Information about developments at snowballing problem

[Blood Gryphon]

@mexxico Hey, I was looking at the new troop upgrade code (in UpgradeReadyTroops and CalculateUpgradeChance), and I wanted to point a few things out that might affect snowballing. I think Khuzait npcs are actually upgrading their recruits to the cavalry branch at a higher rate than they were prior to the change.

Before the change, I think how it worked is that they would just split troops 50/50 between the two upgrade branches. The problem now is that when they recruit a bunch of T1 Khuzait Nomads, this pulls down their cavalry ratio below 0.36 (because Nomads count as infantry), so they start funneling upgrades into the cavalry branch at a >50% rate. This means a higher proportion of their Nomads get sorted into the cavalry branch than the non-cavalry branch than before. Here's an example of what I mean:

Spoiler

Anat already has more Khuzait troops upgraded into the cav branch than non cav branch, but since she has 71 Nomads in her party, her cav ratio is only listed as 0.233, and the game therefore calculates a ~70% chance to upgrade her Nomads into the cavalry branch vs. 30% to non-cav. If you plug in the numbers, even at a 0.3599 cav ratio, the game will still be funneling Nomads into the cavalry branch at about a 63 to 37 split. After her cav ratio rises above 0.36, the upgrade split will return to 50/50. Compare that formula to 1.5.6, where the overall split would be a constant 50/50, regardless of cav ratio.

This means Khuzait as a whole will be getting more cavalry, at a faster rate. If you look at this comparison of the troop composition of two Khuz. armies from 1.5.4 (left) and 1.5.7 (right), you can see that the 1.5.7 army has a higher proportion of troops from the cavalry branch than non-cavalry branch (especially when looking at Tribal Warriors vs Footman):

Spoiler

I'm not offering this as hard proof, since it's only two data points, but it helps illustrate what I mean.

Then the other point I wanted to make is that for the other factions there is a significant amount of "lag" time between when a T1 recruit gets sorted into the half of the troop tree with cavalry, and when it actually becomes a cavalry troop. Specifically for the Empire, they only have one cav troop at T5 in their main line of troops, so a high proportion of recuits get sorted into the archer branch, but relatively few of them survive long enough to become horse archers. This isn't much of an issue on its own, but it makes the tactics perk Tight Formations (Infantry deal 10% more damage to cavalry in auto-calc) less useful against Khuzait since fewer troops get sorted into the infantry branch. Here's another screenshot of an Empire army in 1.5.7 with many more archers than infantry, but relatively few cavalry (only a ratio of 0.04):

Spoiler

So two possible solutions:

1. Only count T2+ troops in the cav ratio. There are no T1 cav troops in the game, but most non-bandit T1 troops have the potential to upgrade into cavalry eventually. This will hopefully prevent the game from sorting so many Nomads into the cavalry branch of Khuzait parties.
2. Allow npcs to recruit more tavern mercenaries (specifically Watchmen and Scouts). This gives factions more abundant access to cavalry troops quicker, but currently npcs don't recruit many mercenaries because there is a high gold requirement in place that only clan leaders can meet (though make sure there are still plenty of mercs left for the player ). As a side note, it seems the mercenary selection code might be bugged, because it isn't spawning any troops from the Scout branch of the merc tree that I can tell (unless that's intended).

That's overstating the advantage a bit though, as any unit can kill any other unit first try, since it's all probability based. If you crunch the numbers from the auto-calc model, a T5 cav unit has about a 40% chance on average to kill a T4 infantry unit on any given hit, but a T4 infantry unit also has a 22% chance to kill a T5 cav unit on any given hit. If you compare T5 cav vs T5 inf, it's a 34% vs 26% chance respectively. So even though the cavalry unit has a 20% bonus applied to its power level, this only shakes out to be roughly an 8% greater chance to kill a non-cav unit of equivalent tier than vice versa.


