there's some Requisition excess, that means that the current build just don't use the single Slave advantage properly. If the A.I could use that excess for more troops, there goes my suggestion's real potential. But if you guys wouldn't be able to work around that then of course let's have a second Slave back, ain't the end of the world. And we've probably got a few more builds to go through so time will tell from you guys, me and other testers I guess.

#5 dreddnott

dreddnott
  • Members
  • 60 posts

Posted 30 March 2008 - 02:44 AM

I think the real problem is that an AI player can't order a single Tortured Slave around in the most efficient manner possible, whereas a human player can easily queue the appropriate buildings and set waypoints to minimise idle time.

It reminds me of the old Traveling Salesman problem, one of those calculations that computers are so poor at.

Frankly I don't see the AI running out of requisition that much in T1, the req tradeoff for purchasing the slave is worth the time the extra builder saves it getting structures up and running more quickly.

#6 Zenoth

Zenoth

    title available

  • Members
  • 469 posts

Posted 30 March 2008 - 07:05 AM

Alright I've had a crash again (with Beta 4 of course), in the Campaign... same story, played as the Eldar on Hard, against the Necrons on Broken Lands as my first mission, it crashed at approximately 10 minutes of play. The AI Log does not end with "End AI Think". But thanks to the simplicity of the situation there's not much else to remember other than it was me against the Necrons on that map in the Campaign.

At the moment I'm typing this I can't attach the file because it tells me that the file is larger than the available space... it's only 1KB, perhaps it's actually too small? But really there's not much to see other than that it doesn't end with "End AI Think" as I said. So here let me show you what it says rather than trying to wait and see if I can upload it:

AI1001 08:48 CpuManager: Starting AI Think
AI1001 08:48	CpuManager: Update unit tactics...

As you guys can see that's all the file shows, there's nothing else. But there's the number, 1001, that's the second player which should be the Necrons of course. I'll go try a Skirmish on that map again and I'll see what the AI Log says that time if it crashes (which will certainly happen).

Edited by Zenoth, 30 March 2008 - 07:06 AM.


#7 Zenoth

Zenoth

    title available

  • Members
  • 469 posts

Posted 30 March 2008 - 07:37 AM

Well it's getting weirder now... I've just tried a Skirmish, again in Broken Lands. Me + Eldar against two Necrons, on Hard... and it did not crash that time, it lasted 25 minutes, and the Necrons even won the match. I don't get it...

#8 LarkinVB

LarkinVB

    title available

  • Members
  • 1,488 posts

Posted 30 March 2008 - 09:31 AM

To add back second DE builder just edit \Dowai_mod\Data\ai\races\dark_eldar_race\strategies\darkeldarstrategyinfo.ai and set

Engineers =
		{
			--number of engineers we're aiming for
			want = 2,
		},


4p the skerries map, 2 x DE vs 2 x eldar. DE at position 2 did build two slave chambers in first minute. I did repeat battle and he did it again (strategy 2). Guess there is something broken. Btw eldar stood no chance. They seems very weak in this matchup, no idea if there can be something improved. They are running around all the time versus mandrakes and hellions, never making a stand. Guess banshees can help but they didn't build them.

All DE build orders start with hal of blood. Why no jetbike rush ?

Edited by LarkinVB, 30 March 2008 - 09:38 AM.


#9 Smokeskin

Smokeskin
  • Members
  • 127 posts
  • Location:Denmark
  • Projects:Soulstorm Advanced AI betatester

Posted 30 March 2008 - 09:45 AM

I just tested a single AI vs AI battle, and the capping behavior seems better so far :p I did see one case of a strong squad capping, but it was fairly close to the main army and there wasn't any fighting going on, so it was fine - if fighting had started, would it have stopped capping and started fighting? I don't know, but if it had, that would be perfect behaviour :)

- Lot's of changes and improvements for DE and Sisters


I just looked at the Sisters AI code, and they don't seem to have changed at all (there's of course the possibility I'm just not looking in the right place though). The behavior don't seem to have changed though.

Battle sister squads still don't use Veteran superiors much, even though these guys give +30% DPS and are absolutely essential, if you're not building these then it's because you don't want battle sister squads. So their infantry gets creamed in t2.
Celestian squads don't upgrade their weapons much, so the sisters completely lack AV and AB in t2.
Regarding BOs, I see only 1 Celestian squad added and after the Manufactorum and an Immolator tank, and they still build both a Manufactorum and a Holy Reliquary in t2.

So, they still lose pretty badly against SM, unlike the tweaked sisters AI I made which completely steamrolled SM (except when it sent the crucial Celestian squads to cap CPs of course).

I'm guessing the other races, it is the same.

Did the race balancing suggestions I made for beta3 not get evaluated or implemented yet, or do you disagree with them?

