Enhanced_AI_v0.26.35:
http://peppe-dg-mods.googlecode.com/files/Enhanced_AI_v0.26.35.zip
http://peppe-dg-mods.googlecode.com/files/Enhanced_AI_v0.26.35.exe
AI will work hard to get map control.
AI will actively work to keep portal counts even or in its favor.
AI will save for critical upgrades like currency, so it can purchase it at or near warrank 3.
AI generals will consider what healer idols the rest of the team owns when making purchases.
AI generals will send one of their healing idols to an AI teammate on summoning (when 2 are summoned) and on teleport.
Just did a local test on Regulus and all of his snipe action had Killshot = true. This apparently means only use the skill for a killshot. Took killshot off the hero target function and it is fire at will now. The default function finds the weakest hero in range, so not really anything else to do.
Put 4 on my team, took off fog of war and they sniped from the center of the map to demigods retreating by the corner flags
I think the AI respects fog of war, so should I add that skill that tags a target, so he can snipe demigods that back into the fog?
During one more test (sporting a quite... umm, weird conception by Darkliath) we were playing at standard Pacov settings on Cataract. The teams were: Oc (human)/TB (human)/TB (AI) v Oak/Oak/Oak (all AI).
Now, the AI was performing quite well, but I've noticed there is a very strange dynamic to its flag re-capture behaviour. The AI doesn't seem to react at all to neutralized (yellow) flags. For example, its Celerity flag was yellow for half a game. I've tested it by neutralizing its gold flag on purpose. The same thing occured: the AI didn't rush to retake the flag in 4-5 mins...
It used to do that because it was flag-lazy, but now it probably does that because it specifically looks for enemy flags to capture. Should probably have it looking at portal flags first (unless very far away from its citadel), then neutral flags (also unless far from its citadel), then a normal enemy flag.
I've noticed that as well. In some cases standing in front of a flag it could simply finish capping, but instead chooses not to. I wonder if its possible to change the logic so that flags in their proximity that are half capped have a higher priority than enemy controlled flags. Not sure if that logic change would cause other problems, though.
That should be an unnecessary exception - only portal flags should have higher priority than neutral flags. Might have to do the distance check for all flag capture actions from both their position and their citadel though (if this isn't already done; I think it might be), to catch flags that they're near enough to, even though they might not be near their base.
Basically, what's happening when they move off of an uncapped flag prematurely (which I don't think I've seen happen when it's clear they went specifically to cap that flag) is that they're being assigned another, higher-priority cap-enemy-flag action, when they should be assigned any nearby / near-base neutral flag cap actions first, then enemy flags.
Every branch of the flag search includes enemy and neutral flags.
Who owns contested flags? Does it have to get all the way to the breakpoint to be 'enemy owned' otherwise it stays ally owned, but contested?
It could be nuetral owned is only until its captured once, then after that if its yellow it is still technically owned by the last team to push it past the breakpoint.
If contested flags that never changed ownership stay in the ally flag list i will need to loop over them and find flags not 100% owned and add them to capture list.
Edit: seems to be counting everything correctly for my offline version. As soon as it crosses midpoint it becomes a neutral flag and part of the capture list.
The flag search might include neutral flags, but it isn't pushing their priority above that of other non-portal enemy flags. I think it should be. I would break neutral flags out of the combined flag search / randomization routine, do them first, but if they're 'hostile' (radius 20 tower check?), then add them into the 'enemy flags' table, which it then proceeds to find a random flag from or however you have it doing that now.
Neutral flags should always belong to the neutral civ army, and thus be gathered by a neutralbrain:GetListOfUnits. They may have a class var on them that shows their last ownership, but I don't know if that should be relevant for purposes of capturing them - we don't want to prioritize prev ownership, as it could result in say, excessive pressure on a random enemy flag that some AI capped and then lost, while other more accessible flags go uncapped.
If you want to specifically prioritize flags the AI 'should' own, there should be a way to search one of the Scenario tables containing markers for the flag unit's Name, checking if it belongs to a particular army. But so far, the only problem I've seen the AI have with recapping its own flags is the accidental-neutral ones (where it goes to attack a demigod, chases the demigod off, and neutralizes the flag - for example).
There is also a deadzone on the flag search. I think it was 800 units, but i upped it to 1200. I think in these units a tower shot is about 600. Anyway inside this zone the flag closest to the unit is selected.
How about add nuetral flags to the capture list, if its empty then add enemy flags and let the routine run as is?
