[EB MP]3.0 Thread - Testing and Updates

[vartan]

Please refrain from making religious references.

What Shak said.

[-Stormrage-]

very well

[gamegeek2]

Though I've heard about stamina (and morale?) effects of knocking down but not killing... and at least in my obsverations there is a difference in the percentage of knockdowns compared to kills for high lethality vs low lethality.

That's probably because higher lethality generally means greater kill rate (that may have something to do with knockdown chances).

[antisocialmunky]

You should just set lethality at a standard value and change attack values around like what Medieval 2 does.

Its not like we use knock down for anything and it'll be easier to (re)balance. But that's a 4.0 EDU issue probably.

[TheShakAttack]

You should just set lethality at a standard value and change attack values around like what Medieval 2 does.

Its not like we use knock down for anything and it'll be easier to (re)balance. But that's a 4.0 EDU issue probably.

That would certainly make life a lot easier for us lay people to interpret unit effectiveness!

[Kival]

That would certainly make life a lot easier for us lay people to interpret unit effectiveness!

I don't like that but that's perhaps just me.

[antisocialmunky]

Unless you're going to make certain weapons have more 'impact' as in fewer bigger swinging weapons like hammers, long swords, and axes using high lethality, low attack.

I guess you could also do stuff like lower heavy infantry lethality, increase attack and increase light infantry lethality, decrease attack so Heavy Infantry vs Heavy Infantry take forever and prone to 'pushing' due to knock down and light infantry have better shock and do more damage initially due to charge value allowing for more successful hits instead of flinches on charge.

I really don't know.

[Brave Brave Sir Robin]

Or just take away good stamina on every heavy infantry unit.

[Lazy O]

I support that. And make a script to remove guard mode.

[Galvanized Iron]

I call upon GI to give us the Interpretation of this hadeeth.

No need for that, hopefully everyone is familiar with Aradan's excellent EDU guide:
http://www.twcenter.net/forums/showthread.php?t=111344

Book of Aradan 2:11

[lethality] : Percentile chance of a soldier to kill an enemy (assuming his strike has found its target). The higher the lethality, the more the kills and the less the knock-downs/knock-backs during a battle. Greatly affects the speed at which melee battles are resolved, since higher lethality means less missed hits and greater casualties in short time. It is not used in ranged combat.

I can through long experience of play around with this attribute assure that it is exactly as Aradan states it, reason everyone is not lying on their backs is becuase a failed lethality can also result in a not so visible knock-back.

The difference between a 0.2 lethality and 0.3 lethality unit is 50% more kills, which is huge and clearly visible in combat.

[gamegeek2]

GI, lethality is not a direct proportion as once thought. See phalanx_man's guide that Aradan links to in the guide.

[Kival]

And make a script to remove guard mode.

But not without some compensation for shieldwall/spearwall infantry like hoplitai.

[Brave Brave Sir Robin]

But not without some compensation for shieldwall/spearwall infantry like hoplitai.

ASM already played around with spacing and mass to create very close order infantry that are effective from the front. I'm not sure how this would affect flanking maneuvers and such though.

[Kival]

Pikemen without guardmode are pretty annoying though because they use unit cohesion this way and make crazy side attacks. Other units can be also annoying for micromanaging without guardmode but for pikemen there is as far as I know no way to prevent that without guard mode.

[Galvanized Iron]

GI, lethality is not a direct proportion as once thought. See phalanx_man's guide that Aradan links to in the guide.

I know from before, that skeleton, unit radius, weapon type etc also affects the lethality, but all factors being equal for two unit then lethality should be quite linear, no? In other way do you have access to a formula for lethality scaling? (Only the pure lethality part, not chance to kill that involves lots of other factors as we all know)

[gamegeek2]

Again, PLEASE read phalanx_man's guide...Aradan links to it in his guide.

[vartan]

I know from before, that skeleton, unit radius, weapon type etc also affects the lethality, but all factors being equal for two unit then lethality should be quite linear, no? In other way do you have access to a formula for lethality scaling? (Only the pure lethality part, not chance to kill that involves lots of other factors as we all know)

It would have been nice if the curve was linear but it's a power function.

EDIT: By nice I mean convenient, not realistic.

[gamegeek2]

It's not a power function, it's linear but has a non-1 slope and a y-intercept.

[Kival]

According to the test of phalanx_man, the effect of lethality is not linear (in a strict sense) but follows an affine function (no power function):

f(x)=2.98 *x + 0.54

(where x is the lethality, and f(x) the killing rate; to be more specific it's the percentage of men of the unit killed in a minute)



This is tested for a solid value of all other effects (attack, defense, etc.), but and that's a very, very big but, there's a ton of methodology problems here:



1) he made only 4-8 data points for every lethality, so the statisical failure is pretty high.

2) he said for himself that lethality showed other killing rates for other battle configurations, so the data seems not be very objective but pretty specific for the situation he tested them in.

3) We only know the effect on killing rate this way. We cannot conclude anything if there is a difference in not dying because of non-working attack or the attack beeing only non lethal. It's still possible, lethality is used as a 1to1 probability after the attack effect has been concluded. I don't see a way to really test this.

4) I don't know how he ruled out more complex non-affine functions. R^2 for his model (and I get the same recalculating it) is pretty high though (R^2=0,9992; so at least his linear model has a high quality).) I did not test for linearity further for now, don't have any real statistic software on this pc.