I'm just a bit confused as to what I'm supposed to test for, since most of the beta3 stuff I adressed is left unchanged. A large part of my playtests have been about how to make AI play better, and I don't really see any point in repeating that stuff if you just haven't looked at it yet - then I'll just reiterate that my beta3 feedback still is relevant. If you feel you have implemented my suggestions, then I've got to test and report on the changes you've made.

It's very hard beta testing when there isn't a list of issues with status of implemented/not yet implemented/not yet examined/rejected.

If I made a list of the issues I've discovered through my testing, will you go through them and mark them?

#10 LarkinVB

LarkinVB

    title available

  • Members
  • 1,488 posts

Posted 30 March 2008 - 09:53 AM

It's very hard beta testing when there isn't a list of issues with status of implemented/not yet implemented/not yet examined/rejected.


I think such a list can be very helpfull. Thudo, I guess it is your job :p ?
I'm not sure wether some of my observations are just not examined or rejected. Expl is missionary just started capping under attack until dead.

#11 ArkhanTheBlack

ArkhanTheBlack

    title available

  • Members
  • 814 posts

Posted 30 March 2008 - 12:48 PM

4p the skerries map, 2 x DE vs 2 x eldar. DE at position 2 did build two slave chambers in first minute. I did repeat battle and he did it again (strategy 2). Guess there is something broken.

No, it is intended. With only 1 slave chamber they don't have enough squad limit. I've noticed it myself and it was also mentioned by a tester.


Battle sister squads still don't use Veteran superiors much, even though these guys give +30% DPS and are absolutely essential, if you're not building these then it's because you don't want battle sister squads. So their infantry gets creamed in t2.
Celestian squads don't upgrade their weapons much, so the sisters completely lack AV and AB in t2.

Sorry, I've forgotten to check out the tactic files. I guess that's the main problem.


Regarding BOs, I see only 1 Celestian squad added and after the Manufactorum and an Immolator tank, and they still build both a Manufactorum and a Holy Reliquary in t2.

Um... yes! You're a building a manufactorum too in T2, so what's the problem. I also don't understand how the sisters are supposed to research stuff if they aren't allowed to build their Holy Reliquary!? Further, why do we need 2 AV squads if the main force at the start of T2 is infantry? I increased their rating instead in the unit stats file. It's not good to set too much units in the build programs. Only what's necessary and nothing more. The AI builds units depending on rating and enemy unit types. This ensures that the AI can react to the opponent. If this isn't working correctly, the unit rating should be increased.

The sisters are also building their dynamic buildings correctly now. That has a huge impact in Kasyr, because they couldn't spam enough vehicles although they had enough resources.


I'm just a bit confused as to what I'm supposed to test for, since most of the beta3 stuff I adressed is left unchanged. A large part of my playtests have been about how to make AI play better, and I don't really see any point in repeating that stuff if you just haven't looked at it yet - then I'll just reiterate that my beta3 feedback still is relevant. If you feel you have implemented my suggestions, then I've got to test and report on the changes you've made.

(Sigh) As long as I still have to adjust 9 races, checking out the core code, solving crash problems, etc. it's hard to keep up with all suggestions and changes. I also focused more on DE than on sisters in this build. You have to understand that releasing a new beta doesn't mean that all previously reported problems are automatically solved. A new beta is released if enough content was included to justify a new build.
If you notice that something wasn't included which you think is important, just mention it again. If I have to read reports, post answers, check out the DOW Wiki to look up stuff, solve coding and balance problems and test them, I slowly get at my limits. I'm just human. I know, that's annoying sometimes, but what we are going through the whole time is not just annoying, it's very often highly frustrating.


Besides that, I asked several times for testing of saving / reloading and campaign problems. The only one who actually seemed to care what I'm asking for was Zenoth. He was also the only one giving feedback for the fortress and massive battles mod. I have understanding that everone tries to focus on the parts of the mod he likes most, but we have to care for the whole thing and not just a specific part. Therefore, if you give a detailed report or analysis about something we didn't excplicitly ask for you might have to wait a bit until we include those stuff.

The good thing is that the old races seem to work now for the most part (except maybe the Necron crash), so we can focus on the new races now. I therefore try to add the missing sisters stuff now.

#12 Smokeskin

Smokeskin
  • Members
  • 127 posts
  • Location:Denmark
  • Projects:Soulstorm Advanced AI betatester

Posted 30 March 2008 - 01:55 PM

(Sigh) As long as I still have to adjust 9 races, checking out the core code, solving crash problems, etc. it's hard to keep up with all suggestions and changes.


I completely understand. I'm not bugging you to get it done, I just need to know what to look for.

It's a bit the same from my side. I've got something like 10-15 things I've found that I think needs fixing. I don't know what you've tried to fix, what you haven't got around to yet, what you don't agree with and so won't fix, and what you just missed. It's hard for me to guess what you're thinking and keeping up with your changes and schedule too.