Strategic reasoning is the hard part though. You can add as many rules as you'd like to the actual auto-calc model (TW has already implemented conditional terrain and unit-countering perks in the tactics tree, for instance), but with those extra rules thrown in, making the AI understand how their party strength compares to every enemy they come into contact with before they attack isn't exactly trivial. I'm not an expert, but you'd probably have to run a bunch of strength comparisons for each of the different troop types between parties for every hostile party that comes close to every other party, or if the ruleset is too complex, just execute the auto-calc code in advance. Considering mexxico said this:

a few pages back, I think they probably want to keep that aspect of the AI code streamlined.

(Sorry for the novel, guys )

Finally read through this, nice find man!

I think both your solutions would be nice adds, obviously the 1st one is the most important. Another solid analysis.

 

[84Slashes]

https://forums.taleworlds.com/index.php?threads/the-siege-of-gallianum-a-play-through-of-a-campaign-with-screenshots.53887/

Is this what you're referring to?

Oh wow, the thought and detail put into that post is striking, exactly the kind of depth I would love to see in a game like Bannerlord. It would be wonderful if we could hear from some of the devs working on the campaign logic about the challenges they found and if/why they decided to abandon that level of depth.

In my opinion having terrain effects, raiding parties, foraging, scouting, etc. would be huge improvements over the current simplified parties/armies system, so if that direction wasn't feasible for TW would be nice to know why.

 

[froggyluv]

Oh wow, the thought and detail put into that post is striking, exactly the kind of depth I would love to see in a game like Bannerlord. It would be wonderful if we could hear from some of the devs working on the campaign logic about the challenges they found and if/why they decided to abandon that level of depth.

In my opinion having terrain effects, raiding parties, foraging, scouting, etc. would be huge improvements over the current simplified parties/armies system, so if that direction wasn't feasible for TW would be nice to know why.

So glad that thread was found as basically everything ive been trying to convey at the Campaign level of bannerlord as well as its snowballing problem -could have been solved had they gone this organic route. Again, everything about BL's strategy tactics is artificial -these lords arent making an hour by hour decision of when to siege, send out outriders to plunder the area or look for counter skirmishers who may be hiding. Of attacking an enemy that is slowed down because of a terrain based hillside etc.. thats riveting tactical play as well as the natural remedy for a world with "too many wars going on" -a strange but very real complaint i hear about this title all the time. Why on earth would anyone complain about "too many wars" in a war game unless.......they felt arbitrary, not well thought out and spammy -artificial...

That thread shows it CAN be done even at 2008 level tech -by one guy -so it could be done bigger and better now should they have chosen to go that route.

 

[stevehoos]

So glad that thread was found as basically everything ive been trying to convey at the Campaign level of bannerlord as well as its snowballing problem -could have been solved had they gone this organic route. Again, everything about BL's strategy tactics is artificial -these lords arent making an hour by hour decision of when to siege, send out outriders to plunder the area or look for counter skirmishers who may be hiding. Of attacking an enemy that is slowed down because of a terrain based hillside etc.. thats riveting tactical play as well as the natural remedy for a world with "too many wars going on" -a strange but very real complaint i hear about this title all the time. Why on earth would anyone complain about "too many wars" in a war game unless.......they felt arbitrary, not well thought out and spammy -artificial...

That thread shows it CAN be done even at 2008 level tech -by one guy -so it could be done bigger and better now should they have chosen to go that route.

Much of this entire game seems to be cutting corners. It's really not very innovative at all.

 

[Nalgasucia507]

Yes. Hopefully the future will slowly but surely proves us wrong.

 

[Bannerman Man]

Shame there is no link to download that mod even if only partial finished -would love to rip open that code and have a looksie

You can actually still download it here if you really want to (it's the TidesOfConquestSicilyNoItems.rar file). I took a look (not at the code, just the gameplay) and it's got some interesting ideas, but there's definitely plenty of wonkiness to the AI. There's a lot of standing around and some back and forth cat-and-mouse stuff going on. I uploaded a couple videos showing what I mean here and here (note, in the first video the reason the pursuit is so slow in the beginning is because both parties have maxed fatigue. There are also some unseen farm plots around each settlement which the AI burns down that explains some of the standing around, but not all of it). But other than that it seems pretty unfair of me to criticize someone's unfinished project that they quit working on more than 10 years ago, so I'll leave it there (it's also straying further off topic). You can make your own judgements though.

