How auto-resolve is calculated exactly ? what is the formula ?

[cedric37]

Hello,

I'm new to these forums, and although i have now spent some hours reading posts; i did not find any correct answer to this question: How auto-resolve is calculated exactly ? what is the formula ?

I don't have time to spend in playing battles and the strategic part of the game is my favorite part; so up to now i used to fight 100% with the "auto-resolve" option (and this just worked well, i have won already 10 campaigns between the different mods and the original game; without making a single battle in the tactical map).
I believed the AI to take into accounts a lot of parameters, but after reading a lot of posts here, i suspect that it is just important to get 19 high quality infantry units and many starred generals.
What is the truth really ?

Does someone digged into the game's mechanics and investigated how the auto-resolve really work ?
I would appreciate if i could get some clues; because i tried to make exhaustive stats of all units using the EDU file and try to make my own units dB with a global score per unit; but this seems to work only if i play battles on the tactical map.

Sorry if this post is a bit long.
Thanks in advance for the answers.

[Goofy]

I think that the computer calculates the amount and quality of your and enemy troops and then calculates the result. Note that the outcome may be different after the same battle if you reload your save.

I don't have time to spend in playing battles and the strategic part of the game is my favorite part; so up to now i used to fight 100% with the "auto-resolve" option (and this just worked well, i have won already 10 campaigns between the different mods and the original game; without making a single battle in the tactical map).

This is TOTAL WAR. I mean that there's nothing as fun as battles.
If you're new with M2TW, see my guide for example.

[cedric37]

@Finnking:
Thanks for the fast answer.
I used to play every battle when i played RTW, but i was a single and now i am married with 2 childrens and i get only 1-2 hours to spend each day on the computer (which is already too much for my wife ^^). That's the all issue... I cannot afford to play all the battles and i am getting used to auto-resolve them, so i would like to continue this way. I am still having a lot of fun playing (currently with SS6.1 which is a superbe mod IMHO but this is quite out of subject).
I know that i loose 50% (if not more) of the game fun, but like i said i just cannot afford to play the battles; i prefer to advance in the campaign (like 10-20 turns a day; so a campaign is more or less finished in a 2 weeks time which i think is enough before getting bored with a faction ^^).

[Goofy]

I used to play every battle when i played RTW, but i was a single and now i am married with 2 children

I thought that you can only marry with fully grown people
Well, then do it in your way.

[cedric37]

oops.. sorry i meant that "i am married and i have 2 children" ^^ But i guess everybody understood anyway. I must apologyze if my english is not really good, it is not my native language...

[Goofy]

oops.. sorry i meant that "i am married and i have 2 children" ^^ But i guess everybody understood anyway. I must apologyze if my english is not really good, it is not my native language...

Not mine too.
Don't worry, I just tried to be funny

[SoulBlade]

Heavy cavalry is also good in auto-resolve.

[cedric37]

@soulblade:
Yes from time to time some of my heavy cav units are doing great (like 0 casualties on my side 116 in ennemy's rank..) in the auto-result but it is very random; so i prefer to rely on heavy inf for at least 50% of my stack. I generally put 9-10 units of heavy inf (3-4 of them being effective against armour), 3-4 ranged units, 2-3 units of spearmen and the rest in heavy cav and until now it works quite well. But in fact i wonder if this is worth to bother except for unit cards diversity. That's the whole point.

I have also noted by examinating the auto-battle results, that crossbow do more casualties than longbow (playing with jerusalem faction in SS6.1 my templar crossbowmen are doing better than my dismounted french archers), and sometimes spear/pole units too are doing massive onslaught. For instance my halbbruder are doing more damage and taking less casualties than the ritterbruder, despite the unit stats in the EDU.

[Silent Assassin]

Roll de dice and that about how it works, it's work as a % and this is also good for Agents like spies when they attempt either a sabotage or a kill. I have seen 26% assassin kill their target, reload and tried it with a 80% assassin and he failed and died so IMO it's like if you were rolling dices or waiting for the card that you need to have the best hand in Poker Texas Hold'em.