Changes coming in H (before i forget them):
Attack demigod actions tweaked to chase in some cases, but retreat in most when towers are encounted.
Regulus Snipe skilled allowed to fire more often than just a kill shot. Skill used only for demigod targets.
Regulus achievement items updated. Removed Renewel added blade of serpent.
TB, Frost Nova and Deep Freeze used mostly for interrupts or severely damaged heroes (to slow them down).
TB, Fire mode/Ice mode Switch based on skills he has rather than random.
Sedna and UB, Interrupt skill usage tweaked.
Process to select flag to capture updated. Tower threat scan range increased. Should result in ally flags being recaptured more consistently as their threat is low.
Removed cloak of night from general's achievement item pick list. Health all the time seems more useful than flee occasionally for the AI at least.
Queen Shield allies: Added condition that ally has to be below 75% health. Maybe she won't blow through all her many as quickly.
Generals Sharing Priests: AI will inspect ally generals and not send a priest to that general if they both own the same level idol. So if you are an assassin with two normal generals at the start, they will both buy monks, they will see that eachother has monk, and each send their spare monk to you.
Probably other stuff i missed.
I'll try to pay more attention if I see it happen again.
I'll likely only have an hour or so to do some testing tonight. Will be free in about 2 hours for about an hour - so, if you have a version ready then, I'll give it a whirl. Otherwise, I'll just review the ability builds and provide some feedback/suggestions there.
Is TB faster in ice mode? I have in there if doesn't have any ice skills, but an enemy is near and he is low health that it is ok to switch to ice mode.
26h:
http://www.filefront.com/17590554/Skirmish_AI_v26h.zip
http://www.filefront.com/17590557/Skirmish_AI_v26h.exe
No, unless he has at least Permafrost or Frost Nova, he shouldn't do this. He should only be switching to ice mode with no skills momentarily, for the weapon damage / mana boosts. And definitely not when he's pursuing / fleeing.
Yeah, that's more or less what I was getting at. If there are any non-dangerous neutral flags, always get those first.
Oh, and yes, I think you should add tracker to a few of the Regulus builds if he now snipes (and if no one else already answered that).
Will be able to test in about an hour.
TY!
also, i agree with pacov, definitely add the tracker, as it is beneficial for allied demigods as well
Fire TB is slower in ice mode. The fire aura benefit (which increases movement speed) lasts for a couple seconds after the switch, and it is sometimes beneficial to switch to ice mode to get the additional weapon damage, but I would say that for A.I. it would be better just to stay in fire mode unless switching back and forth as previously stated to get the 2 buffs.
AI skirmish (mod version) .26h/uberfix 1.03 (last version published by miriyaka)
Settings: cata/High Towers - all other settings on defaultAI settings: normal3v3
rook (human)/reg/qot v rook/reg/qot
Observed results:
noticed something strange right of the bat. My reg made its purchases and announced its build. Then sat at the shop for a little over a minute. Never bought anything else - just sat there. Then, after a min or so, he gets to work. Not sure why he was just standing around.
reg went - sniper mines build. Snipes weren't great, but did see the ai starting to snipe some, so its an improvement.
ended game early as rawrr joined me for testing... doing another test now..
Settings: levi/High Towers - all other settings on defaultAI settings: normal4v4
rook (human)/tb(human)reg/qot v rook/tb/reg/qot
ai reg on our team had autoattack build... but sniped... and did not use angel wings... eg its a bad build or its not being used right (angel wings should be popped immediately.
we watched a replay to see what he did... they both had 2 and 3 tp scrolls early game (wayyyyy too many). Our reg's build was crap. I need to review the reg builds and just keep the good ones.
Saw our reg give up an easy kill on a rook with that had 300 hp. our reg has had full hp and mana, but did not pursue.
Not much else observed - we watched the reg's primarily...
playing 1 more.
I think there is something in the sniper_mines build causing it to hanging running 4vs5 Regulus games and seems to be that build. I'll see if something looks funky about it.
Seems to be tied to the finger of god achievement item. Add that to any class's build and it kills it.
26i:
http://www.filefront.com/17591534/Skirmish_AI_v26i.zip
http://www.filefront.com/17591533/Skirmish_AI_v26i.exe
Regulus Updates: removed finger of god to prevent him from getting stuck in the first shop. Put Auto_Attack build back to an earlier skill plan. No snipe and Angelic Fury maxed first.
Tested all demigods opening shop and sped up shopping speed to .6 seconds per action. Unmodded is about .2 seconds for most actions. Older version of this mod had a full second per action.
