AI Expansion

**tupodawg999** · January 07, 2010, 08:33 PM

AI Expansion

Spoiler Alert, click show to read:

Introduction

This tutorial will consist of a series of posts exploring techniques for improving the AI's campaign map performance with the least amount of modding effort. It's mainly aimed at people who've already modded a bit but maybe haven't looked at AI behaviour but i'll also detail the modding steps involved in some of the changes i make as a general modding tutorial. I'll use the vanilla campaign as the test case.

The general pattern of the posts will be:
1. Description
2. Modding Details
3. Effect

The effect will be based on running automated tests after modding and seeing if the AI takes more or less regions than they did compared with the vanilla files.

The list of areas covered will be:

Intro
Setting up

Part 1

Universal changes that show up in short tests i.e 12 turns
--Movement points
--AI movement trait
--Brigands, Pirates and Revolts (part 1)
--Brigands, Pirates and Revolts (part 2)
--Pre-building
--Rebel garrisons

Changes relating to specific factions to help their individual expansion or edits to particular regions the AI has a problem with. These will be:

--Scotland
--Russia
--Denmark
--Portugal
--Caffa region

Part 2

--New Baseline
--England and Venice
--Mystery Improvement
--Modding for the AI (Rebels)
--Modding for the AI (Missions, Money Shocks and Upkeep)
--Unit Stats
--Campaign AI

+++

Next post: Setting up

**tupodawg999** · January 07, 2010, 09:28 PM

Post 2: Setting Up

The best way to get a feel for how the AI behaves is to start the game (or mod) as a faction that is out of the way and doesn't get bothered in the early game e.g egypt, then pick a faction to watch and click end turn while watching what that faction does. You'll soon see the problems (it helps to have a spy nearby to look into woods) but figuring out the solutions to those problems is mostly guesswork and trial and error.

However, many of the AI's problems relate to random events such as army revolts which do not occur consistently for each faction at the same point in each game. To test to see if edits related to that sort of thing has a beneficial effect on the AI's expansion then i think you need to run sets of tests, average out the results and then compare them to a vanilla baseline.

To do this we need to set up a mod in a way where we can automate the AI for a specified number of turns.

Modding Details

The first step is to get our testing mod set up and ready.

Step 1

1) Create a "mods" folder inside your main MTW2 folder.
2) Create an "ExpansionTest" folder under the mods folder.
3) Create a "data" folder under the ExpansionTest folder.
4) Copy the "world" folder from your main MTW2 "data" folder and copy it into expansiontest/data.

Step 2

1) Make a copy of your medieval2.preferences.cfg file in your main MTW2 folder and rename it to "expansiontest.cfg".
2) At the top of this new file add

Code:

 
[features]
mod = mods/ExpansionTest

Lower down find the section that should look something like:

Code:

 
[hotseat]
autoresolve_battles = 0
disable_console = 0
disable_papal_elections = 0
save_prefs = 0
update_ai_camera = 0
validate_diplomacy = 1

and replace it with

Code:

 
[multiplayer]
playable = true
[hotseat]
autoresolve_battles = 0
disable_console = 0
disable_papal_elections = 0
save_prefs = 0
update_ai_camera = 0
validate_diplomacy = 1

Step 3

1) Create a new text file in the same main MTW2 folder, name it "ExpansionTest.bat" and write in

medieval2.exe --io.file_first @ExpansionTest.cfg

Step 4

1) Run the .bat file

If it's all working right what should happen is the game will load and if you select single player / grand campaign / england / start then you should see the hotseat screen for your first turn as england. You don't need to bother about the password just click start. You'll be in a standard vanilla campaign in hotseat mode running from a mod folder.

Quit game.

Step 5

1) Go to the mods\expansiontest\data\world\campaign\imperial_campaign folder and open the campaign_script.txt file.
2) Near the top of the file, just below where it says:

Code:

 
;----------------------
; monitors

add

Code:

 
monitor_event FactionTurnStart TrueCondition
console_command control england
terminate_monitor
end_monitor
 
monitor_event FactionTurnStart FactionType slave
and I_TurnNumber = 10
console_command control england
terminate_monitor
end_monitor

What these scripts do is as soon as the game starts control of the england faction is handed over to the to the ai. After the number of turns determined by the value you put in the script (in this case 10) control is returned to the player.

Effect

1) Use your expansiontest.bat file to start the game again, select single player / grand campaign / england / start and when the hotseat screen comes up click through that. What should happen if everything is working right is the ai should immediately take control of the england faction as well as all the others and carry on that way for n turns at which point control of england is handled back to you.

2) When the ai takes control hit the space bar once so the game uses fast movement.

