72 replies to this topic

[LarkinVB]

What new changes are you doing in DoWPro that aren't in v1.5

Currently I'm knee deep in just getting all researches/addons to work and fixing BOs. Korabah changed a lot of stuff. I planned to add best of 1.6 if I'm allowed to. I really like to work on my own and I'm no good team player. Therefore I found a good playing field with DoWPro. I also think Korbah did put more thought/work into it than the whole Relic balance team and deserves to be supported. If I will add stuff you might be interesested in I will let you know.

[thudo]

Have to fully check out Korabah's work and yes he should be supported since he's perhaps more mindful of bringing back what was good and cool in DoW into WA. Thats a big deal for this community.

[ArkhanTheBlack]

I planned to add best of 1.6 if I'm allowed to

Feel free to use whatever you like.

[Excedrin]

1.6b2 MoM Eldar vs Eldar
http://dow.lerp.com/...1202.025159.rec

This one shows that old issue where the AI lets its guys get seperated. It's a tough problem to solve. In the bayonet AI, I modified attackplan so that it would constantly move troops towards an enemy hero (unless they're already close enough), but still stay out of range. I think that if all movement code occurs at the plan level, it should be possible to do something like that in a more general sense (read, for all races and strategies).

I think it used a reasonable build order, but I haven't watched the rep. It did seem a little short on troops in the beginning, which meant I got an econ advantage.

Anyway, no glaring issues so far with this one. I'll play a few more games tomorrow night.

[Malkor]

Errr.. Lag in WA?   

Hows that? Time for a system upgrade? Super cheap these day.

As for the RPG elements - thats being discussed. Its more on the faction-specific behaviour when it comes to assisting allies.

<{POST_SNAPBACK}>

All factions should assist allies no matter what.

As for lag, I can gaurantee you I could previously run DoW (or any other game) better than anyone else here. I can record FEAR using fraps and still get around 60 fps at maximum settings, quake 4 on ultramode, ect. Altering settings in WA does not correct the lag issues that are apparent when viewing nothing in a 1v1 but not when watching a FFA slugfest in doom chamber. I thought it was my ram, but that doesn't fit, why would it lag horribly with all the settings on low and viewing absolutely nothing with nothing but two command centers active? It's not my system. It's WA. The lag issues instantly bring my fps to about 1 frame every 4 seconds, and it travels across the network to any players in MP if I am currently in an MP game.

If you want my technical opinion, I'm pretty damn sure relic screwed something up in regards to either system conflicts or driver handling. I'm on a radeon 9800 pro, an athlon 2500 XP, 1gb of PC3200, terratec DMX 6fire, asus nforce2...

[thudo]

Agree on the ally helping each other always. I know it would be nice to have a little "personality" thrown in for this part but Relic did design each faction to be good against defense and assistance so no point in subduing at any point.

All allies should help out no matter what - its your ally after all.

As for the preformance issue.. Hmm.. I have P4 3.1g 1gb PC3500 Dual-Channel Ram, MSI Neo-FIS2R mobo (latest firmware), SCSI-U160 15k drives + SATA2 drive (only running on SATA1 interface though - DoW/WA runs off this SATA2 drive) and a X800XT overclocked to PE speeds.

My system (except for X800XT and SATA2 drive) is June2003 so frig man I NEED AN UPGRADE! hehehe.. Ahh well..No rush yet.. Have no major lag issues running @ 1024x768x32bit with full details per se.

[ArkhanTheBlack]

All factions assist their allies. They differ only in WHEN they assist. For example the AI shouldn't assists if it is annihilating an enemy at the moment.

[thudo]

Agreed on that although as yer annihiliating an enemy but see in the corner of your eye yer ally is also simutaneously being overthrown.. hmm.. Judgement call no? You take an eye for an eye? Remove one enemy ally for your own to balance things?

[ArkhanTheBlack]

There's no general right or wrong decision for such a situation. If you don't support your ally he might get crushed, but you may crush your opponent as well. That would even it out. Worst case, is that your ally is annihilated but you don't manage to destroy your opponent.
The advantage of attacking instead of supporting is that your opponent also may retreat to support his ally. In this case you've saved your ally by attacking and not supporting. You've then even reached a major tactical advatage. If you support, regardless how it works out, you lost the initiative, which is not very good. It can also kill your ally if you try to support him. The opponent is now free from pressure and will try to attack even harder. You can only pray that your ally is fast enough at your base to save you.

