How to persuade AI to do some basic things in a very different ruleset?

[cazfi]

If you's like to try, this is the version of the ruleset for 2.6. The problem existed there, too.

Unrelated to the problem at hand, but with "./fcser --warnings"

Code: Select all

2: /fast/freeciv/longturn/games/LT63Sim/data/LT63Sim/terrain.ruleset: unused entry: extra_resort.gui_type
2: /fast/freeciv/longturn/games/LT63Sim/data/LT63Sim/terrain.ruleset: unused entry: extra_resort.border_sq
2: /fast/freeciv/longturn/games/LT63Sim/data/LT63Sim/terrain.ruleset: unused entry: extra_resort.vision_main_sq

[cazfi]

It seems it considers Tribe all right, but the value for new cities seem to be zero or even negative. So presumably the problem is not in the unit, but maybe some property of the Tribal government makes AI not to build cities.

[cazfi]

Can't spend more time on this now, but the logs from the autogames I run show that level hard AI does select Tribes as the build target in S2_6 HEAD, at least sometimes (hard to tell from the logs alone if that's rare in the context of the situation in the game)

[Ignatus]

It seems it considers Tribe all right, but the value for new cities seem to be zero or even negative. So presumably the problem is not in the unit, but maybe some property of the Tribal government makes AI not to build cities.

Probably having no trade at Tribal has something to do with it, but new cities evaluator should value production as well. Though likely the production alone is not enough to overvalue building other units.

[cazfi]

When I find time, I'll ruleup that S2_6 ruleset to S3_0 (and then maybe also to S3_1) and do some test runs to see how it behaves in newer branch(es). This would also benefit from the better AI behavior inspection tools in newer branches.

[Corbeau]

When I find time, I'll ruleup that S2_6 ruleset to S3_0 (and then maybe also to S3_1) and do some test runs to see how it behaves in newer branch(es). This would also benefit from the better AI behavior inspection tools in newer branches.

If you are referring to SIM ruleset, there is also a 3.0 version on Github, somewhere among Longturn games.

[cazfi]

When I find time, I'll ruleup that S2_6 ruleset to S3_0 (and then maybe also to S3_1) and do some test runs to see how it behaves in newer branch(es). This would also benefit from the better AI behavior inspection tools in newer branches.

If you are referring to SIM ruleset, there is also a 3.0 version on Github, somewhere among Longturn games.

To compare freeciv behavior in different branches, I wanted to have the exactly same rules, so ruleup was the way to go. I've now ruleup'd that S2_6 ruleset to all later branches ( S2_6 -> S3_0 -> S3_1 -> S3_2 ), and am now adding those rulesets to my autogame testing matrix. They will be in server-only autogame runs for now. One they have proven stable there, I'll try to take time to do also such runs where I observe the progress.

[cazfi]

AI is also stuck to Tribal. This goes to typical problem of freeciv AI not thinking more than one step forward. There's no benefit to build the Palace (1 step) under Tribal, but it would need to both build a Palace and change the government to Despotism (2 steps).

[cazfi]

AI is also stuck to Tribal. This goes to typical problem of freeciv AI not thinking more than one step forward. There's no benefit to build the Palace (1 step) under Tribal, but it would need to both build a Palace and change the government to Despotism (2 steps).

- This is very much kind of problem I wanted to resolve with tex AI. Unfortunately during the ~10 years of tex development, I have been able to spend just handful of hours on it
- Depending on how much you want to stick with the current rules for human players, you could maybe make also part of the benefits of Tribal -> Palace -> Despotism to come already with the Palace. E.g. Reduce the Tribal 100% penalty to Gold and Science to 50% when the player has a Palace.

[Corbeau]

At the moment Palace gives a small food bonus. But yeah, I'm considering this. Just didn't get to it yet.