SMACX Thinker Mod

[dino]

> However, the question is whether you want it to be that strong and stomp you every time. Wouldn't you want to eventually win?

I don't think you have to worry about making AI good enough to beat you on difficulty level without cheats. If you've managed to make an AI requiring drop to thinker difficulty from transcend, it would make for a much more satysfying experience, while transcend would remain as ultimate autistic challenge.

Unfortunately I'm too busy to play recently and only follow changelogs of thinker and WtP, I guess I should stop trying to provide feedback untill I have a chance to try your latest efforts.

[Alpha Centauri Bear]

Don't worry about it! Plenty of people express their opinion without having an experience. That is completely natural.
😀

Besides, I am asking for your past experience and thoughts. Nothing specific to a certain version.

[Induktio]

Here is the change in nutshell. At the beginning of the faction turn AI computes "threat level" from other faction combat units combined. The computation is quite complex. It accounts for diplomatic status, unit strength, unit proximity to faction bases, ability to reach it (same or different continent), current defensive power, some cross unit type effectiveness, blah-blah. Many parameters. Then it uses this value to bump military production (combat units, probes, military facilities) priority over non-military.

It seems like you probably rewrote some ideas from Thinker's algorithm with different coefficients. And some parts are probably different. But it looks like there might be a lot of other issues in WTP. I'm not particularly motivated to spend time on analyzing WTP's version unless there's some evidence the solution works better.

There also isn't anything stopping from doing A/B testing with multiple versions of some code. Just have a couple of factions play against themselves and see if there's any notable differences.

Related to that, at one point I considered adding a new feature to the map generator which would produce maps with mirrored layouts, e.g. mirror along X/Y axis or both. Landmarks would cause some issues there because some of them would be messed up so it might require disabling them. Maybe that kind of feature should be added to the mod.

The diplomacy and war choices is a serious AI work, though.

With this in mind, it is theoretically possible to make AI play like a chess pro. However, the question is whether you want it to be that strong and stomp you every time. Wouldn't you want to eventually win?

The branching factor in chess is quite large but amenable for brute force search. In Go it's even larger and brute force approach starts to fail. These kind of strategy games have a way larger branching factor because of the large decision space. Also do you want to treat it as a perfect information game? Technically it isn't so, but sometimes the AIs are given help there. I would use some other comparison than "chess pro" because of these technicalities.

[EmpathCrawler]

I'm greatly impressed by the new world generator. I haven't even had time to sit and play yet, I've just been having fun quick starting over and over to see the results. I like what you've done with the equatorial jungle, too. Should be interesting.

There is a bug, at least while playing with SMAC-in-SMACX enabled. The Fossil Field Ridge fails to spawn correctly (and I don't think it is supposed to with the mod, anyway). There will be a landmark note on the map where it might have been, but no ridge.

[EmpathCrawler]

I played a bit today and I've actually had a lot of trouble getting the game to start properly. I made a desktop shortcut for the launcher and added the -smac flag. Sometimes the launcher does nothing (I get a cursor with an hourglass, then nothing happens), sometimes it'll launch in SMAX mode, and sometimes it'll correctly launch in SMAC mode.

[Induktio]

The map generator is my attempt at procedural content generation, it uses some fancy, newer algorithms.

> The Fossil Field Ridge fails to spawn correctly (and I don't think it is supposed to with the mod, anyway).

What do you mean? It spawns correctly on my tests. What map size are you using and which landmarks are enabled?

> I played a bit today and I've actually had a lot of trouble getting the game to start properly.

This issue would have to be reproducible before anything can be done about it. Need more information etc.

[EmpathCrawler]

Fossil Field Ridge: This is a fresh install with a default Thinker extracted on top. Like I said, I only have SMAC-in-SMACX enabled by the -smac flag which should stop it from spawning at all because it's expansion content. I started the game and clicked Quick Start over and over to get maps without making any changes beyond the flag. There will be a landmark marker on every map I've made, but there's no fossil terrain graphic and the underlying tiles don't have the "Fossil" or whatever attribute, it's just plain ocean. It's as though it's getting incompletely removed.

It spawns correctly in Crossfire games. When I disable it in thinker.ini, it correctly stops spawning. When I disable your map generator, it stops spawning. The Unity Wreckage terrain and marker (correctly) haven't appeared on any of the SMAC-in-SMACX maps, it's just the Fossil Field Ridge and just the marker.


Launcher woes: I'll look for a pattern, but it seems to be random. It could be a Windows 11 quirk. I'll try it on Windows 10 if I get the chance. I changed the install directory from the GOG installer's default, if that matters.

[Induktio]

> Fossil Field Ridge

Yeah that is an oversight, it should not be there because it's also disabled by the default world builder. Instead of Fossil Field Ridge the current version disabled Manifold Nexus and Unity Wreckage from smac-in-smacx.

I thought Manifold Nexus would be expansion content because the description references progenitor stuff, but now that I checked, vanilla SMAC binary seems to also spawn Manifold Nexus on the map for some reason. It's the latest official patch provided by GOG, so maybe it was added to vanilla in the patches? I'm not sure now. Seems a little odd. Hard to say if it should remain in smac-in-smacx mod.

[EmpathCrawler]

> Fossil Field Ridge

Yeah that is an oversight, it should not be there because it's also disabled by the default world builder. Instead of Fossil Field Ridge the current version disabled Manifold Nexus and Unity Wreckage from smac-in-smacx.

I thought Manifold Nexus would be expansion content because the description references progenitor stuff, but now that I checked, vanilla SMAC binary seems to also spawn Manifold Nexus on the map for some reason. It's the latest official patch provided by GOG, so maybe it was added to vanilla in the patches? I'm not sure now. Seems a little odd. Hard to say if it should remain in smac-in-smacx mod.