That's why I suggest a list a issues and status. How about each beta tester (if he feels like it), makes his own thread, with a short, bullet-point list of issues. You reply, getting the 2nd post, and marks each of them with 1 of the 4 status I suggested. You edit this post as the work goes along and something changes status. If a betatester has new issues, he makes a new post with it so you see it, and edits his first post with it too. You then add this to your list in your post #2 and marks a status.

The advantages are that you don't need to go throught the entire "beta x comments" thread to check if you've missed something, or do much work in keeping your own list of reported issues. You can easily go to my thread to see what I've think should be changed and what you haven't looked at yet.
When you change something in the code or choose to reject an issue, I can easily see it because you've changed the status, and so give you better feedback when something gets updated.
I can easily see what haven't been changed yet, so I don't have to report the same thing over and over, and I don't have to wonder if when my changes don't get in a version, what the reason for it was and if I should spend time testing it again. There's no need for me to test race X again if you haven't gotten around to looking at my changes yet.

To me it seems like all advantages, and the little overhead for updating such a list is much less than the work it'll save imo.

So how about it? If I make such a list, is it something you'll look at and update?

#13 Smokeskin

Smokeskin
  • Members
  • 127 posts
  • Location:Denmark
  • Projects:Soulstorm Advanced AI betatester

Posted 30 March 2008 - 02:16 PM

Regarding BOs, I see only 1 Celestian squad added and after the Manufactorum and an Immolator tank, and they still build both a Manufactorum and a Holy Reliquary in t2.

Um... yes! You're a building a manufactorum too in T2, so what's the problem. I also don't understand how the sisters are supposed to research stuff if they aren't allowed to build their Holy Reliquary!? Further, why do we need 2 AV squads if the main force at the start of T2 is infantry? I increased their rating instead in the unit stats file. It's not good to set too much units in the build programs. Only what's necessary and nothing more. The AI builds units depending on rating and enemy unit types. This ensures that the AI can react to the opponent. If this isn't working correctly, the unit rating should be increased.


I think that in general, you build either a Manufactorum in t2 and go vehicles, then t3, or you build a a Holy Reliquary and go heroes, then t3. Trying to do both is spreading yourself too thin - you don't have resources to both keep some vehicles on the map, keep both their heroes on the map, and do the hero upgrades. You have to choose, either heroes or vehicles, followed by t3. There is, I guess, an option to go with both heroes and vehicles in late t2 if it turns into a high income game with many troops, but that's rare imo.

Regarding the Celestian squads, I've been asking about if the AI can respond to vehicles by building certain squads, and how it decides its army composition. I never got an answer, so the only thing I could do to fix it was to add the squads to the BO, which is more effective than what it does now. If the AI can do this dynamically instead, that is much better, I just didn't know about it or how it can be fixed. I think there is a general problem with most races that they don't respond well enough to vehicles because they don't use their infantry AV options.
BTW another advantage of getting AV squads out is that they're mostly also very good vs buildings. This means that if the enemy retreats and it gets the upper hand even shortly, the AV squads will ensure that it takes down one or more LPs or other buildings, letting it capitalize on its window of opportunity.

#14 ArkhanTheBlack

ArkhanTheBlack

    title available

  • Members
  • 814 posts

Posted 30 March 2008 - 03:26 PM

So how about it? If I make such a list, is it something you'll look at and update?

We should have done that from the start, but I'm not sure if we still need it at the current state. The old races are pretty much finished and we can now focus on the problems left with DE, Sisters and the Necron crash. I think we can handle this without a list. However, if anyone starts one, I'll use it.


I think that in general, you build either a Manufactorum in t2 and go vehicles, then t3, or you build a a Holy Reliquary and go heroes, then t3. Trying to do both is spreading yourself too thin - you don't have resources to both keep some vehicles on the map, keep both their heroes on the map, and do the hero upgrades. You have to choose, either heroes or vehicles, followed by t3. There is, I guess, an option to go with both heroes and vehicles in late t2 if it turns into a high income game with many troops, but that's rare imo.

I think we can find a compromise her. I've delayed it now after the T3 HQ addon has started. Therefore it shouldn't delay T3. Further, it's only built with 2000 army strength. That should be enough to ensure that it won't disturb vehicle production in the early and critical T2 phase.


Regarding the Celestian squads, I've been asking about if the AI can respond to vehicles by building certain squads, and how it decides its army composition. I never got an answer, so the only thing I could do to fix it was to add the squads to the BO, which is more effective than what it does now.

Yes, there was so much feedback that I just ignored some questions and focused on the main problems. Sorry! The dynamic building isn't very reliable, but by adjusting the unit ratings high enough, you can get decent results.


