Alpha Centauri 2

Sid Meier's Alpha Centauri & Alien Crossfire => Modding => Bug/Patch Discussion => Topic started by: Question on October 02, 2014, 01:03:14 PM

Title: Is anyone taking over Kyrub's AI patch?
Post by: Question on October 02, 2014, 01:03:14 PM

Apparently hes no longer working on the game, and the parts incorporated into Yitzi's patch at least are REALLY buggy....

Title: Re: Is anyone taking over Kyrub's AI patch?
Post by: Yitzi on October 02, 2014, 06:04:32 PM

If you can talk him into sending me files with everything he's got on the topic of AI (especially the locations of various data), I may be able to fix some specific bugs; more would be a future possibility but would have to wait.

Title: Re: Is anyone taking over Kyrub's AI patch?
Post by: kyrub on November 03, 2015, 09:52:34 AM

Based on the SMAC related activity on RPGcodex forums, I decided to port the AI patch to SMAX myself.

I use the 3.4 version of Yitzi's patch as the basis. I'll include a diff file with all changes and a slightly more detailed description of changes.
I hope the work will be done soon.

Title: Re: Is anyone taking over Kyrub's AI patch?
Post by: Magitek on November 03, 2015, 12:21:29 PM

Quote from: kyrub on November 03, 2015, 09:52:34 AM

Based on the SMAC related activity on RPGcodex forums, I decided to port the AI patch to SMAX myself.

I use the 3.4 version of Yitzi's patch as the basis. I'll include a diff file with all changes and a slightly more detailed description of changes.
I hope the work will be done soon.

Looking forward to it kyrub! I was actually going to ask for your databases and attempt it myself but I would probably take forever.  I was mostly interested in the changes for morgan and university expansion rate because they're really weak factions in SMAX, tracking down the individual change in your patch was too much of a nightmare though.

Let me know if you need help with testing specific features.

Title: Re: Is anyone taking over Kyrub's AI patch?
Post by: kyrub on November 04, 2015, 09:11:03 PM

Thanks, Magitek.
Indeed, I need some help. Testing will be necessary later.

Right now, I could use some important information and assessment from experienced SMACX players.
1) How does the AI Pirate faction work with the new facilities with bonus resources, like Aquafarm, Trunkline, Traducer? Does it build them enough?

2) Are those facilities as desirable as Recycling tanks? Is there a general consensus, as to, in which part of the game one should build them? How much of a priority are they? What other conditions (base size, war status, energy or nutrient situation etc.) should one ponder, when thinking about building them? What should be avoided as a situation?

3) I am trying to implement (very very raw) triple AI strategical approach, as per the classic manual: Build, Momentum, mixed, based on the already present "AI_aggressivity" code in alphax.txt. Can you agree with one of these variants: "The AI in original SMACX is known to build generally such amount of buildings that it is probably closest to a (Build / Mixed / Momentum) game of a human player."

Thanks for any help or opinion.

Title: Re: Is anyone taking over Kyrub's AI patch?
Post by: Magitek on November 05, 2015, 01:49:48 AM

You might want to open a new topic for more visibility.

As for your questions, I'll run a couple of games with the pirates involved, although I'm sure the experienced here will chime in at some point about the pirates AI.

Title: Re: Is anyone taking over Kyrub's AI patch?
Post by: Magitek on November 05, 2015, 04:02:08 AM

I ran a game to 2450:
I have the feeling the pirates should be played as momentum or even a builder to some extent. They can expand unchecked for a considerable amount of time and should make use of it.

Some things I learned about the pirate AI:

They build too many defensive land units, their cities in early game often have five or six land based defenders that cannot move out of the base, which is somewhat damaging to their production speeds. I'm fairly sure players would be building defensive boats instead of armies. These armies were eventually used to raid but I think they did more damage to the pirates than anything else.

They actually built an aquafarm when surrounded by kelp, although building it too early is actually a detriment due to the 2 food resource limit. In 2200 I've yet to see a second aquafarm. I think the aquafarm is a really good facility for them to use, so long as they've planted kelp. The AI seems to understand the kelp requirement for the aquafarm, but obviously it's formers are too few to plant the kelp required to trigger the aquafarm.