 

[froggyluv]

You can actually still download it here if you really want to (it's the TidesOfConquestSicilyNoItems.rar file). I took a look (not at the code, just the gameplay) and it's got some interesting ideas, but there's definitely plenty of wonkiness to the AI. There's a lot of standing around and some back and forth cat-and-mouse stuff going on. I uploaded a couple videos showing what I mean here and here (note, in the first video the reason the pursuit is so slow in the beginning is because both parties have maxed fatigue. There are also some unseen farm plots around each settlement which the AI burns down that explains some of the standing around, but not all of it). But other than that it seems pretty unfair of me to criticize someone's unfinished project that they quit working on more than 10 years ago, so I'll leave it there (it's also straying further off topic). You can make your own judgements though.

Heh i cant even get Mount and Blade 1 to run on my PC. No offense man but those videos dont shown much as its nearly impossible to gather whats happening without having a Key to all those AI codes present and also without further direction from you. Note the way Nijis explains things with such a good breakdown in his post -that would be far more helpful even if the AI is actin wonky. And of course itll be wonky - it started as a proof of concept i remember following it before he decided to make an actual mod ot of it (Sicily) - it was just a new campaign strategic idea he had and many of us followed it in another thread.

The point is 'what can be done' and that is clearly still in proof of concept state by 1just one dev. Again -what could have been made had they thrown the weight of the full dev team behind it -guess we'll never know. The breakdown he makes in his first post far exceeds anything ive experienced as world map lord behavior in bannerlord and its not even close. Ill always take wonky but determined AI behavior than monotonous sludge with zero narrative

 

[Sarissofoi]

Yo fellow human beings

Supply Lines

Allows the player and AI to send food supply caravans between fiefs the clan owns and siege armies. Select troops from your garrison to escort your supply caravan to it's destination or let the AI han

www.nexusmods.com

this mod may be to more to your liking

 

[froggyluv]

Yo fellow human beings

Supply Lines

Allows the player and AI to send food supply caravans between fiefs the clan owns and siege armies. Select troops from your garrison to escort your supply caravan to it's destination or let the AI han

www.nexusmods.com

this mod my be to more to your liking

Wow man -great find how did i miss this?! This guy Cheyron is like a 1 man army of code

Description:

Guerilla Attacks:
Bannerlord lacks that realistic medieval warfare feel. In native castles are absolutely uses. Garrisons with 400 men just watch armies march by and nothing they can do. In reality this was not the case and it was a tactical blunder for an army to march past an enemy garrison and not besiege the garrison. A garrison is dangerous to an army in enemy territory since they can intercept any supply lines. They will not meet an invading army in a pitched battle. They will attack you asymmetrically with guerilla warfare. An army that is marching will be in a long line with large parts of the army not able to communicate quickly to each other. Once you are able to make a response, they will simply run back behind their walls. This is what castles were for and how they protected against invasion from larger forces. They force an army to stop and fight an enemy behind walls rather than a field. If you ignore enemy garrisons, they would use these opportunities to chip away at you while they message their main army your army size and location. You would basically be committing suicide to leave an enemy garrison behind you. Bannerlord does not capture this and it is disappointing but this mod should help now! See the MCM for settings to tweak how this feature works. You can set the effective distance for garrison scouts to sally out and launch guerilla attacks on your party. You can set the frequency or interval in game time hours that the script checks the positions of parties around the map to see if they are subject to these attacks. You can also set the probability these events occur and parties with higher scouting skills will be harder for garrisons to attack.

