SUGGESTION: AI troop building algorithm

[Arralen]

Originally posted by MStavros:
Ah, so anyone came up with any good ideas so far in this topic?

You'll find the idea this thread is all about at the very beginning.

How good it is is debatable, though. And testable, so feel free to do a test game and report your findings.

[ May 09, 2004, 21:36: Message edited by: Arralen ]

[HotNifeThruButr]

How about something to discourage the building of units that don't go well together? Like if the comp has 60% of his already built + will be built forces (measured by gold + resource value, not unit numbers, so one knight will be "more" than 2 militia) as units without missile weapons (exclude javelins and other such minor missiles)and with less than, say, 13 prot, it will feel less compelled to make Archers, Comp bowmen, slingers (excluding Mictlan warriors with slings) and Longbowmen. Maybe have it roll a 3 sided dice, and if it lands on 1, proceed with archer building, if not, abort. Having 60% or more of your army being units with 13 or greater prot makes the three dice roll apply when the comp wants to make crossbowmen and other artillery types.

A way to improve location might be to might be by giving forts that are closest to an enemy province (not indep) that borders on your province a higher priority when building. This will be measured in "provinces away," so a fort that's two provinces away from an enemy will have a higher priority than one that's three provinces away. The capital might be exempt from this rule since there are cap exclusive units.

Maybe this can just work with unit Groups.

Edit: There might also be an impact on troop building depending on the neighbor and his force's composition. If the AI "sees" enemy forces with many hoplites, it will bump down light infantry, and missile units production and train more heavy infantry, heavy cavalry, and artillery/crossbow type units. If it sees tramplers with morale lower than 11 and no berserking, it will build more missile units. But always bump up production in units that are as big as or bigger than the enemy's tramplers.

Is that realistic?

Edit: This also makes it possible to reform the difficulty system, where instead of cheating on pretender points (that's the current AI difficulty setting, no?) we can enable more AI features at harder levels and disabling AI features at easier ones so it actually plays smarter/dumber. For example, an impossible AI might know not to mix Knights with Crossbows and orders non-combat commanders to flee instantly, or give assassins optimal assassination equipment whilte an easy AI wouldn't care about unit mixes in combat and wouldn't equip assassins with good equipment.

Edit: thought of something else. You could also factor in spell to troops synergy (I believe that's the right word for worky-togetherness). If the AI utilizes an army of light infants, it could be less inclined to cast flying shards, while friendly high prot units and hostile low prot units encourage it. It will not summon Air Elems when facing enemies composed of units bigger than the elems themselves. If it's got abysian or other heat resistant units, it will bump up desirability for heat spells.

[ May 10, 2004, 03:18: Message edited by: HotNifeThruButr ]

[wombatsSAR]

From what I recall of the GalCiv AI (I was a beta on the OS/2 implementation - not that that means much of my understanding and programming skills), I believe that the basic module could be moved from game to game - Stardock's first Windows game was ... The Corporate Machine (? - seems odd, but then I recall some problems with the originally proposed name for the game) - a biz empire game - and the same core AI from GalCiv was moved to it. Anyway, I'm wondering aloud if it might be possible to license and adapt that AI. ... Could be too messy and too expensive, but then again, Brad Wardell (GalCiv head programmer) is a nice guy (tm) and it might be possible/not-too-expensive.

That said, the conVersion work involved could be gross - much as I liked the game, the feature set was spartan compared to that of Dominions II and it's possible that there are just too many options for the AI to cope. However, the AI was quite good at what it did. Brad had said that it was possible to put in a nearly "perfect" game player (economically) but that it made a very boring opponent. They had to add some stuff to spice up its antics. And on top of that were the "skins" - a couple of which were actually adapted from the play styles of some of the players. For those that have played the game, "Martzian" (sp?-been too long) AI was from one of the betas.

[Gandalf Parker]

GalCiv is automatically the one to mention in any conversation about AI. But even Brad said it was years to get it converted to Windows. Id hate to think what it would take for windows, linux, mac-X, solaris. Some of Brads talking-papers on the subject might help.

Plugin AIs are great but any plugin has to have something to plug into with the right connections. It could be swappable modules, or importing player game saves (Brads method), or player scriptable (my personal favorite).

But if I want to say

code:

---

IF ArmyA<100 and ArmyA(LI)>50 then buy HI

  

---

then there needs to be labels for storing army and troop info. It needs to have been written or re-written to have lots and lots of grabbable labels.

But if its linear. Going thru each nation one at a time and just spending their resources.. then I can suggest that "Buy LI, LI, LI, LI" might be better as "Buy LI, LI, HI, Calvary".

Of course I should test it first. Anyone should be able to play a game as an AI. Only buying or building by the formula with no extra.. well OK, I couldnt do that. Not for more than a dozen turns or so. But SOMEONE might be able to.

[ May 10, 2004, 19:08: Message edited by: Gandalf Parker ]

[HotNifeThruButr]

Galactic Civilizations? I bought that game and thought it was a waste of 30 bucks. Didn't give it much of a chance though.

[proteus]

It is a good game with a good AI.
But I still recommend HoMM3. Awesome AI. Not as many units/spells like in Dom2, but still lot of different units/spells.
Also in Doms2, the majority of the units are 'useless', just like the spells.
Quantity <<< Quality.

[Gandalf Parker]

Originally posted by proteus:
It is a good game with a good AI.
But I still recommend HoMM3. Awesome AI. Not as many units/spells like in Dom2, but still lot of different units/spells.
Also in Doms2, the majority of the units are 'useless', just like the spells.
Quantity <<< Quality.

Wow you really should play more. In 5 or 6 more years you will learn to say things like 'useless to me' instead of useless. I have found very little in the game that is useless but it has take about a year of purposely looking at things that seemed that way in order to find how they might be used. A player has to find the nation/strategy that fits their playing style (and then of course the rest would be useless to them)

[Gandalf Parker]

Originally posted by HotNifeThruButr:
Galactic Civilizations? I bought that game and thought it was a waste of 30 bucks. Didn't give it much of a chance though.

For nearly a decade GalCiv has been mentioned in scientific papers and serious discussions about Artificial Intelligence. Its the game built around AI algorythms. The original OS/2 Version mostly but now that Windows can do multi-threaded tasking (kindof) Brad has tried to port it to Windows. Its still a work in progress. I own a copy but I must admit Im kindof waiting for the game itself to get more interesting to make the AI worthwhile

[proteus]

That is true...you were right, this was my example. pI am never using LI for example, but maybe it is useful as well, just not for the AI.]

[Arralen]

Originally posted by Kristoffer O:
Hmm, it's a bit difficult to follow without graphical representation.

ok, here it is:


[ May 28, 2004, 15:26: Message edited by: Arralen ]