I think there is a general problem with most races that they don't respond well enough to vehicles because they don't use their infantry AV options.

Yes, that's mostly because we can't control the upgrades good enough. But usually if you set some life saver units in the build programs, the AI doesn't fall too deep in a hole.

Since sisters have noticeable problems with vehicles, it might be still necessary to force 2 celestial squads. But I don't want to do that if I can avoid it. I've already added the missing tactic stuff. I also fixed the auto reinforce of standard troopers. The forced leader should be enough.

#15 thudo

thudo

    Wacko AI Guy!

  • Division Leaders
  • 12,164 posts
  • Location:Lemonville North, Canada
  • Projects:DoW AI Scripting Project
  • Division:DoW
  • Job:Division Leader

Posted 30 March 2008 - 05:12 PM

I think such a list can be very helpfull. Thudo, I guess it is your job

Yes it is but after Beta3 I was utterly overwhelmed by all that. Wow.. Its like we opened up a massive can of worms and now have to baby-sit everything all at once.

I would have to go back and track all issues but 95% of them are NOT show-stopping bugs but rather little gripes people have about this and that and we honestly need to keep reporting to a POINT FORM fashion. Having these massive elaborate paragraphs are insane for logistical purposes. Devs want clear and concise points - not waves of emotion.

Also, I think the testers don't realize this DoW AI Engine can't do it all like most RTSes cannot: there are upper limits. We also run the risk everyday of fixing ONE thing but breaking 2 more unrelated features. This is precisely why Relic doesn't like dealing with code: they fix one major problem only to create little ones down the stream.

So if I can get a clear and concise sum-up I will.. things are SOOO much easier to prioritize when its easy to read.
Advanced Skirmish AI Team Lead for the coolest Warhammer40k PC RTS out there:

Dawn of War Advanced AI Headquarters

Latest DoW Advanced AI Download!

#16 Smokeskin

Smokeskin
  • Members
  • 127 posts
  • Location:Denmark
  • Projects:Soulstorm Advanced AI betatester

Posted 30 March 2008 - 05:19 PM

Yes, that's mostly because we can't control the upgrades good enough.


Can't you control them by the argument for DoBestUpgrade(class_type)? Say space marine squad, feed it VehicleHeavy or whatever form the arguments for the function are supposed to be, and it builds a missile launcher?

#17 Smokeskin

Smokeskin
  • Members
  • 127 posts
  • Location:Denmark
  • Projects:Soulstorm Advanced AI betatester

Posted 30 March 2008 - 06:18 PM

Match: 1v1
Factions: Guard (Smokeskin) vs Ork
Game Modes Used: Standard
Map: Titan's Fall
A.I Difficulty: Harder
Resources Rate: Normal
Resources Sharing: Off
Winner: Me

Comments:
I don't know what happens here, the AI plays very passively, I don't think I've ever seen it play like that before. There's a few, very minor skirmishes in early t1, but it doesn't really take any losses. It just lets me take most of the map without a fight, then sits in its base, without trying to break out or anything, until t4 where I attack.
It is very, very strange. Something went seriously wrong with the AI here.

Attached Files


Edited by Smokeskin, 30 March 2008 - 07:00 PM.


#18 Smokeskin

Smokeskin
  • Members
  • 127 posts
  • Location:Denmark
  • Projects:Soulstorm Advanced AI betatester

Posted 30 March 2008 - 08:23 PM

Match: 1v1
Factions: Guard (Smokeskin) vs Ork
Game Modes Used: Standard
Map: Blood River
A.I Difficulty: Insane
Resources Rate: Normal
Resources Sharing: Off
Winner: AI

Comments:
The Ork AI plays fails to mass, plays too passive, etc. I'm way too bad with IG to last over 20 minutes against INSANE imo. Against the other races, they're pushing and harassing from the start.

Attached Files



#19 Smokeskin

Smokeskin
  • Members
  • 127 posts
  • Location:Denmark
  • Projects:Soulstorm Advanced AI betatester

Posted 30 March 2008 - 08:53 PM

Match: 1v1
Factions: Guard (Smokeskin) vs SM
Game Modes Used: Standard
Map: Blood River
A.I Difficulty: Insane
Resources Rate: Normal
Resources Sharing: Off
Winner: AI

Comments:
Nothing really interesting here, except to see the contrast to the Ork AI the SM plays a lot more aggressively, it is pushing me back not the other way around, from the start.

Attached Files



#20 Zenoth

Zenoth

    title available

  • Members
  • 469 posts

Posted 30 March 2008 - 11:34 PM

@ Arkhan

What do you think of what I've reported in post #6 here? I'm just curious to know about your conclusion. I'll go back at "proper testing" later today, for the moment it's gym time at home, why do we humans have to keep ourselves in shape is beyond me :mellow:



Reply to this topic



  


0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users