 

[Razeo]

Yo fellow human beings

Supply Lines

Allows the player and AI to send food supply caravans between fiefs the clan owns and siege armies. Select troops from your garrison to escort your supply caravan to it's destination or let the AI han

www.nexusmods.com

this mod my be to more to your liking

Oh wow! Thanks for this!

 

[Bannerman Man]

No offense man but those videos dont shown much as its nearly impossible to gather whats happening without having a Key to all those AI codes present and also without further direction from you.

I do have the key to the AI codes though. It's in the text files. The codes do indeed tell you what role the AI is assuming at that moment in time, but don't really tell you what factors led them to perform that action in the first place, and don't always match the action the AI is performing. But even with these codes in mind, in my experience, as the AI are chasing each other around, they are still prone to making bizarre decisions. From the player's perspective it can look irrational and make the outcomes feel more like a fluke than result of strategic thinking or "smart" AI.

The description laid out by the developer is definitely cool, but I think it anthropomorphizes the AI a little too heavily, which projects a bit more strategic reasoning and forward thinking onto them than what they actually possess.

That said, credit to the dev, because the AI do execute their basic roles well, and the AI as a whole is by no means bad. The mechanics still make for an interesting dynamic, and I especially like how the farm plots are used.

Edit: But I don't even know why I'm critiquing such an old mod; there's no point. We can just move on.

 

[froggyluv]

this mod my be to more to your liking

Oh wow! Thanks for this!

Ok its a nice concept but its basically just a message and auto calc of a small guerrilla attack -its not an actual party being dispatched (like the mod above)

 

[froggyluv]

I do have the key to the AI codes though. It's in the text files. The codes do indeed tell you what role the AI is assuming at that moment in time, but don't really tell you what factors led them to perform that action in the first place, and don't always match the action the AI is performing. But even with these codes in mind, in my experience, as the AI are chasing each other around, they are still prone to making bizarre decisions. From the player's perspective it can look irrational and make the outcomes feel more like a fluke than result of strategic thinking or "smart" AI.

The description laid out by the developer is definitely cool, but I think it anthropomorphizes the AI a little too heavily, which projects a bit more strategic reasoning and forward thinking onto them than what they actually possess.

That said, credit to the dev, because the AI do execute their basic roles well, and the AI as a whole is by no means bad. The mechanics still make for an interesting dynamic, and I especially like how the farm plots are used.

Edit: But I don't even know why I'm critiquing such an old mod; there's no point. We can just move on.

Cool that you checked it out so thoroughly but gotta admit -kinda amusing your enthusiasm to how do i say.... "critique" it to such a degree

Again the point was to show more as proof of concept something achieved so long ago with the original iteration of this engine so again...imagine had they invested in that course or better yet decided to now or for future titles. The point was to show how real medieval dynamics create their own storylines and i would guess would give far more variety in how games play out thereby dundundunDAH!!! Prevent snowballing by natural means rather than the artificial ones we have now.

Ok -lets let this rest as i rest my case

Edit: The reason i thought this relevant despite its age is that it was being designed by an actual Taleworlds Developer not just some random modder

 

[Sarissofoi]

@mexxico
Back to snowballing topic:
So there is few problems with current snowball meme.
First:
>>exp AI cheat
>AI Lords get 5+unit Level exp per unit
That is pretty big especially that higher tiers don't need much more exp from lower levels
pic related:

Spoiler: CODE

This help snowball because losing side get fresh troops that level slowly and they lose them all the time and winning side is winning battles which mean most of their soldiers survive(even if wounded) and get exp from battles and on top of it also get exp ticking from cheat.
It lead to whole armies being elite to the point that army that stay at peace will be composed mostly from tier 5 units,
which of course bankrupt AI because it can't support their troops from fief income alone.
So AI EXP cheat favour winning side which of course affect snowballing. There is also effect of higher tier armies having more access to cavalry that is usually higher tier and thus granting it extra global map mobility and on top of that extra battle ability from cav auto calc bonus.


