What I learned from 376 turns of Auction-X
Moderator: Forum Moderators
-
- Posts: 855
- Joined: October 3rd, 2004, 4:52 am
- Location: Portland, OR
- Contact:
What I learned from 376 turns of Auction-X
Recently, in a fit of boredom, I have been watching the AI duke it out on Auction-X. I give each side +5 Gold a turn, but otherwise keep the standard settings.
(5 extra gold is not much, it means that each side can get an extra unit every three turns)
The only constant is that the middle AI always loses, and usually does so fairly quickly, about 20 rounds in. After that, all four AI players will fight over the middle. One of them will usually, through Time of Day, luck, or advancing a unit, take control of the middle. They then send their advanced units off into the other arms of the map, where they fall in a few turns, and then all four sides rush in to fill the vacuum. Some factions have more survivability than others, Knalgans, for instance, and some have much less--- Drakes, as could be predicted, fly into enemy territory and are cut down.
One of the most obvious lessons is that the AI is actually pretty good at tactics, but very bad at strategy. It manages to surround and kill enemy units appropriately...but of course, it has no way to plan for the future, so it doesn't take time to consolidate its units.
The most surprising thing I have noticed in several of these games (and I have watched several, they usually last over 100 turns, 376 was exceptionally long even for this type of thing) is the AI's weird unit choice. I don't know if it is just this setup, but the AI never chooses some units. Some of these are obviously things the AI shouldn't be touching, like Berserkers and Fencers, but across hundreds of turns in all of these games, the AI never recruited a SINGLE Elvish Fighter, Drake Fighter or Spearman. Instead it recruits specialty units.
So, all in all, I learned that the AI is interesting to watch, even in its stupidy. I also learned that I have too much time on my hands.
Also: does anyone want to try a set-up like this and see if they can get the AI to fight for over 500 turns? Over 1000?
(5 extra gold is not much, it means that each side can get an extra unit every three turns)
The only constant is that the middle AI always loses, and usually does so fairly quickly, about 20 rounds in. After that, all four AI players will fight over the middle. One of them will usually, through Time of Day, luck, or advancing a unit, take control of the middle. They then send their advanced units off into the other arms of the map, where they fall in a few turns, and then all four sides rush in to fill the vacuum. Some factions have more survivability than others, Knalgans, for instance, and some have much less--- Drakes, as could be predicted, fly into enemy territory and are cut down.
One of the most obvious lessons is that the AI is actually pretty good at tactics, but very bad at strategy. It manages to surround and kill enemy units appropriately...but of course, it has no way to plan for the future, so it doesn't take time to consolidate its units.
The most surprising thing I have noticed in several of these games (and I have watched several, they usually last over 100 turns, 376 was exceptionally long even for this type of thing) is the AI's weird unit choice. I don't know if it is just this setup, but the AI never chooses some units. Some of these are obviously things the AI shouldn't be touching, like Berserkers and Fencers, but across hundreds of turns in all of these games, the AI never recruited a SINGLE Elvish Fighter, Drake Fighter or Spearman. Instead it recruits specialty units.
So, all in all, I learned that the AI is interesting to watch, even in its stupidy. I also learned that I have too much time on my hands.
Also: does anyone want to try a set-up like this and see if they can get the AI to fight for over 500 turns? Over 1000?
Don't go to Glowing Fish for advice, he will say both yes and no.
- NuclearWorm
- Posts: 14
- Joined: November 8th, 2010, 3:35 am
-
- Posts: 855
- Joined: October 3rd, 2004, 4:52 am
- Location: Portland, OR
- Contact:
Re: What I learned from 376 turns of Auction-X
I saw that phrase on tvtropes earlier, now I have to go to knowyourmeme to find out what it means.NuclearWorm wrote:OVER 9000!
Don't go to Glowing Fish for advice, he will say both yes and no.
- Crushmaster
- Posts: 383
- Joined: August 9th, 2008, 3:38 pm
- Location: United States
- Contact:
Re: What I learned from 376 turns of Auction-X
I've noticed that too. When I've played against Loyalist AI, they never recruit a single spearman - just calvarymen, HI, bowmen, and maybe mermen. It's weird.Glowing Fish wrote: Some of these are obviously things the AI shouldn't be touching, like Berserkers and Fencers, but across hundreds of turns in all of these games, the AI never recruited a SINGLE Elvish Fighter, Drake Fighter or Spearman. Instead it recruits specialty units.
Definitely.Glowing Fish wrote: So, all in all, I learned that the AI is interesting to watch, even in its stupidy. I also learned that I have too much time on my hands.
That would be a bit...tragic, I think.Glowing Fish wrote: Also: does anyone want to try a set-up like this and see if they can get the AI to fight for over 500 turns? Over 1000?