That's why I don't want a support in any situation. It can hurt you more than it helps.

[thudo]

Maybe perhaps a randomizer so you don't know how the AI will react to reinforce his ally? No biggie... this will have to be observed and tested.

[ArkhanTheBlack]

I don't think this is necessary, since the result of the calculation itself isn't so obvious. Besides that the race specific behaviour adds more strategic behaviour to the game. For example, if the ally of a human player is under heavy threat and he decides to attack another enemy because he KNOWS that the attacker is an eldar and will most likely break its attack to support his own ally, then the player has made a decision on highest strategic niveau and I definetly want him to get an advantage through that.
Tactical decisions based on the psychological behaviour of the opponent are almost non present in RTS games today. I think if we have the chance to give the game more strategical deepness, we should use this chance. The AI isn't weaker because of this. It only allows the player to maybe get an advantage when he made a good tactical decision. And this should be honorated.
This is also a bit more than pure racial behaviour. The Eldar are the best supporters because they've the by far fastest army. If a chaos army has to support, it has to spend one hell of a lot more time on just walking than Eldar. Therefore I would consider the AI worse than before if Eldar wouldn't have a better chance to support as the other races.

[LarkinVB]

Some observations, please take them as constructive critism. Arkhan is a very good (much better than me by skill) and dedicated coder but I have problems with some concepts :

1. Congrats for building at maps which usually had problems at certain start points. Seems this is fixed. Great !
2. Build programs. I just don't get what's the real advantage of them. I watched two big AI vs AI battles (4vs4 burial grounds, 3vs3 testing grounds). Some AI played strong, some weak. This is because some build programs fit the map/enemy, others do not. As they are mainly randomized you get an overall average AI which doesn't seem to be stronger than with 1.5. Build programs are very static and only give any advantage if they fit. sometimes they just don't.
One chaos was only building marines, lots of them though he could have build zerkers or raptors. His IG enemy finished him quickly cause vanilla chaos marines don't stand a chance shooting it out.
While his base was threatened he also did sent two squads to fight an enemy elsewhere ?
Another chaos was building two infantry barracks before starting the first pit ?
3. Attack targets. I had one IG (testing grounds) attacking the enemy opposite on the map (far far away) though he had a neighbor enemy. It took him years to cross the map, meanwhile his base was attacked by his neighbor.
4. Jump. I did see very very short jumps where it would have been much better to move normally and save jump for better usage. I did also see jumps at broken tiny squads which were already engaged in CC by other troops.
5. Some turrets were build at distant LPs which did not live very long. If you must build them it is better to do this at home.

v1.5 was inspired by Arkhans directive to tech to vehicles fast. It tried to preserve a basic starting infantry force and was very flexible regarding what units to build.

v1.6 has some build programs which mass infantry and because those are not used for map control it fails miserably with some setups.

I would like to see a general flexible build order as in 1.5 with some static programs triggered by certain maps/setups.

[thudo]

2) BuildPrograms - this can be set to only use one (the all-arounder) BUT wait a sec..
The other RUSH programs are designed to happen only once thus only the initial wave of hand-picked "rush" units are being used. The AI does not or should not be ALWAYS be attacking with the same kind of units? Diversity is the spice of life, no? I don't believe I saw the Chaos AI always attacking with one or two types of units especially Tier3+.

3) I believe this has always been an issue where the AI will something think a far away enemy is actually his "nearby enemy" and thus attack it.

4) Hard to get this perfect.. we're getting there..

5) Well yes and no. Remember our discussions back in ancient times about turrets in general" best spent on units instead? Btw, AI always builds turret back if he loses them BUT show concentrate on units first. Turrets are a luxury. Still.. easy as pie to have turrets built at homebase.

The great thing about the BuildProgram code is for the first time we can fully control outcomes.

More comments welcome about playing the latest Ai build!

[LarkinVB]

2. The chaos AI was just plain marines. It had the armoury and was under attack by IG. It did build at least 4 to 5 squads. Just them until it died.

