The Wizard's Lair - AI Fixes and Thoughts

[KaiserMuffin]

Just remove them from the befriend field entirely

[Faulty]

Judging from the behaviour I have seen liberate = and liberate_nation = control whether independence is granted and whether puppets are released, but which is which?

[^_AC_^]

Judging from the behaviour I have seen liberate = and liberate_nation = control whether independence is granted and whether puppets are released, but which is which?

liberate -> independence is granted
liberate_nation -> puppets are released (meaning they're puppets no more)

[KaiserMuffin]

liberate -> independence is granted
liberate_nation -> puppets are released (meaning they're puppets no more)

liberate -> puppetnations freed
liberate_nation -> puppet status revoked

(sorry AC, I found your explanation unclear without several re-reads)

[razorbird789]

Is there anything in the AI file that dictates how peace treaties outside events are made? Can you decrease the chance of a nation asking for peace, like if i wanted to force China to fight to the death (events not included) how would i do that?

[Pershing]

In my Kaiserreich game as Italy, I've given some southern France provinces to my National France ally. I've even sold them transports to move troops there. Even if I control their military and move troops there, they just start moving them back to Algeria when I let go.

Shorter version: How do I change what provinces the AI will garrison/defend?

[Faulty]

Does setting combat = to a very high number override the game rules about declarations of war from democracies requiring belligerence?
Someone is reporting that this is happening in my mod http://img840.imageshack.us/img840/5...ancemaniac.jpg

In my Kaiserreich game as Italy, I've given some southern France provinces to my National France ally. I've even sold them transports to move troops there. Even if I control their military and move troops there, they just start moving them back to Algeria when I let go.

Shorter version: How do I change what provinces the AI will garrison/defend?

There are fields under the "garrison" section named "priority" and "ignore" that control this, I think the SOV ai has several of these set by default if you need a template

[LiamRiordan]

So I'm looking at some of the parameters at a glance and I'm currently testing a quick and dirty AI file in-game, but I'd like to either streamline or improve testing. Are there any tips to improving AI testing or good tips to actually create them?

[Arturius]

Is there anything in the AI file that dictates how peace treaties outside events are made? Can you decrease the chance of a nation asking for peace, like if i wanted to force China to fight to the death (events not included) how would i do that?

ferocity = yes #prevents ai from accepting peace deals

[Yoy21]

Is there a way to mod the AI files so that the AI of all countries are a lot more likely to launch an amphibious assault?
Example: German AI defeats the Soviets, but never bother to land on the UK. How can I make Seelowe more probable?

[Fernando Torres]

Is there a way to mod the AI files so that the AI of all countries are a lot more likely to launch an amphibious assault?
Example: German AI defeats the Soviets, but never bother to land on the UK. How can I make Seelowe more probable?

There is a whole "invasion" section.

[KaiserMuffin]

This is actually a failure of an event. Find the German Seelowe AI (there is one) and bring the dates forward a bit, then hook it into the bitter peace events, so once that fires, the Germans will start developing their navy and concentrating on the UK again.

[Barbaelectrica]

Hi guys. I think that the developers should check the AI of Spain and Portugal. In peace, and even in war, the AI always keep all the troops in the capital city (Madrid and Lisbon).

So, when there are a war, Spain and Portugal always lose terrain and chances of win a war, because the AI spend a lot of time moving all the troops from Madrid, or Lisbon, to the front.

Thanks to all and hello to everyone.

[Fernando Torres]

Hi guys. I think that the developers should check the AI of Spain and Portugal. In peace, and even in war, the AI always keep all the troops in the capital city (Madrid and Lisbon).

So, when there are a war, Spain and Portugal always lose terrain and chances of win a war, because the AI spend a lot of time moving all the troops from Madrid, or Lisbon, to the front.

Thanks to all and hello to everyone.

Not really. Axis Nationalist Spain sends troops to Gibraltar and protect beaches..

[kami888]

Hi guys, hopefully there is still some activity around here...

Basically I'm having two separate problems which I've described here.

