Latest Posts

Topic: AI seafaring issue?

Tibor

Topic Opener
Joined: 2009-03-23, 23:24
Posts: 1377
Ranking
One Elder of Players
Location: Slovakia
Posted at: 2017-12-23, 23:14

Hi, while testing some improvements to AI, I noticed that AI colonization does not work. A colonizing ship is just waiting next to a probable port field - forever. And no colonies at all, in the time when there should be many...

This would be big issue and I wonder if I am the only one who encountered this.

EDIT: using branch derived from current trunk (8543)

Edited: 2017-12-23, 23:16

Top Quote
hessenfarmer
Avatar
Joined: 2014-12-11, 23:16
Posts: 2646
Ranking
One Elder of Players
Location: Bavaria
Posted at: 2017-12-26, 00:11

In former revisions when I tried to train AI I saw the AI building colonies. So I propose first to give them more tries. However the ship does a lot of weird things before building a port. I hope that this could be influenced by training.


Top Quote
Tibor

Topic Opener
Joined: 2009-03-23, 23:24
Posts: 1377
Ranking
One Elder of Players
Location: Slovakia
Posted at: 2018-01-01, 21:22

I am testing it again, AI builds ships but does not start any expedition. Am I really only person who sees this issue?


Top Quote
kaputtnik
Avatar
Joined: 2013-02-18, 20:48
Posts: 2433
OS: Archlinux
Version: current master
Ranking
One Elder of Players
Location: Germany
Posted at: 2018-01-02, 11:00

Just tested map "The Nile" with multiplayer as spectator an r8548. The AI starts expeditions and build colonies round about 4 hours game time.

  • Player1 (blue Barbarian): Build port and colonies
  • Player2 (red Empire): Was long in a deadlock, no ports, no shipyard, so no colonies
  • Player3 (yellow Atlantean): Deadlock, no trees
  • Player4 (green Barbarian): Build ports and colonies
  • Player5 (black Empire): 1 Shipyard, no port so no colonies
  • Player6 (ocre Atlantean): Build ports and colonies
  • Player7 (pink Atlantean): Deadlock, don't know why
  • Player8 (white Atlantean): 1 Shipyard, no port so no colonies
Edited: 2018-01-02, 11:02

Fight simulator for Widelands:
https://wide-fighter.netlify.app/

Top Quote
Ex-Member
Avatar
Joined: 2014-09-12, 10:53
Posts: 184
Ranking
Widelands-Forum-Junkie
Posted at: 2018-01-02, 11:56

It seems from some testing that I have been doing that some AI DNA is good at colonies and some are not. This may be what accounts for the differences noted by kaputtnik. I had several AI players, some get deadlocked, two had built 20 ships each but no port, one of these had all the ships in a small lake so they could never sail anywhere. One built a port but no ships and another had got stuck with the port 90% finished, I think he as short of gold as he was only mining iron and stone.

It seems training the AI for one task can often reduce their abilities in other areas. There seems to be one DNA that stops expansion after creating basic buildings, it dismantles almost all military buildings and the soldiers all retreat to the headquarters.

It is not very clear how the AI works, there does not seem to be much in the wiki about DNA and what the numbers mean or even about what AIhints are, for example there is a limit setting for weak and very weak but is there a limit setting for normal AI?


Top Quote
Tibor

Topic Opener
Joined: 2009-03-23, 23:24
Posts: 1377
Ranking
One Elder of Players
Location: Slovakia
Posted at: 2018-01-02, 12:55

Well, this is not about deadlock. This is about situation, when a player has a port, has a shipyard, has few ships built and building new ones and still not launching an expedition. This is what I saw. Moreover player should not keep building new ships when previous ones are idle.... This is not how I intended it. But there was one change to seafaring by I dont remeber who... Maybe somebody should look at it....

Seafaring is not controlled by genetic algorithm, so this is not matter of training. Yes, deadlocks is completely different matter, here genetic algorithm plays a big role...

It seems training the AI for one task can often reduce their abilities in other areas.

Yes, this is true, but the changes incured from one round of testing are quite minimal... moreover, testing is done with sequence of different maps, it is matter of what maps will one use and in what order...

BTW I resumed the training with use of actual maps, I am trying to use very different ones.... The latest stage of training is only with minimal mutation rate so it should not happed that there will be one (of four) fully screwed DNA...

there does not seem to be much in the wiki about DNA and what the numbers mean

some decisions are made as coumpond of ~20 DNA numbers (e.g. scoring and so on) so it is not suitable for manual tweaking and understanding what individual DNA numbers are responsible for...


Top Quote