2200:
In my game they did not produce sea formers until 2200 which puts them at a disadvantage (probable research requirements keeping them down). They keep cancelling the formers in favor of defenses (I assume this is triggering due to vendetta). The defenses at these bases already number 7+ (typical AI behavior).
They built a land-based colony pod at a sea base that could never leave.
Expansion seems fairly strong. If it chose better base locations it would be many times stronger however.

2265:
Expansion seems to be slowing down in favor of military. Although the pirates are doing OK in ratings, it's all downhill from here due to their poor economy.

2300:
1/13 bases have aquafarms. 5 bases would greatly benefit from it being built. The rest don't seem to have formers to plant the kelp in the first place. I can't imagine how much stronger the pirates would be with aquafarms. It has six fungicidal formers now, which is probably ok (these were built too late though).
The 2 colony pods are still stuck at bases but many of the infantry were finally moved out for raiding.

2350:
Actually seems to use marine detachment a fair bit for capturing ships which is good.
Not even a second aquafarm yet. Many opponent AI bases are reaching 9+ population levels, while the pirates best is at seven (the aquafarm base naturally, the only one to reach 7 so far).
Data roze has a 14 pop city, properly utilizing tree farms (I'm sure this is just a coincidence). She even has a bore hole, though I've no idea where it came from.
Nearly every faction has five or more facilities per base, while the pirates have barely 3 on average.
Most of the land bases have more aquafarms than the pirates which is baffling.
18 bases total. The leader has 23 (Aliens of course), although many are captured bases. Average is 19 bases so the pirates are probably falling behind drastically due to lower populations.
The colony pods finally manage to make landfall! .. and now is expanding on land, something that probably should have happened in 2100.

2400:
The pirates are getting pulverized by air, the majority of bases don't have AAA or ECM defenders (just stacks of plasma sentinels), even though the pirates have the tech. AAA units seem a critical necessity for the pirates in particular, but they don't care to use it.
The pirate faction is pretty much irrelevant now. If they don't build aquafarms, they should at least be occupying more territory. Preferably both.
Having drone problems at multiple bases.

He never gained access to trunkline technology, but would have been significantly more powerful given the number of mines built on the sea. No one but the pirates would be able to build these due to the mineral costs. He ignored traducers.

Early game, his research really needs to be focused on build once he has Explore 1 for aquafarms and formers, rather than explore and conquer. Build 3 and 4 provide him with incredible sea base potential, although he refused to build traducer when he got it, even though he would benefit greatly from it as well.

Overall, I guess he didn't do bad. At 2400 he was fighting the top 3 factions and not losing much ground, but it would have been entirely different if had those three facilities for every base. I would pin his late game 'success' more down to the AI doing a poor job of actually attacking sea bases.

Perhaps we can get some input from someone who actually plays as the pirates, but it seems to me as if the Pirates AI should really focus on sea expansion + those key facilities with some skirmishing here and there. The pirates really are unique in the fact that sea bases actually have some merit and taking advantage of it could result in a late-game monster.

Title: Re: Is anyone taking over Kyrub's AI patch?
Post by: Magitek on November 05, 2015, 04:34:20 AM

2nd game:

Played out similar to the first.
The pirates never built an aquafarm or traducer despite having access to both.
Surprisingly in 2400, the pirate faction took the lead position by having a superior air force.

I noticed Marr builds recycling tank on his sea bases, despite the fact they do nothing when paired with pressure domes. Even with that malus, he basically out-boated the Pirates.
The free drones had an extensive array of sea bases, all with zero facilities.

Title: Re: Is anyone taking over Kyrub's AI patch?
Post by: kyrub on November 05, 2015, 07:42:07 AM

Thanks, that is very thorough information, quite valuable! Those below stand out for me and could hopefully find their way into the patch.

Quote from: Magitek on November 05, 2015, 04:02:08 AM

They build too many defensive land units

They actually built an aquafarm when surrounded by kelp

In my game they did not produce sea formers until 2200 which puts them at a disadvantage (probable research requirements keeping them down).

If it chose better base locations it would be many times stronger however.

AAA units seem a critical necessity for the pirates in particular

The colony pods finally manage to make landfall! .. and now is expanding on land, something that probably should have happened in 2100.

