34 replies to this topic

[thudo]

Morning UserBytesBack...

hehe.. more leaked betas of our work. Love it. That build ya had was old and should have been retired.

When our AI is published, anyone can use it as they wish although it would be honourable to give the team some credit for the work so acknowledge us if you could.

I'm not sure what build you played but the AI does have attack/retreat routines already there. Oh sure he'll send a throng of squads/vehicles "into the maelstrom" but the AI does retreat - even at the squad level. I mean, we could script it so the tolerance level of the AI is lower so he'll save himself more quickly then waiting for more of his own to perish.

[Corsix]

How about an encoding the zip files and only gives passwords to the team to prevent leaked betas?

[thudo]

As long as the team here can receive PMs from me (to get the username/password to access those zips) OR I send it in email directly. Fortunately, any leaked betas seem to be restricted to a few people at least.

[Corsix]

Or have the password for each zip to be <common password><build name> so if our common password was "fredbob" and this build was nicknamed "thudo0.64" the password would be" fredbobthudo0.64"

[thudo]

hehe.. well now the secret is out! I think its best to email it to everyone on the team with the password.

[thudo]

I believe the difference in the way the AI does things in T&H and CA victory condition lies in the Skillset files.

For buildbasestrategyinfo.ai.. this is what it is for HARDER/INSANE:

 --how much requisition I have before I try to build an uber unit
  uber_unit_cost = 900,
  
  --minimum amount of power before I build another generator
  min_power = 50,
  
  resourcers = 2,
  
  --reserve this amount for building units/buildings (not for use in upgrading/ reinforcing)
  req_reserve = 120,

  ETA =
  {
 	 cost = 250,
  },
  
  --attack rating needs to be greater than this prorated to attack
  attack_rating = 200,
  
  --need at least this at all times
  squad_cap_threshold = 2,

  --need at least 3 at all times
  support_cap_threshold = 3,

For HARD:

 --how much requisition I have before I try to build an uber unit
  uber_unit_cost = 1000,

  --percentage of opponents army to build up to
  max_army_percentage = 0.9,
  
  --minimum amount of power before I build another generator
  min_power = 50,

  resourcers = 2,
  
  --reserve this amount for building units/buildings (not for use in upgrading/ reinforcing)
  req_reserve = 500,

  --attack rating needs to be greater than this prorated to attack
  attack_rating = 100,
  
  --if you have this amount of army, attack anyway
  max_army_size = 3500,
  
  --need at least this at all times
  squad_cap_threshold = 2,

  --need at least 3 at all times
  support_cap_threshold = 3,

And For Standard:

 --how much better they should be before I build an uber unit
  uber_advantage = -1200,
  
  --how much requisition I have before I try to build an uber unit
  uber_unit_cost = 1000,

  --don't build secondary buildings that allow you to quickly amass troops
  no_secondary_buildings = true,
  
  --don't win by strategic objective or control area victory
  no_alternate_objective = true,
  
  --percentage of opponents army to build up to
  max_army_percentage = 0.6,
  
  --minimum amount of power before I build another generator
  min_power = 50,

  resourcers = 1,
  
  --reserve this amount for building units/buildings (not for use in upgrading/ reinforcing)
  req_reserve = 180,

  ETA =
  {
 	 cost = 400,
  },
  
  --attack rating needs to be greater than this prorated to attack (always attack)
  attack_rating = -500,
  
  --if you have this amount of army, attack anyway
  max_army_size = 3000,
  
  --need at least this at all times
  squad_cap_threshold = 2,

  --need at least 3 at all times
  support_cap_threshold = 3,

See the differences? I think we need to tweak these setting for HARDER/INSANE skillsets so they are more competitive in T&H and CA game modes .

Also.. Not the attackstrategyinfo.ai file.. for HARDER/INSANE:

 --my prorate against yours before I stop attacking you and retreat
  stop_attack_rating = -200,
  min_units = 2,

For HARD:

 --my prorate against yours before I stop attacking you and retreat
  stop_attack_rating = -100,
  min_units = 2,

  --if you have this amount of army, attack anyway
  max_army_size = 1500,

and for STANDARD:

 --my prorate against yours before I stop attacking you and retreat
  stop_attack_rating = -1500,
  min_units = 1,

  --if you have this amount of army, attack anyway
  max_army_size = 800,
  
  --percentage of opponents army to attak with
  max_army_percentage = 0.5,

I believe we aren't far off in fixing this problem!

Edited by thudo, 09 February 2005 - 06:55 PM.

[Corsix]

Then hurry up, fix it, and make the release...

[quiet_man]

Then hurry up, fix it, and make the release...

<{POST_SNAPBACK}>

better to do a release with a known bug than with many unknown

[thudo]

Just doing some final code testing..

Can you check that math.random thread regarding rebuilding and re-upgrading destroyed turrets? I think as Giskard mentioned we need an Update function for those so they get re-updated upon re-construction.

Btw, the random turret upgrader updates based on the final total of turrets right? So if the AI picks 14 turrets to build in the game, then he updates 7 of them to Tier1 anti-vehicle (based on math.random 50,50). Once turrets are destroyed and recreated they fall out of that realm and are considered just regular turrets. Thanks!

[thudo]

Well I did some testing with a 3vs3 and 4vs4 (oh man.. 4vs4 is sooo incredible!) and the AI was actually upgrading turrets to Tier1 (anti-vehicle) as it expanded to captured LPs. Wierd. Maybe its a money issue as the map I played yesterday was VERY limited to the # of LPs but this one each starting point had 4-5 LPs to start hence his ability to upgrade.

Nevertheless, a good test for the AI is playing the 8p_Penal Colony map.. Oh man.. slim pickings! The AI is still brillant on that map as it both satisfies both the infantry aggression Larkin is looking for while being also defensive and good at base building. That map is a toss up as resources are VERY limited so its up to many elements to surviving that one.

Anyway, UserBytesBack.. we'll still doing internal testing - you can understand its among the project members. A release is iminent - whether the T&H and ControlArea game modes are fixed before then is unknown. I should look at it based on my message below. I'm just enjooying the full AI test at present.

We will be looking at the T&H and ControlArea game modes - thats one of our objectives. Its probably due to those parameters mentioned in my second last message.

[thudo]

Btw.. another section that may shed some light on why the AI isn't aggressive on T&H and Control Area..

buildbasestrategy.ai

function BuildBaseStrategy:__init( baseinfo ) super( baseinfo )

	--set maximums, if needed
	if self.info.no_alternate_objective ~= nil then
  CapturePlan.Maximums.points =	resource_manager:GetTotalNumStrategicPoints() / 2
  CapturePlan.Maximums.objectives =	resource_manager:GetTotalNumStrategicObjectives() / 2

The thing that boggles me is only Easy/Standard difficulty levels are the ones with the

buildbasestrategyinfo.ai

             --don't win by strategic objective or control area victory
  no_alternate_objective = true,

I think some minor testing is needed here to finally fix it. Alot of people do play T&H and ControlArea so some resolution would be cool..

Edited by thudo, 10 February 2005 - 07:42 AM.

[thudo]

Thanks UserBytesBack.. I will run some quick tests tonight to fix this since we are so utterly close to release. To fix the AI's behaviour in T&H and CA victory conditions is one of our main objectives + many people out there do play these conditions as they find simple Anhiliation boring.