There is also another snowball factor.
>loot coin aka war economy
Winning side get:
>extra cash that come DIRECTLY from defeated lord treasury(which can easily bankrupt poor clans) up to 10k denars(I think but I may be wrong)
>cash from ransoms from prisoners and lords or influence
>cash from loot from defeated soldiers(conveniently they get straight gold conversion, not items that they need to sell in cities)
by the mexico word loot coin is about 35% of lords income in war.
Of course its more for winning side and thanks to that they can afford their elite armies and fill their chests with gold but losing side not only gain nothing but also lose their change and savings they have in their small chest.
The losing savings directly contribute to poor clans having little money because they can't save any cash.
I would get rid of it(the DIRECT transfer). Both player and AI get loot(or loot coin) and prisoners - that should be enough(if not please increase value of prisoners). Poor clans then have to decide if they can pay ransom for their members or not and have their secondary member raise party. Right now they are straight robbed and from all their gold.
It obviously help winning side more as losing side is unable to keep any coin in their war chest.

So this two factors help each other snowball
>winning faction get extra helped by exp cheat(especially because higher tier troops have higher chances to get wounded than outright killed, do it help them a mass higher tier units)
>and they can sustain their elite armies because of loot coin(and direct robbery of defeated clans treasury), where poor side rely on losing their higher tiers troops to avoid bankruptcy

Proposition.
>remove direct robbery of defeated Clan treasury
>reverse the EXP cheat from
current 5+unit Level to 31-unit Level
that way 1-2 Tier units will level up quickly but it will slow down for higher Tier units making AI Lord armies mostly average Tier which are cheaper for upkeep when being reasonably effective. It would also have bigger effect on losing side when being less effective for winning one.

To be honest facing armies composed of mostly T5 units is not really fun as it force player to have max grinded units too and its costly both for player and for under performing AI factions.
Also making Cavalry troops having higher upkeep(like in Warband) could maybe help some weaker factions.

 

[Bannerman Man]

The point was to show how real medieval dynamics create their own storylines and i would guess would give far more variety in how games play out thereby dundundunDAH!!! Prevent snowballing by natural means rather than the artificial ones we have now.

I guess I don't see any reason to believe the mechanics or AI from that mod would prevent snowballing just by virtue of having more moving parts. Feedback mechanisms (and balance tuning) prevent snowballing, but I don't think that mod is an example of that.

 

[Sarissofoi]

What we need is a Civil War mechanic.

 

[Bannerman Man]

So you can get from 5 to 6 tier 3 times faster than from tier 1 to tier 2.

Hey can you explain your math on this part? I think you might be misinterpreting the XP requirements.
Tier 1: 300 / (6 + 5) = 27 days
Tier 5: 1500 / (26 + 5) = 48 days

What we need is a Civil War mechanic.

?

 

[Sarissofoi]

Wait. So exp isn't threshold just straight cost from tier by tier?
I stand corrected then.
Well that is even better and it not change that Exp AI cheat help army with higher tier units more than with the lower one.
I would say changing bonus to scale in reverse would be even more beneficial for AI because of upkeep cost and would help losing AI side to catch up.

 

[Apocal]

I guess I don't see any reason to believe the mechanics or AI from that mod would prevent snowballing just by virtue of having more moving parts. Feedback mechanisms (and balance tuning) prevent snowballing, but I don't think that mod is an example of that.

I feel like a lot of people use "prevent snowballing" to add legitimacy to what they think are cool ideas. And some ideas are really cool but... they don't really address snowballing. The core of snowballing is winning factions being able to exploit their initial successes into permanent dominance, so any mechanic that doesn't directly gut a large faction's power doesn't do much.

Like, I tried to see if just gutting the army numbers helped and it did, somewhat, but it was a bandaid fix and nothing like a balancing mechanic.

 

[Sarissofoi]

I do not care much about snowballing TBH. I want to see emergent story telling and organic created situations more.
Problem is the whole current system isn't really good for that.