The great thing about the BuildProgram code is for the first time we can fully control outcomes

NO ! You control build order. Outcome can be crushing defeat which certainly isn't what was intended.

3.) NO ! AI never attacked across the field unless it uses CommonEnemy routine which wasn't the case as I played at hard level. With 1.5 it will attack closest enemy. This new behavior seems related to the new attack routines.

4) I agree. Just wanted to help.

5.) Very minor issue indeed. But since turrets are now part of the build program and not just build because of excessive resources they should be placed more carefully.

I *REALLY* like those mass jump squad programs with Arkhams very good jump code BUT they can backfire. More problematic are the mass reaper/marine programs.

If my observations are not related to the programs but bad flexible squad counter picks I apologize. I suggest using flexible programs most of the time and adding special programs triggered by map/enemy setups.

[thudo]

2) I do not see a problem with this. Who doesn't build sometimes a large CSM squad with just marines then just spams em? The Chaos AI does kuzz IG Guardsmen are nasty against other infantry which is a balancing nightmare. Perhaps then for the early infantry rush program instead of the AI making 5 squads lower it to 3 so then the AI will proceed to ramp up normally. However.. if the enemy was Eldar or Ork the outcome would have been different. It would give the CSM enough time to tech to vehicles afterwards. And I meant the OUTCOME OF THE AI-controlled faction - not game outcome.

3) Will have to address this.. Arkhan now knows about this so changes can be made.

BuildProgram quantities could be toned down.. if the AI is waiting to make too many squads then this could pose a time issue. This can be tweaked hence the reason for beta testing before release.

Thanks a ton Larkin for the feeback. We need more? Excedrin?

[LarkinVB]

Last comment as I still disagree regarding 2)

Chaos marines are WEAK ! Mass chaos marines will do no good.

[thudo]

You don't get any disagreement from me. Actually, what about this..

For Chaos Infantry Rush..

3 Marine.. 2 Raptor Squads? Nice balance no? Remember, you don't just need to specify one build type. Its all about tweaking it to get the right values for each faction. Too bad SM don't suffer from this prob.

You haven't mentioned anything about the other factions - seems CSM got the raw end of the deal after WA. :(

[ArkhanTheBlack]

Thanks Larkin for the constructive criticism. I've observed some of the issues you've mentioned myself, but wasn't able to fix them so easy.

2.) One definite advantage of the build programs is that buildings are always rebuilt if they are destroyed and you can better control the army strength / teching relationship.
The other point are the rush programs. The choice of them ARE influenced by the opponent. There's a higher chance of assault/raptor built against guard than the mass marines build. But the chance is still there. Since I had to write 4 build programs for every race, I took the most obvious and different ones. One mass ranged, one mass melee, one vehicle rush and one alround. I would also lie if I'd say that every build program is perfect. I influenced the chance of them depending on map size and opponent. But I only consider the closest enemy, not all of them. My main attempt was to offer the possibilities, but I wouldn't say that I'm a balancing god. If someone knows the perfect build order for an anti guard rush I'm at least able to simulate it with a build program. For the moment I'm at least happy that they try different stuff to give the player some diversity, but I'm of course free for all suggestions to improve the quality of the rushes.
The second barracks is a dynamic built. He only builds it if he has too many requisition. All second troop production buildings are dynamic, since there's too many variation in the maps and the situations that one all-pefect build order would exist.
The troops attacking elsewhere are mainly a problem of the defense strategy. I know the problem but to be honest, it's too much for me at the moment to rework this too.

3.) You can't imagine how much problems I had with exactly this behaviour. The main problem is the crappy terrainanalyzer routine. The distances I get from this method range from okay to complete nonesense. It's also dependent on buildings since the results change during the game. According to this routine the 'closest' enemy base on Kasyr Luthien is the diagonal base in some situations.
I tried to at least decrease the problem by calculating a roadmap at the start of the game because the ways aren't influenced by buldings and troops at this time. But this was not enough, I was also forced to average the result with the direct airline distance to at least reduce the 'complete nonesense' results. On most maps it works okay so far, but not on all. I've seen it on testing grounds too, and I nearly started to cry as I saw it. I fear I have no ideas anymore to solve this problem. The only thing that would help in my opinion is a Relic fix for the terrainanalyzer routine.