In Christ,
Crushmaster.
-
- Posts: 855
- Joined: October 3rd, 2004, 4:52 am
- Location: Portland, OR
- Contact:
Re: What I learned from 376 turns of Auction-X
And this is even the case when there were pierce-vulnerable opponents, such as Drakes or horse units to fight.Crushmaster wrote:I've noticed that too. When I've played against Loyalist AI, they never recruit a single spearman - just calvarymen, HI, bowmen, and maybe mermen. It's weird.Glowing Fish wrote: Some of these are obviously things the AI shouldn't be touching, like Berserkers and Fencers, but across hundreds of turns in all of these games, the AI never recruited a SINGLE Elvish Fighter, Drake Fighter or Spearman. Instead it recruits specialty units.
Don't go to Glowing Fish for advice, he will say both yes and no.
Re: What I learned from 376 turns of Auction-X
Yeah, the AI seems to have bizarre aversions to certain units, and favour others (all-gryphon initial recruit, anyone?). It's wacky.
Re: What I learned from 376 turns of Auction-X
It only recruits the best of the best.
But yeah, I noticed this already a year ago, especially with the Loyalist AI, that it only recruits the same unit over and over, even if that unit isn't very good against the enemy faction.
I personally solved the problem by modifying the AI to just recruit random units. And from my investigations, players have it harder to defeat that AI than the one with fixed recruitment!
But yeah, I noticed this already a year ago, especially with the Loyalist AI, that it only recruits the same unit over and over, even if that unit isn't very good against the enemy faction.
I personally solved the problem by modifying the AI to just recruit random units. And from my investigations, players have it harder to defeat that AI than the one with fixed recruitment!
Wesnoth
The developer says "no".
The developer says "no".
-
- Posts: 855
- Joined: October 3rd, 2004, 4:52 am
- Location: Portland, OR
- Contact:
Re: What I learned from 376 turns of Auction-X
It is hiring gryphons because the AI's first hire is for scout units to grab villages. Of course, they have a good other scout, the footpad, which is much cheaper, but they don't hire it...Atz wrote:Yeah, the AI seems to have bizarre aversions to certain units, and favour others (all-gryphon initial recruit, anyone?). It's wacky.
Don't go to Glowing Fish for advice, he will say both yes and no.
Re: What I learned from 376 turns of Auction-X
Which would make sense if it didn't recruit four gryphons at once, thus blowing its entire budget and not being able to recruit anything else.Glowing Fish wrote:It is hiring gryphons because the AI's first hire is for scout units to grab villages. Of course, they have a good other scout, the footpad, which is much cheaper, but they don't hire it...Atz wrote:Yeah, the AI seems to have bizarre aversions to certain units, and favour others (all-gryphon initial recruit, anyone?). It's wacky.
-
- Posts: 855
- Joined: October 3rd, 2004, 4:52 am
- Location: Portland, OR
- Contact:
Re: What I learned from 376 turns of Auction-X
Hmmm, that is true that it does seem hire more scouts on the first turn, and its scouts are also very expensive, so then it doesn't have money for anything else.Atz wrote:Which would make sense if it didn't recruit four gryphons at once, thus blowing its entire budget and not being able to recruit anything else.Glowing Fish wrote: It is hiring gryphons because the AI's first hire is for scout units to grab villages. Of course, they have a good other scout, the footpad, which is much cheaper, but they don't hire it...
Don't go to Glowing Fish for advice, he will say both yes and no.
Re: What I learned from 376 turns of Auction-X
Gryphon Rider: neutral potential damage of 24. Tops everything else. That's why it's AIs first choice.
Greetz
HomerJ
Greetz
HomerJ
Six years without a signature!
Re: What I learned from 376 turns of Auction-X
The AI just isn't considering cost and versatility enough.
Wesnoth
The developer says "no".
The developer says "no".
Re: What I learned from 376 turns of Auction-X
What you never play gryphon swarm?
I play it from time to time, particularly on maps with two areas divided by water.
I play it from time to time, particularly on maps with two areas divided by water.
CHKDSK has repaired bad sectors in CHKDSK.EXE
Re: What I learned from 376 turns of Auction-X
Cool. Could you please show how to change the AI to just recruit random units?Rya wrote: I personally solved the problem by modifying the AI to just recruit random units. And from my investigations, players have it harder to defeat that AI than the one with fixed recruitment!
Re: What I learned from 376 turns of Auction-X
Inside the side tag:
Code: Select all
[ai]
recruitment_pattern=
recruitment_ignore_bad_combat=yes
recruitment_ignore_bad_movement=yes
[/ai]
Wesnoth
The developer says "no".
The developer says "no".