Sorry, just noticed your message.

It is here:

cai_task_management_system_task_generator_groups_generators_junctions

...but I think I went back to the decimal values for one reason or another.

Yes, I went back to the more narrow spectrum because I think it was causing the AI to get confused. But I scaled the smaller spectrum to the previous 1-9 values, so it should be similar in regardles to priority:

"
Update: Dec 10, 2014 @ 9:53pm
Change XXVII.

-Reworked the AI personality priority system so that the spectrum is smaller, and thus the AI's actions are more abundant. There seemed to have been some manner of in-activity on the part of the AI before in the past update of this mod, but this has been corrected now.
"

Where can I see a version of your work? I have been trying to mod this table and would like to know the bounds and how to balance things.

Not sure if the link will be allowed here, or if it will even help, but I found my old XLSX database file where I scaled the AI behavior.

https://www.dropbox.com/s/wlouacxgpyk9msx/cai_task_management_system_task_generator_groups_generators_junctions.zip?dl=0

As mentioned in the comments though, unfortuanately this knowledge is a lost art for me (5 years ago).

In that xls, you were giving top priority to tasks by assigning 0.1 and you assigned up to 3 to low priority tasks. For ex, aggressive ai was given this:
2.5 aggressive_with_regions_task_generator_group CAI_TMS_TASK_GENERATOR_DEFEND_ALLIED_TERRITORY and

0.1 aggressive_with_regions_task_generator_group CAI_TMS_TASK_GENERATOR_ATTACK_ALL_ENEMY_SETTLEMENTS

Naval_Supremacy was given:
0.1 naval_supremacy_task_generator_group CAI_TMS_TASK_GENERATOR_DEFEND_SEA_REGIONS

Is this a mistake or did it actually work opposite the way CA had it set up in Vanilla? Have you seen any documentation from CA about the upper and lower bounds of these fields?

I am asking because I am trying to build an ai mod that actually does something significant for Attila.

Its not a mistake, no. Nothing in this mod is a mistake, simply by the fact that I was experiementing and playing it at the same time (hence the 155 chances, and the 900+ hours of playtime). But who knows, maybe I misunderstood something, but all I know is that the end result plays well to me. :)

I don't recall what what "vanilla" had it set at anymore, but the values that are there are from experimenting. I mean you can see, while using this mod, that the upper "3" behavior is not done as often as the lower < 3 value behaviors. The AI becomes more active in general. So, I don't know if CA themselves misunderstood what their own programmers had designed here, and maybe that is why in Rome 1.0 the AI sat around and did nothing. I suspect so.

I have not seen any CA documentation on the matter. It was just trial and error.

I think if you take what I have done, and then expand on it in Atilla and experiment a big, then you will get what you want. In the end too, the AI priorities only serve to nudge the AI into making a decision. It is assumed that if all the situations don't apply, then the AI will do what does apply, only using the applicable situations in its comparison of priority levels, but who really knows (apparently some people even at CA don't know). Even having the AI do random things sometimes makes them look intelligent.

What I mean is, its not like these priority values make the AI smarter, it just attempts to focus them and keep them from being stagnant. Just keep in mind that is the basic goal here (beyond that, it is probably down to die rolls and relying on the programmed AI behavior routine). But I suppose this is relatively obvious...

"then expand on it in Atilla and experiment big"
I think it works the opposit in Attila/Warhammer. It seems the order is reversed in Attila and Warhammer. Numbers over 1 are encouraged tasks while 0.1 is low priority. Negative is actively discouraging dumb moves like resacking. 0.1 is for chasing non human horder armies. 1.5 is for attacking enemy forces.
It is almost as the Rome2 task management table is just an afterthought. You expanded it nicely here.
But can you comment on the values for Attila : 0.1 vs 1?

I missed this message before.

Unfortunately I cannot comment on the Attila values because I have never edited them there. The AI seemed generally active in it, so I suspect that they changed something significantly in it.

(whoops, accidentally deleted the post)

Well, you got me curious so for the past month I have been experimenting with reversing the AI priority. It seems like you might be right in your assumption that the priorities were backwards. Very strange, perhaps way back in version 1.0 it was backwards, and I replicated it.

Then again its hard to tell exactly what the AI is doing, these priorities basically just nudge the AI to do something and what they decide to do after that is up to the code level behavior, but it certainly *seems* to play better. I will continue to investigate. Thanks!

The best way is to pick 1-3 metric values and make them do the exact opposite of what is being discouraged. For Attila, I like CAI_TMS_TASK_GENERATOR_ATTACK_LOW_FERTILITY_REGIONS
CAI_TMS_TASK_GENERATOR_RAID_RECENTLY_SACKED_SETTLEMENTS

CAI_TMS_TASK_GENERATOR_ATTACK_RECENTLY_SACKED_SETTLEMENTS
Then I give them a 2 instead of -2 to -5 and watch. That's my current plan. It takes at least 20 turns to see this in action in Attila.

Yes, that is the trouble (actually seeing the AI get into a situation to use the priorities).