Utility Single Player Improvement - Page 3

momfreeek · #21 July 5th, 2012, 07:48 AM

Quote:

Originally Posted by JonBrave

The external AI program will never get written. Without the actual codebase of the real Dom3, the amount of work you'd have to do just to generate consistently legal moves would be huge, before you even get to producing a sensible and better AI opponent. Or, Andy Murray could win Wimbledon this year.

Thats why it would make sense to use a program like autohotkey to interface with the running program. So the ai could give commands and the dom3 executable would generate the .2h file. Generating legal moves would not be an issue.

- read gamestate from .trn (big task.. not sure)
- program autohotkey to give commands (still a big task.. but probably better than trying to fake .2h files)
- develop the ai (open ended task).

Yeah, not likely.

Gandalf Parker · #22 July 5th, 2012, 08:51 AM

Part of the problem with the devs giving access to the files, or even more to the code, is that they would then be responsible for making sure that the knowledge is not used for evil. Otherwise they would face the wrath of half of the community for uncaring.

Gandalf Parker · #23 July 5th, 2012, 08:52 AM

The problem with programming the AI, even in AutoHotKey, is that it would be HUGE. Even if it was done in a decent programming language with access to the code it would be huge.

Part of the complaint about the AI now is that there is only one. It makes decisions with very generic labels. Purchases are simply infantry, mounted, scout, mage based on cost. Scripting pretty much the same for all. So the improved AI would need to decide differently for Caelum vs Neifelheim. And even different for EA Caelum vs LA Caelum. And even then, to avoid players posting "how to beat the AI" you would want variations for Rainbow, SC, Blessed, Thug, or push to a specific spell. If there is no separation then we would have another catchall AI but just a different one (which I guess would at least be something)

At least Im happy that the problem was tackled and mostly fixed with the SemiRand program. At least for starting AIs. But not for players turned AI.

Fantomen · #24 July 5th, 2012, 09:48 AM

My approach wouldn't be to try and actually make the AI "smarter" in a general sense, rather use the current AI as base but teach it to handle typical situations from more accurate parameters. Then try to reduce the variables of AI play to make it limit itself to those typical situations as often as possible. Something like this:

Remove the randomly made AI pretenders. Instead premake three gods for each nation, one for each difficulty level. Preset battle scripts for these gods. The AI nations would always use these gods, but these gods and their scripts would be accessible for modding via a set of new mod commands (#AIpretender, etc).

Have preset reseach targets for each AI nation, also moddable. Example: thau 4, ench 3, const 6, thau 6, evoc 2... and so on. The AI would always research in the order preset for it's nation, and the order would be determined by the player/community to find the most optimal pattern for how the AI uses mages.

Make a recruits priority list for each nation, moddable. For example prioritize mages over other commanders and national troops before indies and so on.

Have two moddable spell priority lists for each nation, one for rituals and one for battle. The battle list could also be used for human controlled nations, allowing each player to preset spell priorities for his own mages before the game starts, which is used whenever mages can't use their battle scripts.

For human controlled nations this list should be accessible on a per game basis, to allow different spell strategies for different games, which would also improve MP.

Alternately make a reasonably effective script for each mage in the game, and have the AI default to that.

If the AI cannot be taught to bless it's sacreds, make all AI sacreds always blessed.

If the AI cannot be taught to build forts, make them appear automatically following a preset pattern for time and placement.

Preset a number of good thug builds, with preset battle scripts similar to the pretenders. Make a priorities list for each nation for building them, and make the AI try and achieve this builds as close as possible. If the AI cannot be taught to plan it's forging, make items appear in their lab automatically following a preset time schedule, and multiplied by number of forts.

I don't play SP anymore, but improved AI could make it fun again perhaps, and make it a little less painful to set nations AI in MP.

Options such as these could allow players/the community to design a playstyle for each AI nation. And over time we could reach a decently challenging AI

This is all theory of course, since illwinter sharing the secrets of the code seems unlikely. But I don't really understand what Gandalf mean by "used for evil" from a community perspective. I can imagine ****nel considering a movement towards decreased control over the client evil, but the community would benefit. The theoretic risk is cheating of course, but if the code was only shared with a single trusted person prepared to put the workload in to build this kind of utility, would there really be a big risk?

Gandalf Parker · #25 July 5th, 2012, 10:09 AM

The closest thing to that would probably be the one of the cloning projects.

momfreeek · #26 July 5th, 2012, 11:09 AM

Yeah, there's case for designing a game to be played by an AI rather than the other way round. Not sure I want though. IMO GalCiv2 was as dull as solitaire.

jimbojones1971 · #27 July 5th, 2012, 08:47 PM

Quote:

Originally Posted by momfreeek

Yeah, there's case for designing a game to be played by an AI rather than the other way round. Not sure I want though. IMO GalCiv2 was as dull as solitaire.

Yeah, that is unappealing.

I do like Fantomen's idea of simply limited what the AI does to a number of well known, optimised paths. Its never going to be much good, but it can be improved.

momfreeek · #28 July 5th, 2012, 09:30 PM

Oh, I think I misread Fantomen, sry. Hard-coded research goals, army compositions and suchlike definately sounds like the best way to develop an AI for such a complex game.. not that I've ever programmed an AI.

It gave me an idea.. would it be possible to create some sort of "survival mode" on a set map where preset armies attack the player at intervals? Might be possible with modded units that cast rituals that spawn whole armies or something. Would the AI use the units to attack or just do something stupid?

Gandalf Parker · #29 July 5th, 2012, 09:36 PM

The scenario options arent in Dom3 like they are in the newer game.

But you might be able to get what you want by using the tactic for switching the server mod in mid-game to slightly change the abilities or priorities of the units that are already in the game.

jimbojones1971 · #30 July 8th, 2012, 09:16 PM

Quote:

Originally Posted by Gandalf Parker

The scenario options arent in Dom3 like they are in the newer game.

But you might be able to get what you want by using the tactic for switching the server mod in mid-game to slightly change the abilities or priorities of the units that are already in the game.

Yeah, all sorts of interesting options would open up if you had a mod like this on the server. :-)