Note that with the (basic SMAC) AI patch, all AIs already build more formers and expand more on sea in general. The seaformers plant a lot more kelp as well. I guess we will have to wait and see how this changes some of the issues you have mentioned. Maybe, it's best to make a simple port of the AI SMAC patch first and reconsider the issues afterwards. One thing is sure, I will strengthen all recyling-tanks-based facilities for sea bases.

As for landfall, that is one of the hardest points. The AI is programmed to expand from land to sea, at one point. It does it simply by producing a seapod. But the other way is far from simple, it needs planning and combining units etc. I wonder, could a marine colony pod be designed for bases close to land? Would something like this work, would AI use it?

Title: Re: Is anyone taking over Kyrub's AI patch?
Post by: ete on May 22, 2016, 07:15:14 PM

Quote from: kyrub on November 04, 2015, 09:11:03 PM

3) I am trying to implement (very very raw) triple AI strategical approach, as per the classic manual: Build, Momentum, mixed, based on the already present "AI_aggressivity" code in alphax.txt. Can you agree with one of these variants: "The AI in original SMACX is known to build generally such amount of buildings that it is probably closest to a (Build / Mixed / Momentum) game of a human player."

Thanks for any help or opinion.

Unsure about original, but SMAX AI generally builds too few buildings, maybe as much as a hard momentum player would *during a rush*, but rather than building rovers and probes it spends all its supply on defensive units and transports. Even an aggressive player needs to consolidate their position and build infrastructure, and they don't seem to do that. I think this is critical; the AI needs to play as a builder some of the time even with an aggressive personality or they'll fall behind on tech and infrastructure and lose to mid-lategame exponential growth every time, even if they got a good start and stole a bunch of bases.

Early bases want to be supporting 1-2 formers, one defensive/police unit, and maybe one scout. If a base is spending more than 2-4 minerals a turn on support outside an urgent conflict, something's going wrong. If there's a nearby active war, pumping rovers is fine for a while, but otherwise buildings and terraforming should be prioritized.

Title: Re: Is anyone taking over Kyrub's AI patch?
Post by: Yitzi on May 22, 2016, 07:54:51 PM

Quote from: ete on May 22, 2016, 07:15:14 PM

Quote from: kyrub on November 04, 2015, 09:11:03 PM

3) I am trying to implement (very very raw) triple AI strategical approach, as per the classic manual: Build, Momentum, mixed, based on the already present "AI_aggressivity" code in alphax.txt. Can you agree with one of these variants: "The AI in original SMACX is known to build generally such amount of buildings that it is probably closest to a (Build / Mixed / Momentum) game of a human player."

Thanks for any help or opinion.

Unsure about original, but SMAX AI generally builds too few buildings, maybe as much as a hard momentum player would *during a rush*, but rather than building rovers and probes it spends all its supply on defensive units and transports. Even an aggressive player needs to consolidate their position and build infrastructure, and they don't seem to do that. I think this is critical; the AI needs to play as a builder some of the time even with an aggressive personality or they'll fall behind on tech and infrastructure and lose to mid-lategame exponential growth every time, even if they got a good start and stole a bunch of bases.

Early bases want to be supporting 1-2 formers, one defensive/police unit, and maybe one scout. If a base is spending more than 2-4 minerals a turn on support outside an urgent conflict, something's going wrong. If there's a nearby active war, pumping rovers is fine for a while, but otherwise buildings and terraforming should be prioritized.

Colony pods and teching too (depending on faction and situation)?

Title: Re: Is anyone taking over Kyrub's AI patch?
Post by: ete on May 22, 2016, 09:23:47 PM

Getting enough CPs to found a decent empire is also vital, yes. As is teching, though that's not a build choice (other than indirectly via NNs and the like).

Title: Re: Is anyone taking over Kyrub's AI patch?
Post by: Yitzi on May 22, 2016, 11:06:53 PM

Quote from: ete on May 22, 2016, 09:23:47 PM

Getting enough CPs to found a decent empire is also vital, yes. As is teching, though that's not a build choice (other than indirectly via NNs and the like).

And more indirectly by the choice whether to focus on energy or minerals, and whether to focus more on tech or economy.