122 replies to this topic

[Flenser]

1. I noticed a single AI cultist/scout/slugga squad standing near hq in a 4vs4 at Kasyr Lutien doing nothing while other squads of the same AI was fighting elsewhere. Any idea, Flenser ? They were 3+ strong so it couldn't be the new code preventig small squads to be included in attacks.

<{POST_SNAPBACK}>

Was this near the start of the game? If so, then it is a problem with the AI not getting out of BuildOrderStrategy quickly enough. While it is in BOS it will neither attack nor defend - in the sense of creating Attackplans and Defendplans. Fortunately any threat to one of its bases will cause it to leave BOS immediately.

I would guess in the case you describe that (if it was near the start of the game - remembering that it can take the AI a while to work through everything in BOS) the AI had made all the capture plans it wanted to make so there was nothing else for the squad to do. If they were attacked they would have fought back, but otherwise they will just stand there until BOS ends and BBS starts.

If it was not near the start of the game, then I got nuttin'

Flenser

[oozish]

well, I just played a game with resources set to 'high and indeed there is NO problem building vehicles if you do that. However, i hate using this option it's totally unfair and takes away the strategy of the game in my eyes.

However, for the record, if you increase the resources from standard to high, the chaos did build plenty o' vehicles and advanced units on blood river. I just didn't enjoy the game.

Edited by oozish, 13 July 2005 - 10:41 PM.

[Markoso]

Just got back from work and finally have some extra time, I'll give it a test myself, it'll be wonderful to see 1.3's "issues" fixed.

[Obake]

Oozish, I'm not quite sure how you figure that the game becomes 'un'fair' and takes away the strategy element with the resourcing rates increased. You are still gaining resources at the same rate as the AI. In fact, everything remains equal with the exception that SP's and the like are worth more, enabling the AI to stay on par with a human player.

I'm not trying to be sarcastic, I'm genuinely curious about how you arrive at the conclusion.

Speaking for myself, I find that a higher resource rate makes the game more challenging, as the AI is less predictable. With money to spend, it will do things that it wouldn't be able to do under normal resource rates unless you sat back and let it have map control.

[oozish]

Oozish,  I'm not quite sure how you figure that the game becomes 'un'fair' and takes away the strategy element with the resourcing rates increased.  You are still gaining resources at the same rate as the AI.  In fact, everything remains equal with the exception that SP's and the like are worth more, enabling the AI to stay on par with a human player.

I'm not trying to be sarcastic, I'm genuinely curious about how you arrive at the conclusion.

Speaking for myself, I find that a higher resource rate makes the game more challenging, as the AI is less predictable.  With money to spend, it will do things that it wouldn't be able to do under normal resource rates unless you sat back and let it have map control.

<{POST_SNAPBACK}>

maybe because I'm used to playing with resources at the standard level, it feels totally different, i don't have any idea how fast the resources are coming in. It completely changes the feel of the game for me. That said, 'cheating' isn't the right word, if the resources are the same for AI and human.

I didn't know they were....I"ve never played that way.

To me, using that option makes it like a new game; which isn't necessarily a bad thing. It's just I have tactics familiar with standard resource production that work vs. the AI as I've come to know it. I like to control the LP's ect., as you correctly commented in a previous post. Once you change this option, my old tactics don't seem to work anymore..probably because the AI get's so much resource from it's captured LP's I can't impact it anymore by my normal strats. Maybe that's more a reflection of my weakness if the AI is truly let loose, but nevertheless, it's not the same feel as I've come to know the game. Gimme a break, i'm resistant to change.

I"ll probably keep it on high if that's the only way I can get the AI to consistently make vehicles. To me, the vehicles are the coolest thing about DOW.

[thudo]

I'll be submitting RC2 to the DoS ftp site in a few hours then I believe we're releasing as its just mint. I tweaked the builds for various buildings and have been playtesting.

Just to let ya know Oozish: every map I've played the AI builds vehicles BUT I'm probably not as super-aggressive as you are in 1vs1. 2vs2 and above is where I love playing as the AIs for some reason are just bloody animals!!!! Something about DoW's AI and playing together as allies makes them pro-rate themselves differently OR maybe they share resources together? Hmmmm..