3) Open up the console - on my keyboard it's the @ key that brings up the console but i believe it varies - then type "toggle_fow" so you can see the whole map.

nb The number of turns to use is tricky as you need to use different numbers to test different types of edit. The testing process is a lot faster if the number of turns is low so it's a good idea to test changes that show an effect within a few turns first with a low number of turns, then lengthen the test runs later if neccessary.

4) Count something.

What to count will depend on what you're trying to test. If you're testing general AI expansion and you've set the number of turns to be low then it's quicker to count the number of regions the AI factions have taken. If you've set the number of turns higher then it will eventually become quicker to test something else like the number of rebel regions remaining.

5) Repeat the test a number of times and average the results to try and iron out random events.

Baseline

For my initial tests i'm going to set the number of automated turns low and count the number of regions taken by the AI factions.

For the initial group of tests i set the number of automated turns to 10. As the game starts at turn 0 this returns control after 11 turns (i meant it to be 12 but counted wrong).

Six tests using the vanilla files gave me a baseline total of 102 regions gained by the AI factions.

nb The numbers per test were 18, 17, 20, 18, 12, 17. Generally in my testing the numbers tend to mostly cluster but with dramatic outliers. I think this is because in an average game maybe 2-3 factions have a major glitch. In some games maybe only one does and the expansion is particularly good. In other games maybe 4-5 factions have a major glitch and there's a big drop.

Given the level of randomness in the game i don't think six tests is really enough but it takes too long to do more. However i think if a change produces an improvement of 10%+ then it's reasonable to assume there's some benefit.

Lastly instead of adding each change on top of each other i'll make a change, test it individually, then remove it again before the next change so each test will be compared purely with the vanilla game. This is because the same problem can have multiple work-arounds. One fix might help with 10 AI problems and another might help with 30 but some of those problems will be the same. It's easier to see the general scale of improvement if it's just compared to vanilla.

+++

Next post: Movement points

**tupodawg999** · January 08, 2010, 01:46 AM

Post 3: Movement Points.

The AI in MTW (and RTW) has major path-finding problems. One of the main reasons for slow AI expansion is the AI armies are constantly getting stuck trying to reach their targets.

I believe the main problem is simply that the path-finding calcs take too long when there's a lot of units to move in the AI turn so the game uses a strong path-finding method to find a route to a target up to one turn's worth of movement points away and if it hasn't found the target within that range then the ai switchs to a weaker but faster method of path-finding. Therefore increasing the army movement points increases the range the game uses the strong path-finding method for each AI each turn.

The downside is this can have a dramatic effect on the length of the AI turn to the point where the game becomes unplayable (at least to me) so there's a compromise to be made. Strict agent limits can help with this as the AI factions have a habit of clogging up the game spamming spies and other agents.

The default value for army movement is 80 movement points. What value to increase it to will depend on map size, scale, region density etc and will have to be found by trial and error and tested late game when the map is swarming with AIs. From previous testing with the vanilla map i think 120 is a good value so i'll go with that.

Modding Details

Step 1

Copy the vanilla descr_character.txt file from the vanilla mtw2 data folder and copy it to our mods/expansiontest/data folder.

Step 2

Open decr_character.txt and edit the values for movement points to be 120 for armies and 180 for admirals and agents:

Code:

starting_action_points 120 ; default value for all characters and pathfinding calculations

Code:

 
type named character
actions moving_normal, moving_quickmarch, garrison, assault, attack, besiege, entrench, ambush, diplomacy, bribe, exchange, building_fort, building_watchtower
wage_base 100
starting_action_points 120

Code:

 
type general
actions moving_normal, moving_quickmarch, garrison, assault, attack, besiege, entrench, ambush, exchange
wage_base 0
starting_action_points 120

and for the agents and admirals

Code:

 
type spy
actions moving_normal, spying
wage_base 100
starting_action_points 180

Code:

 
type admiral
actions moving_normal, quick_sail, blockade, disembark, exchange
wage_base 50
starting_action_points 180

repeat for all the others.

Effect

Run the ai-controlled tests and count the number of regions captured by the AI.

In my case the total was 139 which is a nice improvement over 102.

(20, 23, 27, 23, 23, 23)

Increasing the base movement points will give the biggest improvement for the least modding effort.

+++

Next post: AIMovementBonus trait.

**konny** · January 08, 2010, 06:16 AM

A lot of interesting stuff here. I would suggest to move this thread to the scripting section.

Originally Posted by tupodawg999

Another possibility might be to have just one level of the trait that gave a 100% bonus right away but with an anti-trait and triggers that removed the bonus. The anti-trait triggers might be:
1) Post-battle - to hopefully reduce retreating too far
2) Ended turn in settlement.
3) Ended turn outside a settlement but with < 50 movement.

