AI Contest & Exchange

[Autolykos]

Introduction

From what I've seen until now, a lot of mods suffer from poor/inadequate AI (mine included), mainly because of the difficulties involved in editing and tuning AI files. The documentation is almost nonexistant, and while there are a few AI wizards out there, they seem to work on their own and only test their ideas against their own (or vanilla) AI. To advance AI editing from an arcane art to something even vaguely resembling a science, we need an open exchange of ideas and a (mostly) unbiased way to test them against each other. For this reason, I propose:

ACE - AI Contest & Exchange

I will host a contest between multiple AIs on a slightly modified Abyss scenario (LINK). The scenario is modified to give all factions roughly equal starting positions:
- No faction is crippled by being extremely short on resources (like UTC in vanilla)
- All capitals have 100% Infra
- All provinces have at least 80% Infra

General Rules

1. Each contestant may submit one (or more) sets of AI and Event files.
2. The files may be submitted openly or via PM, but will be published after the contest (the whole purpose is learning from the submissions)
3. I will post updates on the situation after every year and save/reload at that point. This should automatically load the AI file for that year if one is provided.
4. I will run the game as handsoff with Tannu Tuva. If an AI DoWs me by accident, I will use acceptall to make peace again. AIs DoWing me on purpose (via event or combat = {}) are disqualified.
5. The country with the highest amount of base IC after 15 years (1.1.1951) is considered the winner.
6. The game is played on normal/normal and democracies can declare war (no TT or IC takeover).
7. Each contestant may pick/reserve one (and only one) country on a first-come-first-served basis.
8. Using other people's (or vanilla's) AI as basis is fine, but all sources have to be declared as a comment in the AI file.
9. Do not submit your entry as a self-extracting archive (.exe). Only .zip, .rar and .7z are accepted.
10. Make sure your files are actually valid, load properly and do not crash the game. I may correct small mistakes (like wrong charset), but I won't debug them.
11. Countries for which no AI is submitted will use their vanilla AI.

Rules on AI Files

1. All AI files have to be named "TAG_*.ai", with TAG being the tag of your country. They may be placed in subfolders.
2. Blatantly gamey tactics (Zerg Fleets, completely ignoring whole branches/pages of the tech tree, etc) are allowed and encouraged.
3. Outright cheating (like negative values in enemy_handicap = {}) is not.
4. No backstabbing of (potential) alliance members. You may not attack countries with the same starting ideology, even if they're run by another contestant. Not joining their alliance or trading with them is technically legal (I can't possibly control/enforce it), but I'd prefer you not doing it.

Rules on Event Files

1. All Events must be contained in the same file "TAG_events.txt", and may use only the ID range I assign to the respective country (see below).
2. The Events are allowed to load AI files (your own or vanilla's) and do everything a human player could do (declare war, move sliders, change ministers, ...)
3. Triggers may only contain conditions a human player could see (or guess using spies/experience). If you're unsure, ask.
4. Minister/Idea changes are free at the beginning (1.1.1936) and have to pay the full dissent cost afterwards.
5. The ai_prepare_war command is banned, since it can unpredictably crash the game.
6. Events may only target your own country.

Rules on Multiple Submissions

1. You may enter the contest with multiple AIs, under the following conditions:
2. Your AIs are not allowed to cooperate. Especially, they must have different ideologies and may not share a border or a common neighbor.
3. Other contestant's first pick of a county has priority over the second and third of someone else, even if those were made earlier.

Reserved Event ID Ranges

Spoiler

800001-801000: UES
801001-802000: UBO
802001-803000: UPR
803001-804000: URO
804001-805000: UER
805001-806000: UGS
806001-807000: UCS
807001-808000: UPS
808001-809000: UCH
809001-810000: UTC
810001-811000: UTL
811001-812000: UTO
812001-813000: UAU
813001-814000: USN
814001-815000: UIC
815001-816000: UPE
816001-817000: UIR
817001-818000: UAP

Participants

 

[Mr_B0narpte]

Great idea! I guess the ultimate aim is to improve the AI's abilities in combat.

 

[Autolykos]

Improving the AI's combat skills is definitely one of the aims, but there's only so much you can do with AI files (still, I'm prepared to learn from hacks using clever combinations of events and AI file switches to guide combat tactics). Finding out what units, brigades and doctrines suit the AI best and how to get the most out of them is probably just as important. Balancing IC investment and proper timing in declaring wars are also areas with much room for improvement.
But the most important thing is testing what works and what doesn't. I have some preconceptions about each of these questions, but I might be completely wrong and would never notice it unless I test my AIs against someone else's.

 

[Pang Bingxun]

This thread might help to achieve some valuable knowledge.

 

[Mr_B0narpte]

Finding out what units, brigades and doctrines suit the AI best and how to get the most out of them is probably just as important.

Wouldn't the best (land) brigades be Artillery and Sp-Artillery for the AI? Less to research and they are generally the best brigades. But I guess what I am saying is an assumption (but based on practice and the combat values of the brigades) and thus goes against the 'spirit' of this thread.

Balancing IC investment and proper timing in declaring wars are also areas with much room for improvement.

Aren't all the countries already at war in this scenario? Sorry I cannot check as I cannot install your edited version because I have so many different versions of AoD at the moment.

But the most important thing is testing what works and what doesn't. I have some preconceptions about each of these questions, but I might be completely wrong and would never notice it unless I test my AIs against someone else's.

That's a good mode of thinking.

 

[Autolykos]

Wouldn't the best (land) brigades be Artillery and Sp-Artillery for the AI? Less to research and they are generally the best brigades.

Very probably - but even with that, I'm not sure enough to take it for a fact. Art and SpArt are usually considered the "ultimate" brigade because of flexibility - but the AI will probably not use the Artillery Bombardment mission a lot, making RctArt more cost effective (although late in the tech tree). Also, Art slows your units down and SpArt uses fuel (and out-of-fuel-penalty sucks hard), so I can see unbrigaded (or unorthodox options like AA) as a viable alternative, too. On top of that, there are advocates of Armored Cars, Assault Guns and Tanks, and the question of when to prefer Art and when SpArt. That still leaves a lot to try out.

Aren't all the countries already at war in this scenario?

Nope. There are two scenarios (The Abyss and Armageddon), and in The Abyss, they start at peace and are not part of an alliance (I chose that one on purpose).

 

[Pang Bingxun]

SpArt is a very fine brigade, but becomes available relativly late. Art is available earlier and more tc efficient. Eng Brigade has a much lower tc load and works well on offensive and defence, therefore its is a good brigade for AI. If manpower is no issue vanilla Inf without brigade gives a very good BWA/TC.