TB: Removed going to ice form if it doesn't have any ice skills.
Thanks for another release. I'm going to start looking at ability builds now. Anyone know where the file is? Going to dig around, but I imagine I'll find it shortly.
edit - found it. Taking a look.
Skirmish_AI_v26i\hook\units\heroes\
If you have windows 7 go to that folder and search for *AIBlueprint to get the blueprint for each demigod. If not just go into each folder and load the blueprint file in a text editor.
Here comes an ugly paste. New skill tree for DA:
SkillBuilds = { Warp_Spine_Swap = { 'HDemonWarpStrike01', # 1 'HDemonSpineAttack01', # 2 'SAVE', # 3 'HDemonWarpStrike02','HDemonSpineAttack02', # 4 'HDemonShadowSwap01', # 5 'SAVE', # 6 'HDemonWarpStrike03','HDemonSpineAttack03', # 7 'SAVE', # 8 'SAVE', # 9 'HDemonWarpStrike04','HDemonSpineAttack04', 'HDemonShadowSwap01', # 10 'HDemonWarpArea01', # 11 'HDemonWarpArea02', # 12 'HDemonDemonsSpeed01', # 13 'ADD THE NAME FOR EVASION HERE', # 14 'HDemonDeadlyWarp', # 15 'HDemonDemonsSpeed02', # 16 'ADD THE NAME FOR EVASION HERE', # 17 'HDemonDemonsSpeed03', # 18 'ADD THE NAME FOR EVASION HERE', # 19 'ADD THE NAME FOR ASSASSIN's SPEED HERE', # 20
Updated sniper_mine build for reg:
sniper_mines = { 'HGSA01Snipe01', # 1 'HGSA01ExplosiveMine01', # 2 'SAVE', # 3 'HGSA01Snipe02', 'HGSA01ExplosiveMine02', # 4 'HGSA01Betrayer01', # 5 'ADD NAME FOR TRACKER', # 6 'HGSA01ExplosiveMine03',# 7 'HGSA01Snipe03', # 8 'SAVE', # 9 'HGSA01ShrapnelMine01', 'HGSA01Snipe04', # 10 'HGSA01Betrayer02', # 11 'HGSA01Maim01', # 12 'HGSA01Maim02', # 13 'HGSA01Maim03', # 14 'HGSA01Betrayer03', # 15 'StatsBuff01', # 16 'StatsBuff02', # 17 'StatsBuff03', # 18 'StatsBuff04', # 19 'StatsBuff05', # 20
AI skirmish (mod version) .26i/uberfix 1.03 (last version published by miriyaka)
rook (human)/reg/reg v rook/reg/reg
hmm... regs performance was so so leaning towards bad. I'm thinking that based on the way the ai plays, it might be best to have the ai always pick up a level of scope so it can peck away at towers at the least (probably at level 2). Snipes still really aren't where I'd like them to be. I'm seeing alot of super close snipes (can be good, but generally really bad). Watched 1 reg walk right up to a rook and choose to snipe while rook was swinging away at him. Wonder if we can get some overrides in place. For instance, perhaps its ok to snipe up close if the opponent is down to 10% hp or so. it's probably a difficult rule to program, but the underlying principal is that its always best to snipe at max distance. I didn't see any ai spamming snipes on a single target ever, though. I'd be pretty happy if we could see that working. Maybe having tracker in kits will help that, but we'll see. The way qot is shielding (still needs some tweaks, but good) makes me want to have her as an ai teammate. If we could get reg so that he's harassing regularly with snipes, then I'm game.
saw a rook port into a recently locked flag and try to relock it while the lock was still on there (eg no effect).
question about portal control logic. I took the ai's portals 2 times prior to ws 6 or 7 or so. The ai made VERY little effort to try to retake. Is that intended?
I tried to take a portal late game and the ai was all over me.
Another thought - ai rook could really use a build tower farm routine. It would start after it has lev 3 towers. Basically, the ai would camp a flag IN A LANE (eg creeps are coming by) and it would spam towers for like a minute or so.
Also, a farm routine would be good for early/mid game. Basically, this means that an ai would plop down in a lane or just behind one of its towers and wait for creep waves to come back and attack them. Generally, when farming, unless you are quite strong, you avoid fights with other dgs. This would would be a short override if the ai is 1) behind on levels or 2) has > flag control than the other team. Anyway, farming makes a big difference imo (in good MP games, it means the difference between a win and a loss).
There are many great features available to you once you register, including:
Sign in or Create Account