I'm not that familiar with anti-traits so i did it this way with four incremental bonus movement levels to dilute any unintended consequences a bit.

An antitrait reduces points of his opposing trait first before accumulating points itself. That is, when you have GoodCommander and BadCommander as antitraits and a character gains one level in GoodCommander for winning a battle, the next battle lost will not give him a level in BadCommander (what would result in him being a good and a bad commander at the same time) but take away the level acquired before in GoodCommander. Further battles lost would then give him points in BadCommander.

For your purpose I would indeed suggest to use one level in AIMoveBonus and a trigger with a negative value for reset:

Code:

 
;------------------------------------------
Trait AIMoveBonus
Characters family
;Hidden
 
Level AIMoveBonus1
Description AIMoveBonus1_desc
EffectsDescription AIMoveBonus1_effects_desc
Threshold 1 
Effect MovementPoints 10
;===============================================================

Trigger Reset_AIMoveBonus_trigger
WhenToTest CharacterTurnEnd
 
Condition Trait AIMoveBonus = 1

Affects AIMoveBonus -1 Chance 100
;--------------------------------------------
Trigger AIMoveBonus_trigger
WhenToTest CharacterTurnEnd
 
Condition IsGeneral
and not EndedInSettlement
and RemainingMPPercentage > 95
and not FactionIsLocal
 
Affects AIMoveBonus 1 Chance 100

That way the bonus trait will be reseted for each character who has it and instantly afterwards given anew for anyone qualifing, allowing him to use the bonus movement the next turn.

**tupodawg999** · January 08, 2010, 01:13 PM

@Konny

A lot of interesting stuff here. I would suggest to move this thread to the scripting section.

Ty. I posted here because some of the changes will be map edits so the changes will be both text file and map.

For your purpose I would indeed suggest to use one level in AIMoveBonus and a trigger with a negative value for reset:

Yes on reflection i think you're right. I'll redo it.

**tupodawg999** · January 08, 2010, 05:53 PM

Post 4: AIMovementBonus Trait

(Hat-tip to Konny for suggesting an improvement to this which i've incorporated.)

Most players will have seen the basis of the next change many times without realising the significance - an AI stack will join a crusade/ jihad and suddenly attack a nearby city that has nothing to do with the crusade / jihad target. A lot of players will have seen France do this with Bruges early in a vanilla game but it will happen all over the map at some point or other during a game.

What happens is one of the most common pathing glitches involves an AI stack moving to attack a city gets stuck at a point more than one but less than two turns movement away from the city. However anything that brings the target city within one turn's movement of the bugged army fixs the problem and the stack moves again. Many things can do this like a road being built in the region but one of the most common in MTW2 is the stack joins a crusade and gets its movement points doubled, and it seems, because the army was stuck when it joined the crusade, the original target city takes precedence over the crusade target city.

This will only happen with a general army and not a captain army obviously.

The AI much prefers to have its generals as governors - unless it has spare generals it will usually move all of them into cities and use mostly captain armies. However one advantage the AI seems to have is it often gets captains promoted to generals on the move and the captains that get promoted seem to be almost always those armies that have been told to capture a city. What this means is the critical AI stacks - the ones that get stuck on the way to attack settlements are the ones most likely to get general promotions.

This kind of promotion is not rare but it's not common either - it's somewhere in between. I've seen an AI captain getting promoted en route to attack a city many times while testing but the frequency overall doesn't seem to be more than one per faction at a time and my guess would be when the faction already has a governor in each of its cities. There's also either a random element to it or it's very specific i.e when the faction has exactly x generals in x cities and captain armies moving to attack cities. If there's already excess generals it may not happen at all.

Regardless of the details there will be many times during an average game where an AI captain army will get promoted to general while stuck more than one but less than two turns of movement away from its target.

This provides the opportunity for a trait fix to help the AI expansion.