That's right, the Manifold Nexus and also the Borehole Cluster were both added by patch to SMAC to foreshadow the upcoming Alien Crossfire lore. Which is why I love your mod's ability to disable those, too. Fossil Field Ridge and the Unity Wreckage were the only ones added directly by the expansion.

[PvtHudson]

Seems, changes in diplomatic status do not cancel probe teams (including probe foils) orders. This leads to odd attacks: for example, mind controlling a base next turn after cease-fire, or tech steal immediately after signing alliance.

[Alpha Centauri Bear]

AI does not save their plans over turn. Unit order recalculated every turn.

[Induktio]

> This leads to odd attacks: for example, mind controlling a base next turn after cease-fire, or tech steal immediately after signing alliance.

Hmm yeah, that might be another oversight. It is probably quite easy to reproduce it but if you have some save game that exhibits this problem, you can also send it to me.

> AI does not save their plans over turn. Unit order recalculated every turn.

That is not accurate with Thinker. Probe units will usually continue towards their assigned move target without checking if the treaty status changed in the meantime, which is the cause of this problem. And there are also some other plans that are preserved in the save game.

[JoGr223]

> topic: AI can't fight even when they are "on paper" strongest faction and initiate war
..
Dry maps are generally worse for AI performance too. If you want more challenge, try also using blind research, otherwise with the default tech tree it might be too easy to rush some techs.
..
Generally the AIs might not have the time to build up during the first 80 turns if the game settings are unfavourable for them. You can try changing conquer_priority=150 or something so the AI will build more units. Sane values for that setting are somewhere in the range 50..250.
..
The AI already has some priorities which cause it to increase combat unit production if it loses bases, but it's probably not going to help much on smallest map sizes.

ok, so I took this info and based on that designed challenge with better developed AI  result of which I describe here:
2268: total war against 3x stronger AI with cloning vats
https://alphacentauri2.info/index.php?topic=21766.0

I comment on Thinker AI there too, it's too big to keep everything in this single thread here

[dino]

With the original ruleset it's impossible to make AI capable of defending against the player, just do the math:

The odds of attacking an armored defender in a base with perimeter defense are 1:1. To destroy 10 armored defenders, you need 15 attackers: you wil loose 5, 5 near death and 5 almost unscatched.
As an attacker you'll suffer half the loses of a defender if you have 50% advantage in a number of units brought to a base assault, ridiculous !!!
If you bring few more multi moves units to the battle you can also after destroying armored units wipe out with no loses as many low armor units as you have moves.

As a human player you can theoretically wipe out in the first few turns of war 3x the size of your army and take over 30-50% of bases.
In mid and end game if the player / ai would play properly, the war is simply won by the side that starts it, it's all there is to it, unless attacker is much smaller.

The *active* defense is even more ridiculous, between combat odds and collateral  damage, cheap rovers and copters can wipe 2-4x the amount of approaching units with no loses.
So the player can easily play builder game and and fend off few times stronger AI opponents with minimal army, then easily wipe them out in few turns mid game.

***
The combat rules in SMAC are fundamentally broken, no amount of AI improvements will change it, so there is no point in both playing and developping combat ai without addressing it:

1) Edit armor values in alphax.txt to change att/def ratio from 2:1, to around 3:2.
Now assault odds for a base with perimeter are 3:4, doesn't look like much change from 1:1, but with the way combat and health points work, to take out 10 armored defenders you'll need 20 units.
15 will be destroyed, 5 will have 50% health, so now as an attacker you'd need 100% numbers advantage and suffer 50% more losses then the defender, to conquer a base.

You won't be able to win a war against an opponent of comparable size in few turns and as long as AI bring enough units through it's disorganised stream of units to conquer even a single base and park some undamaged defenders in it. It'll be almost as effective as mass assault, since you'll have to loose as much resources to take it back as AI lost, at this point it's much more crucial who can replace units faster, then who attacked first.
This also changes the odds of defending in a rough terrain, or a base without perimeter to 1:1, much more sane than vanilla were the odds are in a favour of the attacker.

2) No instant healing with command center, only 20% per turn.
Further limits how many bases you can take out in a single turn, since if all your assault force will be damaged, it will be destroyed in a counter attack.
It both limits the advantage of initial mass assault and forces you into 2-3 turns break to repair your forces, giving AI time to counter attack.

3) Disable collateral damage from loosing a unit in a stack.
This change, combined with att/def ratio of 3:2, makes *active* defense with fast unarmored attackers no more effective than passive defence with armored units behind perimeter.
Which makes easy no loses defence against ai during early game impossible.

I'd make this configuration default in thinker, with vanilla rules either ai will be always defenseless, or player will be if you teach it to attack first with mass assault properly, playing both options is, or would be unfun and pointless. These changes, makes ability to make an initial mass assault advantegous, but the ability to produce units faster more crucial and between ai bonuses and thinker improvements AI is good at producing stuff. It should also make smaller maps more viable for AIs, since it shouldn't be possible to wipe anyone with 2-3 impact rovers even early.

[EmpathCrawler]

Launcher woes update: So I've had intermittent failures to launch with thinker.exe on two separate PCs, one on Windows 11 and another on Windows 10. What I've done on both was set up a desktop shortcut, added the -smac flag to the shortcut path, and used the shortcut. It will often not start anything, and when it does start the game it will not always respect the -smac flag. Am I the only one experiencing this?

What's really weird is if I run thinker.exe from a PowerShell prompt, the launcher almost always works, but it still doesn't consistently respect the -smac flag.