Anyway.. we're on-track now. Hell.. Larkin is happy too so I'm busting out baby! BUSTING I TELL YA ! ! ! hehehe

Edited by thudo, 14 July 2005 - 03:31 AM.

[thudo]

Update: RC2 is on the site! There are some minor changes that should make the AI even more challenging. Don't expect miracles - we're still under the spectre of v1.30 but the AI is 80% better than without these fixes. VERY VERY pleased with how it turned out.

When Relic does fix this shiet the fixes can easily be removed - one of the bonus things about LUA scripting.

So go grab.. remove your old \AI folder on yer system and install this into a new \AI folder. LETS GET THIS THING RELEASED ! ! ! !

[LarkinVB]

I checked 1.2RC2 and it looks better than 1.2b9. I wonder wether neither Flenser nor me had the idea to ignore the complicated prereq system and just build missing buildings as soon as possible. It's not as flexible as the old prereq (dynamic build vrs static build) but IT WORKS

However Thudos new building code may need some more finetuning after release of v1.2

Two points which come to my mind :

1. It should check for a minimum of squads on the field before trying to build buildings.
2. The new code makes part of the prereq system obsolete so it might be good to check if BuildLater() is still needed at all or alternating time can be set back to 30.

Oozish : I think its your playstyle which does prevent AI from building vehicles at small maps. You seem to have constant map control and the AI is just low on resources.
On a second thought this should be true with DoW 1.2 too.

Edited by LarkinVB, 14 July 2005 - 10:46 AM.

[oozish]

I think I agree Larkin. 1.2 vehicles do come out, but not that many; whereas with 1.3 usually I don't see vehiicles at all. As I said above, setting resources to high can solve this problem, tho it throws off the gameplay/strats as I've come to know them.

The cool thing about 1.3 though, is the squads are alot tougher all around. So al in all it makes the game interesting.

[thudo]

Thats what I've never understood: v1.30 works on-par with v1.20 if in v1.30 you throw more money at the AI at a higher rate than over v1.20. So... has the resource gathering rates for either HARDER / INSANE or the Resource Rate changed from v1.20->v1.30? Has it been lessened in v1.30 so that in order to compensate you have to increase it to make it on-par with v1.20? Something is fishy.. it should like its all about the gathering rate to me.

Testing is moving forward and it seems waiting the extra week will help immensely to cut down the inquiries on why the AI under v1.30 didn't preform to spec.

[Flenser]

I think introducing the new building code was a mistake at this late stage. It has caused uncertainty once again as to what the AI is doing. At least with the v1.2 code we knew how it behaved, so we could word the release doc accordingly. Now we either release it with a woolly doc that hems and haws about AI behaviour or we test it for another week and delay release again.

I'm not disregarding your efforts, Thud, or suggesting your changes do not work, just saying that we need to lock code and release and then start tweaking again.

Flenser

[thudo]

I would have agreed with you however I playtested it for 3 hours last nite in various scenarios and each time vehicles came out and did everything v1.20 did. I see nothing wrong with it. Even tried a 4player FFA on >gasp< 4p_Biffy's Peril and the AIs kicked my arse to town even though I had control of my side. Orks struggled more (unfortunately since they were nurfed its hard to keep 'em alive in v1.30, silly Relic.. ).

My code changes shouldn't break anything but at least its fully working well 80% better than on the weekend. We've had 2 solid days of testing from a few of you and I plan to put in just a bit more tonight although last nite was a sweet marathon for me testing the final tweaks before publishing to the FTP.

I think we're really good especially given the angst that which is v1.30. I feel far far more comfortable releasing it this way then throwing pre-Monday code out there that works so half-as*ed in a game code everyone likely has. Yes I wish I did this sooner but the scripting ideas came to me and eagerly wanted to ensure our project worked well with the latest game code.

I think we're ready for launch.. playtesting on various maps, in alliance games has proven it. The Chaos AI has never been so good since v1.20.