One is that I can't get the front AI to be less reckless when attacking. No matter what I do it seems to throw everything forward and leave no troops to cover the flanks. Case in point:


What you see here is that the Nationalist Italians attacked that mountain province in the center of Italy and after a long battle, they won. Hurray! Oh wait, turns out they've left nobody to defend Rome, oh no!! So the AI realizes this and immediately tries to move some troops back to cover the holes in the front line that it had just created, and often it is successful in bringing troops there in due time, but equally often it is not. This is a problem. Here's the relevant portion of the front AI:

	support_defense_prop	= 0.6  # Proportion of divisions in a front province that may be used to support the defense of neighboring provinces. Default 0.33

	recklessness 		= 0
	distrib_vs_ai 		= op_defensive
	distrib_vs_human	= op_defensive
	# Even - Distribute forces evenly along the whole front.
	# Reactive - Try to match enemy troop concentrations.
	# Defensive - More troops in strategically important provinces (IC, Victory Points, Oil, etc)
	# Offensive - Concentrate more troops where the enemy us weak
	# Operational Defensive - The fewer avenues of escape, the less troops (to avoid encirclement.)

	enemy_reinf_days	= 6	# Count enemy units less than X days from a target province in odds calculations
	reserve_prop		= 0.1	# Maximum proportion of total divisions in a front allowed to remain behind the front line ( HQs and divisions needing to recover.)
	panic_ratio_vs_ai	= 2.5	# Front requests more troops if outnumbered by this factor. NOTE: this results in troops being drawn from garrison duties, etc, so set HIGH.
	panic_ratio_vs_human	= 2.5	# Front requests more troops if outnumbered by this factor. NOTE: this results in troops being drawn from garrison duties, etc, so set HIGH.
	base_attack_odds	= 1.2	# Will attack at X odds or better
	min_attack_odds		= 1.0	# If stalemated for a long period, may go as low as these attack odds
	max_attack_odds		= 2.0	# Don't use more force than necessary to reach these odds
	reinforce_odds		= 0.6	# Send in more troops if the attacking force is at these odds
	withdraw_odds		= 0.6	# Break off attacks and withdraw at these odds

Yes, I've deliberately set the attack odds pretty low to encourage aggressiveness, I have also raised the defense proportion to make the AI respond a little better to being attacked among other things, but neither one seems to be the problem. As in, the AI's defensive behavior seems to be fine more or less (occasionally it STILL does not send in reinforcements when it has them and seems like it should, but that's another issue and mostly it behaves as one would expect), the aggressiveness itself is reasonable as well. And I remember having this problem before I made these modifications too, it's not a recent development. The main problem is that the AI always does direct attack where a player would do support attack. What could be causing this? Does the AI even know what support attack means? I'm not sure if I've seen any evidence of it ever using it..


And here's the second problem:


What's going on here is that Britain backed by its allies which all have tons of reserves to draw on have invaded Ireland, successfully repelled the initial Irish attack and went on the offensive, conquered most of the island, but got stuck on Cork in a perpetual stalemate. The main reason why is pretty simple - they've stopped bringing reinforcements to Ireland as soon as they about matched the overall number of troops that Ireland has per province. In fact, I'm not sure but it's possible that they've actually withdrawn some of them since after the initial invasion. They have plenty of reserves on which to draw, that goes without saying, but the only way to compel them to bring any of that to Ireland is to either make Ireland have even more troops (in which case they will once again bring in just enough to roughly match the Irish numbers per each province) OR use cheats to spawn lots of rebels in Cork, which seems to have the same effect.

Here's the relevant part of the AI:

max_front_ratio = 1.3 
max_front_ratios = { 
	IRE = 15.0
	CAN = 15.0
}

Going by the numbers it would seem reasonable to expect the AI to bring much more troops to Ireland than this, but nope, it does not. In fact, I'm not even sure if that's a 1.3 ratio it has there, it seems more like a one to one ratio to me.
Part of the reason why I set the attack odds lower is to try to go around this problem by encouraging the AI to just attack Cork relentlessly even if the odds aren't very good, as that would cause the Irish to collapse very quickly since they can't reinforce themselves and their infrastructure is bombed to oblivion. However as you see, even with 1.0 minimum attack odds, the AI still deems it too risky to launch an attack.

What is the AI doing and how to fix that?

