Who is good or bad to have buy ash?  I can do what miriyaka suggested and make it a priority for all assassins, or a priority for everyone and remove it from specific builds, or leave it at 0 globally and add it to specific builds.

another thing.  On nightmare, the ai is very reserved.  It should spend more money on upgrades imo.  Too often do I see the ai walking around with crazy amts of money. 

Another thought - and this is batshite crazy - can you teach the ai to buy and drop items at the crystal?  If it is loaded (eg nightmare), it could probably buy items and drop them for teammates, which could add a fun factor for the team to run back and pick up, etc.  This isn't a "please do" requests - its a can you do request.  Thanks.  And of course, we wouldn't want the ai to pick up what it dropped. 

26p:

http://peppe-dg-mods.googlecode.com/files/Skirmish_AI_v26p.zip

http://peppe-dg-mods.googlecode.com/files/Skirmish_AI_v26p.exe

With citadel upgrades, journeyman, and godplate Nightmare AI has plenty to shop for now.  Should not float as much money mid game.  Ashkandor added globally.  Can refine later, but the nightmare AI's I tested didn't get to 18k gold to buy it before the game ended.

Updated globals on some artifacts and upgrades to miriyaka's latest formulas.   AI pretty much buys the whole upgrade tree.  Troops and their armor/attack upgrades are delayed until 8.

There is a drop function in the files.  Not sure if the AI can call it or if its only for the player through the UI.  Don't think it will be of much use if the AI is buying upgrades and is going to have enough items to survive well.

Conducted a Nightmare AI test on Cata with standard settings + High towers (v26p + Uberfix 1.03, of course).

UB (h)/Sed v UB/Sed (UB, since Darkliath recently mentioned AI-controlled UB is 'weak', probably because of spending on cit upgrades, which won't happen in any non-noob MP game)

It was actually a fun game. The opposing team acted more human-like than ever. Silence okay (maybe even good!), Pounce hard to judge, as both Sednas went Pounce-Heal (so it didn't stay at level I for long). Fire focusing all right. Sedna/UB interaction all right. And, well, don't think the AI UB was weak. Overall, very positive impression.

Problems:

(1) Capping vs destroying towers. Well, no human player will just pass hp flag and concentrate on the tower. Likewise, I don't think AI tries hard enough to capture portal flag. They fought to death over the right to destroy towers in gold flag proximity, and over the flag itself (reasonable, but human would clearly prioritize portal -- they could have pushed me out easily, have they tried).

(2) The counter to having a portal captured should not be automatically "capture enemy portal". My AI did not lock the portal, after all. However, the AI spent a lot of time trying to capture our portal instead of retaking its own.

(3) No idea how on Earth could my AI purchase all the creeps before the opposing team (2 Nightmare AIs should be > 1 Nightmare AI in terms of gold and cit upgrades...).

(4) The AI does not care much for its valor flag. I could capture and lock it totally unopposed.

Replay: http://www.gamereplays.org/demigod/replays.php?game=51&show=details&id=172329

If your teams portals are defended, yes... tp and lock as needed.  Priority should be portals first, then valor. 

skirmish (mod version) .26Q/uberfix 1.03/FavorMod 2.0G

Settings:  cata/High Towers - all other settings on default
AI settings:  normal/nightmare
ub(h)/oak(nigthmare)/sed(normal) v occ(normal)/da(normal)/tb(nightmare)

ub(h)/sed/reg v ub/sed/reg

Observed results:

Saw some pretty annoying behavior from my ai's.  We'd be doing well holding all of the flags so they grouped up at the middle, moved towards the enemies towers... drew some damage, then backed down.  They need to stay in lanes to level, not rush the middle, get dmged and have to run away.

This behavior happened multiple times (as of ws 3 - 3 times).  Seems to occur if my team owns all of the flags.

Another thing - it would be good if the ai got a little better at trying to get kills.  It still prioritizes flags over kills, even if a kill is often very close to happening. 

Late game - my team is ws10 - their team is ws6.  We walk in and take their portal with no resistance.  Is it possible to make the ai super protective of their portals IF the OTHER team has ws 8 and above?

Center push on Cataract might be a result of the valor flag change, unless that change only affects 'owned' valor flags.

AI should definitely prioritize its own portals over enemy portals, unless it's much closer to the enemy portal, and even then, it should still try to retake its own portal immediately afterwards.  Is the portal flag code treating 'owned' portals differently from 'enemy' portals, other than preferring portals closer to the citadel?

What exactly is wrong with Sedna's heals?  Focusing too much on demigods not currently in combat?  If so, could add a small radius check (<15) for enemy demigods and have it prefer any friendlies with <75% health and an enemy nearby over those with lower health, but keep in mind that will occasionally have bad results as well (enemy just out of range casting an ability on the wounded DG, etc).