[CobraStallone]

Roll de dice and that about how it works, it's work as a % and this is also good for Agents like spies when they attempt either a sabotage or a kill. I have seen 26% assassin kill their target, reload and tried it with a 80% assassin and he failed and died so IMO it's like if you were rolling dices or waiting for the card that you need to have the best hand in Poker Texas Hold'em.

Almost but not quite I think, autoresolve 2 units of pesants against a general, an archer unit and an infantry one like 100 times and the pesants will always loose.

[John Doe]

Roll de dice and that about how it works, it's work as a % and this is also good for Agents like spies when they attempt either a sabotage or a kill. I have seen 26% assassin kill their target, reload and tried it with a 80% assassin and he failed and died so IMO it's like if you were rolling dices or waiting for the card that you need to have the best hand in Poker Texas Hold'em.

A little of this will explain the load/save difference in result (can be as dramatic as victory to defeat or vis versa).
Second battlefield is taken into account, cavalry are bad in siege battles and archers good in mountains.
Third there is a rock/paper/scissor effect, cavalry against pikes don't do well, pikes against xbows either, xbows against cavalry...

So if you take those 3, add experience, armour, generals command/dread/chivalry/traits, different army composition, and even weather (gunpower in the rain??), you have a rather complex formula.

[sherlock]

Are you all sure it's just some formula to calculate a probability?
I was rather thinking the application runs a full no-visualization simulation of the battle.

The reason of this is:

1) In battle AI sometimes behaves horribly stupid when it has just 1 unit of bodyguards. They tend to stay idle, awaiting to be attacked (at least in vanilla): thus series of charge/retreat with other bodyguards or any stupid unit of archers can win an apparently impossible-odds battle.

2) When I autoresolve attacking with bodyguards alone a stack of bandits, often I saw, against stacks made of only 1 peasant archers, with odds like 10:1 for me, that autoresolve gave methodically clear defeat.

It's like autoresolve did exactly the same errors as battle AI. No amount of bad luck can justify the number of disastrous autoresolves I saw in conditions as above described.

This led me to think that autoresolve is a full battle simulation, AI Vs AI.
All in all, battle AI algorythms probably need a minimal fraction of computing power compared to graphics (although the latter obviously is delegated to graphic chip) so it's well possible that battle simulation and thus results can be calculated in just some tenths of seconds.

[John Doe]

That won't explain the difference when you save/load. You can get a complitely different outcome on the second autoresolve. And I had some battles with multiple full stacks on both side, it would take longer to calculate those autoresolve on my pc. It's got to be a fornula.

[sherlock]

That won't explain the difference when you save/load. You can get a complitely different outcome on the second autoresolve. And I had some battles with multiple full stacks on both side, it would take longer to calculate those autoresolve on my pc. It's got to be a fornula.

Even more then. If it is just a formula then were would be no difference in outcome with save/load, as it happens with assassins missions when you reload/retry without doing anything different (see this post)

Don't underestimate your PC. The lowest PC that can run M2TW is many, many times more powerful than the supercomputers that helped to take the men on the moon 40 years ago.
If you run a large battle at 6x speed, you cannot usually (because humans are not quick enough to take decisions and control units at that speed unless there are just a few units), but battle AI has no problems. If you could see visualization of AI vs AI battle at 100x rest assured your CPU wouln't have problems in taking decisions for units and calculate results of the blows. It's just the graphic chip who would suffer having so many perspectives to calculate and would "lose" fluidity of movements.
With no visualization an AI vs AI battle can probably be run/computed in a very small time.

Also, from the point of view of developers, if they have ready the program modules/routines for battle AI and they know PC can run it very quickly with no graphics, why should they invent some complex formula if they can simply run a simulation?

[cedric37]

Thanks for your insight sherlock (you are well named ^^).

At start i too thought that computer was making rapidly a battle AI vs AI; and your analyze seems to confirm this. Your example with a general bodyguard loosing against peasants is like a proof for me since in formulas, the stats of the general would give hime always a clear victory against peasants.


What kind of units are stupidly used by the AI (don't say all ) ? This will give me some tips on how to recruit units to compose my armies.