I think we'll all agree: its better now than before Monday. Chaos was a joke before then. At least now its super-competitive and near par with v1.20.

[LarkinVB]

Now that we (Thudo) made some drastic changes I vote for another week of testing and tweaking to optimize the new direction.

I modified order sequence in buildbasestrategy.ai to great effect (IMHO).

function BuildBaseStrategy:Update()

	profile_start("Update")
	
	--superclass
	Strategy.Update(self)
	
	--always want more strategic points
	if self:PlanCount( "Resource Plan" ) < self.info.resourcers then
  self:AddPlan( ResourcePlan() )	
	end

	if self:PlanCount( "Capture Plan" ) < self.info.resourcers then
  self:AddPlan( CapturePlan( cpu_manager.cpu_player:GetStartingPosition() ) )	
	end
	local req = resource_manager:GetResourceAmount():Get( ResourceAmount.RT_Requisition )

    self:DoBuildBuildings()
	self:DoBuildUnits()
	self:EvaluateSquadCap()
	--only research if enough req
	if req > self.info.req_reserve then
    self:DoUpgradesAndAddons()
	end
	self:DoBuildGenerators()

	self:FinishBuildings()
	
	--can be flagged to not build secondary buildings for easier difficulty settings
	if not self.info.no_secondary_buildings and cpu_manager:IsInSecondTier() then
  self:DoSecondaryBuildings()
	end

I also changed midcapture only to be valid for the first squad, resulting in much faster capturing of nearby points.

I suggest we wait to see what beta testers say to RC3. I'm currently busy testing/tweaking it.

[thudo]

Sounds good considering the considerable progress we've just made by reinvigorating our project under v1.30 -> that says alot. At least we can subtly say "Screw Relic and their lack of AI patches, we'll work AROUND the bug".

I'll test the above as well. I'm glad we're working at this once more although its BS we had too in the first place BUT since Relic isn't obliging its more sugar-added for us. No Game Code is gonna put this team down!!!

EDIT: So essentially the changes you did look like:

       self:DoBuildBuildings()
	self:DoBuildUnits()
	self:EvaluateSquadCap()
	--only research if enough req
	--if req > self.info.req_reserve / 2 then
	if req > self.info.req_reserve then
    self:DoUpgradesAndAddons()
	end

        --only build generators if enough req
	--if req > self.info.req_reserve / 2 then
    self:DoBuildGenerators()
	--end

	self:FinishBuildings()

You moved up DoBuildBuildings() to the top of the heap (does the order make a difference then before?) and made is easier for the AI to build Generators. You also made it slightly harder for the AI to build upgrades/addons but thats no big deal.

Edited by thudo, 14 July 2005 - 06:42 PM.

[LarkinVB]

It's all about the right balance between building the right stuff in the right order. My code seems to work better. Even better would be a dynamic order according to map size, number of players, type of ememy, quickstart etc. pp.

Good to have for v1.3 but we should at least find a good static balance.

Here my ratio (chaos as example) :

1. Put buildings at top to ensure armoury/circle/pit as soon as possible
2. Squads, we need squads as much as possible
3. If we need more cap we should get it.
4. Now we can research for our squads as we have some and of best type.
5. Now we check for needed generators and since we are at bottom of queue we don't need another req check for them.

I tested and it played nice. Again it would be better to make it ALL dynamic on circumstances but this is MAJOR work with 80% testing on various maps/configurations and just 20% coding.

We have to test, check for wrong build order and correct. Rinse and repeat. My 2 cents.

[Flenser]

Guys, this is great, but we are never going to release this if we keep tweaking it. Each tweak = more testing.

Read that Valve guide to mods I posted the link to. They probably know what they are talking about.

Flenser

[thudo]

Thats why this must be a limited tweak. We should give it a couple of days and then we're done. We can keep re-tweaking every little bit but then we're just holding back a pretty-much solid product anyway.

Lets wrap this puppy off and ship 'em out !

[oozish]

...says oozish (above)

[thudo]

They will be in a couple of hours. I'm testing them for the moment. Hang in and RC3 will be out in a moment.