4.) I've pretty much reworked the jump code a hundred times but as you've seen it's still not perfect. The super short jumps should only occur if they are in direct combat. In this situation they couldn't move away even if there's a better victim. But I admit, that the better solution would have been to modify the infantry tactic. But it would have also been a lot more work since I would have been forced to add all the situation analysis code. I think it's at least much better than before, though I agree it's not perfect.

5.) I didn't touch the turret placement code. The front LP is on several maps at the start a good idea, but later for distant LP's it's not that great. I agree, the placement code is not perfect, but I think it could be much worse.


There are even more issues. Yesterday I had to rework the take and hold code, because the AI wasn't clever enough to reliable decap a critical and the timer wasn't reliable activated. It's still not perfect, but it has to be enough for this release. To be honest, at the moment I'm completly exhausted and want 1.6 out and off the table.

[LarkinVB]

I tried to at least decrease the problem by calculating a roadmap at the start of the game because the ways aren't influenced by buldings and troops at this time.

That was the reason I added cpu_manager:GetShortestPathingDistance() some month ago. It does count pathing distance from all bases to all LPS and all bases to all bases at gamestart. One problem (which I wrote to Relic buglist month ago) is that cpu_manager.terrain_analyzer:GetPathingDistance() is returning strange values since DoW 1.2 or 1.4 (not sure) They are much smaller than they used to be. Something like 80 from furthest base - base at Kasyr Lutien. You can still use them for comarison but can't use them as absolut values if you still expect values > 500.

To be honest, at the moment I'm completly exhausted and want 1.6 out and off the table.

I completely understand. It was same for me with 1.3 and 1.5.
I just fear that 1.6 may be slightly inferior to 1.5 with certain setups. You told Excedrin that 1.6 (like 1.5) isn't well suited for 1vs1 with good players. So it should be good at big battles BUT some build programs seem to be inferior to 1.5 using a general fast vehicle tech approach.

I will be honest. I will use parts of 1.6 for dowpro AI (thanks for your ok) but the build programs just don't cut it for me - yet.

Do you remeber our discussion while 1.4 was slow teching to vehicles ? You stated that fast vehicles are very important - and you were right on. So I tried to streamline 1.5 to keep a basic army just to survive going tier2 to vehicles. Players seem to like it though good players killed it fast with rushes. I just fear that infantry heavy build programs just don't cut it. The AI is dead on doing them BUT fails taking advantage for map control and resource dominance. Therefore it gets behind in the tech race, much more than with 1.5 IMO.

I just wanted to let you know. I would love to see 1.7 using 1.5 flexible tech to vehicles with build programs triggered by setup and AI fighting for map control. Currently AI (both 1.5/1.6) is too passive fighting to gain the upper hand at SPs.

Thanks for your great work and good luck for the future. I was a bit pissed while you took 'control' of the AI mod by your superior coding abilities while I felt it was kind of 'my' baby but I see it is in very good hands and I found my place with DoWpro. Don't dare to invade this :-) !

[ArkhanTheBlack]

I just fear that 1.6 may be slightly inferior to 1.5 with certain setups. You told Excedrin that 1.6 (like 1.5) isn't well suited for 1vs1 with good players. So it should be good at big battles BUT some build programs seem to be inferior to 1.5 using a general fast vehicle tech approach.

That shouldn't be the case anymore since I restricted the fast vehicle build to 4player+ matches.

I will be honest. I will use parts of 1.6 for dowpro AI (thanks for your ok) but the build programs just don't cut it for me - yet.

Believe me, the point you don't like are some rush builds and I admitted that they might not be perfect, especially the marine build. But you don't have to use rushes. The build programs without rushes do not have any disadvantages compared to the old system, only advantages like rebuilding bases and restricting troop builds if necessary. And the dynamic researches and builds were pretty much your idea, because I had to learn it the hard way that completly static builds aren't perfect all the time. And considering the marine rush..., I'm honestly thinking about removing it completly because I got a similar impression like you: The AI s*cks with them. Thud what do you think?

I was a bit pissed while you took 'control' of the AI mod

Yes, I suspected it... But I know the 'my baby' feeling.