6 replies to this topic

[thudo]

W00T ! ! ! ! !

Just had a 2vs2 Direct Connect Game (2 Human SMs vs 1 AI Chaos+1 AI Eldar) on HARD skill (all skillsets are now modified to work properly now) and the turret upgrade randomizer worked flawlessly! No issues. AI upgraded all tech as usual. No synch errors as I thought there might be. Map space around Startpoints was tiny so AI was decent at base design (this 4player map has a little smaller reestate then normal) but otherwise it was fab. NOW IMAGINE PLAYING HARDER+ Skills! OUCH!

So we're good with MP! Yippee!

[LarkinVB]

Back from LAN party.

Observations :

1. Couldn't play with -modname dow_ai. 'Using different version error'. Had to copy into W40k/data folder and then it worked. Since we also used Aetherfox balance mod I'm not sure which caused the trouble.

2. In general harder AI played well but it heavily depended on map. Played 4 vs 4 AI most of the time. Eldar seems to be the weakest as AI can't take advantage of some of its powers. It was strange that some AI build full grown bases while others were significantly smaller though we played with shared resources.

3. AI needs to learn teamwork as it didn't support well if one of their bases is under attack.

4. I suggest discussion of buildorderstrategyinfo.ai and buildbasestrategyinfo.ai as these are crucial for the AIs research/attck/defend behavior. We should get rid of any research in the tree which the AI can't hande. We should also optimize the initial build order. Building turrets before barracks isn't smart IMO. Even better we should make buildorder variable depending on mapsize and distance to next enemy base.

5. I would like to get information on how the troop value is determined as lots of scrips use something like

 --attack rating needs to be greater than this prorated to attack
  attack_rating = 300,

Do you know the rating of a full sized space marine squad without heavy weapons ? Is it hidden in unitstats.ai ?

[thudo]

Apparently, the reason for the "you have a different version" is due to the .module file having the "ModVersion = 0.1h" designation. It should like: "ModVersion = 0.10" with no letters after the = . That should fix that.

Yea I usually test with the HARDER skillset and its generally excellent. Of course its dependant on the map where sometimes it excels while others plop a different civ there and it'll build differently. Don't know why the Eldar side of things is always the weakest.

Definitely agree about AI supporting each other to help defend. This, however, is hit-and-miss but generally I'm seeing the AI come and help me if I'm under attack ALTHOUGH if my base is closer to his (map dependant) this happens more often.

The research tree is fine and should hardly impact the AI. Its essential to keep it intact as its a test to ensure it works for future modders. It can be adjusted to go between buildings so one building isn't researching 2-4 items in a sequence. This can be tweaked but seeing the AI research everything (even if some seem trival) is key to ensuring reliability. Yes Grot Infiltrating seems silly but other things like Banshee Warshout are not (AI does use it).

BuildOrder is always changing. I find if the AI does build one turret then a barracks this is ideal as at least he'll have a turret to defend and troops to complement it.

The attack_rating thing never really made sense to me except it seems to say: I can only attack if my force is better than the enemy's rating.

Still.. AI has undergone some changes since you last downloaded the last build. Been working on it for 3-4 hours per nite. Coming along fine. Sometimes inconsistent but usually excellent.

[Corsix]

I have modded a file and set attack_rating to a nice high number - the result: the AI dooesn't attack me; it builds a large force, puts it in the middle of the map but doeesn't actaully attack.

[thudo]

I have the SM, CSM and Eldar at 200 Attack_rating. Ork are at 300 and it works out just fine. Total destruction all around. Don't see any difference regardless with setting it between 200-300. Haven't tried any lower or higher but right now they're pretty brutal regardless.

[LarkinVB]

I found the AI is attacking later with more force while I had it on 300 during our party. Also did set retreat from -200 to -100. But this should be checked by debug log or rec comparison and not by watching a single battle.

[thudo]

Attacking later with more force with a 300+ attack_rating makes sense. It prorates the enemy's value (you can see it in the on-screen ai debug) and compares it to its own army so when it goes over a certain value then it knows it can compete for a successful assault. Mighty brilliant idea on Relic's part actually.

I would prefer the AI to be a little "ballsey" and risk a full frontal fight. I mean, we could randomize this value from 100-400 if we wanted to so no game would ever play the same. It would be a rusher verses a build-up guy who waits around first to amass then strikes. Lots of possibiltiies!