Is there any rule in place where the at at WAR SCORE 6 would not prioritize saving its portal when the opponents are at WS10?  In MP, if that was the case, its an all hands on deck sort of thing to SAVE the portal.  If NO ONE ports, then its GG.  The AI doesn't seem to consider porting as an option - at least in the game I'm referencing, where I KNOW 2 dgs had enough hp and mana to port. 

Peppe - with regards to sedna - can we do anything to make her more supportive, though?  For instance, even if its idiotic, can we have her heal any dg nearby with a lower % of hp than her?  When we see QoT play now, we know she will generally shield the crap out of herself (which means she doesn't die as much).  If we can have senda do the same, but perhaps focus more on teammates unless she is more at risk, then good.

And back to the QoT thing - her shielding herself is super, but an ideal support player (and one that is ranged at that) spams shields on their teammate and attacks from a distance.

Another advance QoT strat (though, granted, its only ok) is to start the game, immediately get shield, and shield up 1 player while still at the crystal and shopping.  If you can add that into the code (pause 5 sec if Qot, then shield random dg at start of game), that would be good too. 

Anyway, if you care to watch a replay, check out one from the tourney (I can link for u if needed) where it was my team vs mcshanes team (ub/sed/occ v ub/sed/oc).  If you note, sedna TYPICALLY stays a little back and provides support heals.  That's the ideal sedna unless she's just holding a flag.  I wonder if it would be useful to run some replays for you and give you some feedback on the fly as to why expert MP guys do this or that. 

Anyway, I'm still loving the mod. I spend more time cursing because of the reg AI now as I NEVER know if he's going to snipe me or not, so I'm SLOWLY starting to respect him.  Bastard seems to hit me more often than not. 

26u:

http://peppe-dg-mods.googlecode.com/files/Skirmish_AI_v26u.zip

http://peppe-dg-mods.googlecode.com/files/Skirmish_AI_v26u.exe

QoT: Spam shield when she is around healthstone (fire at will).  Her build already started with shield, so no change to her build.

Sed: Increased her threshold to self heal when an ally is near.  I think i did 65% or 70% for self heal if an ally is near.  If an ally is near she will heal them at 85% (might be a little spammy).    Played with weight of heal self and heal ally (may cause heal ally action to be selected more)

Flipped flag vs attack grunt in the overrides.  So if grunts are around it should attack them instead of pursue a new flag until warrank 8.

Tweaked portal priority when down a portal.  Unless the AI is 3/4 of the way to the enemy portal already it should go for the portals closest to its citadel.

I didn't see much attacking of the center towers that seemed out of the ordinary.  The valor flag wasn't given anything special in the capture routine, so it shouldn't be going after the flag in there.    Let me know if you see it more and maybe i can watch the replay and turn on some logging to find which action is taking it in there.

yes the teleport in to defend portals and valor flags threatened in a larger radius was disabled until warrank 8.  Added a check to look at both teams to see if its time to defend those flags more aggressively.

QoT firing shield around health stone is easy... positioning and awareness type micro is hard.  I've considered making the AI more fearful when its out numbered, but it already retreats so much I thought it might get too boring.

I like the changes you detailed and am looking forward to giving this a go after work.

The precise scenario is this - my team owns all 3 middle flags on cata.  The my ai ends up at the middle flag (either because they just capped it, or whatever).  The ai then walks up and engages creeps/towers at mid, pushing past the towers.  The effect of this is that the ai actually takes all/some of the creeps out of the mana/hp lanes (eg those creeps never make it there).  I'm not sure what the coding change would be to do this.  I think this ONLY happens if all of the mid flags are owned and the AI is at middle already (i think it could be related to the valor flag weighting, though). 

AIs are easily kiteable,   even if a player is at 100hp  if someone else joins the fight and does some dmg the ai will instantly turn around and not do the last auto attack to kill the player.

As a queen player I would prefer groundspikes over shield even if there random (random spike on tower at level 10 = gg tower ).

Just don't expect the shield and u will live.

Haven't read the whole thread and just started using("testing") this and uberfix mod.  For all I know, Rook and Beast AI is standard and any differences I'm seeing are psychological.

Rook vs Beast - beast cowers from the towers.  I kept beast fenced in most of the game on Crucible.  I think he stunned me once the entire game.

Beast vs Rook - Rook doesn't drop towers enough, strays from them too much, etc.  Never set up a tower field although he seemed to take towers as his primary (he never hammered)

That said - the AI seems much better than the standard set especially when against multiple AI and I'm not spending all my time trying to mess with them.