[John Doe]

Even more then. If it is just a formula then were would be no difference in outcome with save/load, as it happens with assassins missions when you reload/retry without doing anything different (see this post)

Don't underestimate your PC. The lowest PC that can run M2TW is many, many times more powerful than the supercomputers that helped to take the men on the moon 40 years ago.
If you run a large battle at 6x speed, you cannot usually (because humans are not quick enough to take decisions and control units at that speed unless there are just a few units), but battle AI has no problems. If you could see visualization of AI vs AI battle at 100x rest assured your CPU wouln't have problems in taking decisions for units and calculate results of the blows. It's just the graphic chip who would suffer having so many perspectives to calculate and would "lose" fluidity of movements.
With no visualization an AI vs AI battle can probably be run/computed in a very small time.

Also, from the point of view of developers, if they have ready the program modules/routines for battle AI and they know PC can run it very quickly with no graphics, why should they invent some complex formula if they can simply run a simulation?

from http://www.twcenter.net/forums/showthread.php?t=292140

Have you retried after doing something different?
Whenever there is something with a 1-100 percentage PC "seems" to roll a random number, but really take the next in a predetermined randomized series of 1-100 numbers.
If his next numbers are, e.g., 68-12-47-...., each time you retry with your merchant with, say, 45% chance, you'll always pick 68 and fail (you must pick lower than your success percentage).
If you reload and FIRST send a spy for an easy 95% mission, the spy picks 68 and succeeds and THEN your merchant picks 12 and succeeds too.

This mechanics (no "random roll") are implemented, I guess, to avoid an easy exploiting by players of save/reload trick whenever there's a random roll, but are not exploit-proof anyway.

I was looking for the link to answer you (didn't know it was from you!! +rep for that one). The difference in outcome (between curent game and first reload, the second and third reload will be same as first) would be explained by the use of random number within the formula (like when reloading a spy or assassin mission).

The problem I have with your theory is that they do already use a formula for naval battles (no one can deny that). They just needed to extrapolate to fieldbattles. Second, when you are outnumbering in an attack siege battles, you will get almost no casualty (sometime 0, impossible if AI vs AI, AI has to kill them all in the city square), that one does not fit in your explaination, but in a formula you can approximate to the nearest integer (sometime 0: not much divided by a lot = 0 , times another variable = 0). As for my argument about PC speed, I meant that the result is almost spontenious, even when huge army are involve (I knew there is no need to calculate the video display). There should be at least a small lag and some computation noise.

[sherlock]

.... There should be at least a small lag and some computation noise.

I think (if I just knew if it's possible to run M2TW in a window and how to do it) a good test would be to run in autoresolve a quite large battle AND meanwhile check Windows "task manager".
If my "AI vs AI simulation" theory is correct, I'd expect a brief but evident peak both in CPU work and in memory usage by M2TW application.
If it's just a formula, it shouldn't use much RAM memory. If it's a battle simulation there would be quite larger RAM memory involved.
If it's a battle simulation, CPU usage should have a few tenths of seconds of high activity when you click "autoresolve"; that wouldn't be justified by a simple formula (that would just use some nanoseconds of CPU work).

[Goofy]

What kind of units are stupidly used by the AI (don't say all ) ? This will give me some tips on how to recruit units to compose my armies.

See the link in my sig, then check "Units" and "Forming an army" sections. They work against human as well as AI.
Tip: if enemy has archers and you have horse archers, use the "circle and shoot" ability; the AI can't shoot your horse archers when that's done. That's one of "AI unit using bugs".

[cedric37]

Yes but in that case you use your human brain to overcome stupid AI.

My question was more oriented toward an AI vs AI outcome.
I mean, in your example if AI command horse archers and opponent AI command foot archers, the horse AI will maybe never use the circle thing, so you don't gain any advantage by putting horse archers in your stack, when planning to auto-resolve the battle.

Ooops that's a long sentence, keep your breath ^^

[Goofy]

Then use powerful troops like (a lot of) heavy cavalry and good spearmen to protect you against enemy cavalry and to kill their spearmen.