Modding Details

Step 1

1) Add a "text" folder in "mods/expansiontest/data".
2a) Either copy the export_VnVs.txt.strings.bin file from the main mtw2 "data/text" folder into mods/expansiontest/data/text and use the strings_bin_converter from http://www.twcenter.net/forums/downl...o=file&id=1006 to convert it to export_VnVs.txt. Delete the .bin file afterwards.
2b) Or download the text files http://www.twcenter.net/forums/downl...do=file&id=951 , then copy the export_VnVs.txt file and paste it into "mods/expansiontest/data/text".
3) Add at the top:

Code:

 
¬--------------------
{AIMoveBonus1} AIMoveBonus1
{AIMoveBonus1_desc}
AIMoveBonus.
{AIMoveBonus1_effects_desc}
+100% to Movement Points

Step 2

1) Copy export_descr_traits.txt from the main mtw2/data folder into mods/expansiontest/data.
2) Under

Code:

 
;===============================================================
;== TRAIT DATA STARTS HERE ==
;===============================================================

add

Code:

 
;------------------------------------------
Trait AIMoveBonus
Characters family
;Hidden
 
Level AIMoveBonus1
Description AIMoveBonus1_desc
EffectsDescription AIMoveBonus1_effects_desc
Threshold 1 
Effect MovementPoints 10

3) Lower down the same file underneath

Code:

 
;===============================================================
;== TRIGGER DATA STARTS HERE ==
;===============================================================

add

Code:

;------------------------------------------
Trigger Reset_AIMoveBonus_trigger1
WhenToTest CharacterTurnEnd
 
Condition Trait AIMoveBonus = 1
Affects AIMoveBonus -1 Chance 100
 
;------------------------------------------
Trigger Reset_AIMoveBonus_trigger2
WhenToTest PostBattle
 
Condition Trait AIMoveBonus = 1
Affects AIMoveBonus -1 Chance 100
 
;------------------------------------------
Trigger AIMoveBonus_trigger
WhenToTest CharacterTurnEnd
 
Condition IsGeneral
and not EndedInSettlement
and RemainingMPPercentage > 95
;and not FactionIsLocal
 
Affects AIMoveBonus 1 Chance 100

nb1 The live version of the trait would be Hidden and AI only but those parts are commented out for testing.

nb2 There seems to be an odd glitch with the "RemainingMPPercentage" condition. In the original version of this trait when i had "RemainingMPPercentage = 100" in the trigger it wouldn't fire with base movement points of 120 plus a previous 20% movement bonus - some kind of strange rounding error somewhere maybe - however setting it to "> 95" works. This may be useful info for anyone getting odd results with that condition.

nb3 The reason for the postbattle trigger is one of the unintended consequences of the trait i already knew about was the common situation where an AI army goes move->siege->siege->assault. In this case the AI would get the movement bonus at the end of the turn of sieging and then if the assault on the next turn fails the stack would sometimes retreat the full distance given by the move bonus. This could then lead to situations where the AI gets stuck at the spot they retreated to but the move fix won't help them because they already have it. Removing the movement trait postbattle hopefully prevents that.

nb4 Getting stuck after retreating is the second biggest AI movement problem.

Effect

I removed the previous changes to bring the mod folder back to the vanilla state, made this change and then ran the six tests.

My result was 114 regions gained (compared to the baseline of 102).

Actual values 18, 22, 16, 17, 17, 24

+++

Next post: Brigands, Pirates and Revolts (part 1)

**El Hamburglo** · April 03, 2010, 08:20 PM

Any chance you could upload a version of your map with the extra settlements?

**Gorrrrrn** · April 05, 2010, 04:49 PM

re Mission Shocks
http://www.twcenter.net/forums/showt...51#post6709051

As far as I'm aware the AI factions don't get missions as such.

--

re settlement levels I'd recommend starting with a minimum level of towns and no villages - the latter are a pain in public order terms.
ie their population isn't big enough to allow for public order buildings, and being captured usually causes a population drop.
the AI doesn't keep a big enough garrison to maintain order, and sooner or later, the place rebels, the population drops and the settlement remains a village.
and so on and so on ...

**irishron** · April 05, 2010, 05:07 PM

I naven't gotten into the mechanics, but in my mod the ai will grow villages. In fact, I sieged one that was a village by the time I got there was a large town so it can be done.

We are using BBB 4 and XAI if that helps.

**Methoz** · April 06, 2010, 07:50 AM

can downlaod any file for new mod?

**VINC.XXIII** · July 25, 2010, 03:45 PM

Very interesant thread, I would know how modd the power of the loyalists, who sometimes push the conqueror out from the city.
Thanks

**z3n** · December 30, 2019, 01:47 PM

Moved this thread here.

Pretty interesting read.

**bitterhowl** · January 02, 2020, 11:48 PM

I already read this thread about 2-3 years, but couldn't realize it's wisdom for that time.

Well, that's a very interesting and illustrative example of AI improvement whith such a small changes. And also an argument of how disbalanced original game is.

**bitterhowl** · March 30, 2020, 08:57 PM

One more old hint (I suppose scince Rome I) - in descr_strat if you set different faction_creator for settlement then faction it owns at start of the campaign seems that faction set as faction_creator will try to capture this settlement. Need more tests but first drafts says it's true.

So we can set principal targets for factions this way.