Thanks

[knightofni]

Hi kami,

I'm in no way an AI expert - i actually came here to ask a question - but you can try the following commands

1. Combat

It orders the AI to focus more on a country as a whole. It's possible that the AI is not sending troops to fight the Irish because they are seen as a tiny opponent (in terms of IC, total army size, etc.. )

ex :

combat = {
IRE = 100
}

2. Province focus

front = {

province_priorities = {
'provinceid' = priority

}

This is some sort of multiplier for the front priority. Set something high, like 50 - 200. You can use your own provinces or the target province. To get the provinces ID check in Modding documentation or download DH companion

3. Switch AI

The province focus will make the AI reinforce Ireland whether there are enemy or not. To control it better, you can create an event that fires if ireland is invaded, that will teach the AI by telling it to 'switch' to the ai file that specifically tells it to reinforce ireland. Check the event commands / in game examples to learn how to do this.(db/events/1914_AI for instance)

Good luck

[kami888]

Hi, thanks for the suggestions. The combat thing didn't really help, it just made the British declare war on Ireland before schedule but not commit more troops.
As for province focus, I have actually tried that before, i forgot to mention in the post, but i did put in some province_priorities and even area_multipliers for the Irish provinces, and it wasn't helping.

However, following your advice, I have further raised those priorities and multipliers to something ridiculously high like 200 for province_priority and 50 for multiplier (before they were at something like 30 and 4.0 respectively). And... It appears that might have done the trick. Ireland doesn't seem to be a problem any more, although maybe I'm just getting lucky. But for now let's assume that this problem is solved, unless it surfaces again. Of course, even so, it's not at all a pretty solution, as the developer should never have to hardcode specific provinces for the AI to concentrate on so that it doesn't fail miserably, but it was one of a kind situation for the scenario which I had in mind and so it seems sufficient.

The other problem still stands though. The AI's recklessness is making it hard to have a viable scenario where there can be an organized retreat or even stalemate. I've tried the default AI (taking the default 1936 England AI), I've also tried making modifications to it as shown in my previous post, neither one performed anywhere near satisfactory at not getting themselves perpetually cut off and encircled by each other and losing the strategic provinces left and right. What can be done about it?

[froglegs]

Not quite sure where to post this AI question.

Playing 103 as USA 1933 it seems that Japan never seems to build enough CVs. I always give the AI +20 IC, res, manpower in the diffculty file and wrote an event that fires for Japan to build another +20 across the board making +40% total yet they still seem to only have 2 CVs by Pearl. I have edited the AI Japan and certain switch files for instance telling AI Japan to build 20% less inf and 20% more ICs to CVs in several HOI2 games successfully but it doesn not seem to work here in DH. I know to look under mods/DH full etc to find the working files.

Bottom line -- how to make AI Japan build more CVs?

[Supeerme]

In the Japan ai files there should be a build order like this:

(1936 ai file shown)
	#### Divisions etc...
   	infantry     = 75 
	cavalry 		= 5
	motorized 		= 0
	mechanized 		= 0
	light_armor	 	= 0
	armor 			= 0
	paratrooper 	= 0
   	marine       = 0 
	bergsjaeger		= 0
   	garrison    = 0 
    hq 				= 5
	militia 		= 0
#				0 %
   	interceptor    = 0 
   	multi_role     = 5 
	cas 			= 0
	strategic_bomber = 0
   	tactical_bomber = 10 
   	naval_bomber   = 0 
	transport_plane = 0
	flying_bomb 	= 0
	flying_rocket 	= 0
#				0 %
	battleship 		= 0
   	carrier     = 0 
	escort_carrier	= 0
	destroyer 		= 0
   	light_cruiser   = 0 
	heavy_cruiser 	= 0
	battlecruiser 	= 0
	submarine 		= 0
	nuclear_submarine = 0
	transports 		= 0
#				0 %

As you can see Japan will not be investing in any CV divisions. However if you change the number, Japan will build CV as a %. (Warning! The entire Division list MUST reach 100% in total so if you want to get 25% CV you must take 25% away from something else.)

[froglegs]

Supeerme -- We are on the same page here. I have done this with other HOI2 games in the past and recently with AOD. However, I get not immediate response here with DH. I can normally see a difference after a month of game time running with the change at the start of the 1936 scenario but I see no effect with DH.

I guess I'll try it again with a larger change and run it for a year of so and look again. Again -- what I see is that Japan has but 2 completed fleet carriers by Pearl when historically they had 6 CVs at Pearl so the CV production has to get juiced up by quite a bit.