EDIT: My conclusion is, that phalanx men tests (on lethality) are not very conclusive :(

[Galvanized Iron]

According to the test of phalanx_man, the effect of lethality is not linear (in a strict sense) but follows an affine function (no power function):

f(x)=2.98 *x + 0.54

(where x is the lethality, and f(x) the killing rate; to be more specific it's the percentage of men of the unit killed in a minute)



This is tested for a solid value of all other effects (attack, defense, etc.), but and that's a very, very big but, there's a ton of methodology problems here:



1) he made only 4-8 data points for every lethality, so the statisical failure is pretty high.

2) he said for himself that lethality showed other killing rates for other battle configurations, so the data seems not be very objective but pretty specific for the situation he tested them in.

3) We only know the effect on killing rate this way. We cannot conclude anything if there is a difference in not dying because of non-working attack or the attack beeing only non lethal. It's still possible, lethality is used as a 1to1 probability after the attack effect has been concluded. I don't see a way to really test this.

4) I don't know how he ruled out more complex non-affine functions. R^2 for his model (and I get the same recalculating it) is pretty high though (R^2=0,9992; so at least his linear model has a high quality).) I did not test for linearity further for now, don't have any real statistic software on this pc.



EDIT: My conclusion is, that phalanx men tests (on lethality) are not very conclusive :(

Exactly, I was thinking "wtf?" When I read about his tests, a formula for kills per minute that is even leaving out the attack rating(!). First of all, all tests vs the computer is only approximate at best as it can't be considered an experiment in a fully controlled environment and secondly unless CA actually would share us their formula it is just speculation.

It's not a power function, it's linear but has a non-1 slope and a y-intercept.

In English please, do you even know yourself what that means?


As you may have noticed some units also has the hidden stats critical hit, which was visable in older mods, I think RTR 4, which seem to indicate that hit is done through an attack + die roll vs defense calculation whereas lethality is a seperate after-calculation.

[Galvanized Iron]

As a further proof of linearity I would make as an example my cavalry charge tests made for RS II MP. Starting out from 0.2 lethality a frontal charge of cataphracts vs marines resulted in about 10-12 casualites each first charge, increasing to 0.4 lethality casualties rose to 20-24 kills per hit, at 0.5 where it was good enough kills landed at a 25-30 kills per first time charge.

(These numbers are higher in the current state of the mod, but due to an increase in cavalry mass outside the EDU)

[Kival]

Exactly, I was thinking "wtf?" When I read about his tests, a formula for kills per minute that is even leaving out the attack rating(!).

That's actually fine. When you control other variables you can test for only one without any problem. You just need to take care what it does exactly mean. Phalanx man tests are nice, some results especially regarding other issues than lethality. No need to rage about it. The tests are because of the mentioned things to be taken with caution though.

In English please, do you even know yourself what that means?

He does know what it means and is right as far. He's using pretty general school terms, nothing crazy as afine functions :p.

[vartan]

Kival, you'd be surprised that phalanx_man's data is near-representative. That's why he got close to actual numbers most likely used by CA. Like I was telling gg2 in conversation recently, it's not as if someone at CA sat down and said, "Today I think I'm going to make a linear proportionality model connecting attack, lethality and defensive values such that the first coefficient is 2.98742." No. You start simple and complicate things from there to refine them. That's why they most likely landed on 3 and 0.5, not 2.98 and Heaven knows what other fancy numbers.

[gamegeek2]

GI, I learned what that means when I was about 12 years old. Please don't insult my intelligence.

[antisocialmunky]

GI, I learned what that means when I was about 12 years old. Please don't insult my intelligence.

If you paint things red, they go faster, this is why Saka and Rome are OP.

[gamegeek2]

But Rome isn't OP :p

[vartan]

But Rome isn't OP :p

True dat.

And I don't think that was cool either GI. Why you be hatin' brah?

[Kival]

Kival, you'd be surprised that phalanx_man's data is near-representative.

How do you know that? Only according to his test data we can't know that, actually for lethality he said that he got different restults for other battle configurations. I don't know how different though.

No. You start simple and complicate things from there to refine them. That's why they most likely landed on 3 and 0.5, not 2.98 and Heaven knows what other fancy numbers.

Agreed. If phalanx-man was conclusive, the real numbers would be 3 and 0.5, but we don't know if they used a linear model at all. As far as I can see without real testing, to use a linear model for this data is okay. What we can be sure of is that f(0)>0, so lethality obviously isn't the chance itself.

Still I'm not sure about the validity of the data itself. If I have the time, I'll try to re-model the system with some hypothesis and see if it fits with the inductional "models" there. Do we know (or at least have a good assumption) if the actual animations are directly relevant? Is one solider attacking another soldiers the real mechanic or ar the real mechanics only unit based?

[vartan]

Still I'm not sure about the validity of the data itself. If I have the time, I'll try to re-model the system with some hypothesis and see if it fits with the inductional "models" there. Do we know (or at least have a good assumption) if the actual animations are directly relevant? Is one solider attacking another soldiers the real mechanic or ar the real mechanics only unit based?

That's what I've been wondering, too. I want to say that the visuals don't make any difference, and that the two models have only to collide at one spot in their collision detection overlays (very simple, probably elongated spheres) in order for combat calculations to start running. So it's not that the combat is pure numbers, or entirely based on animations, but rather depends on both. The animation itself doesn't contribute, it's the position of the model relative to the enemy model. I wish CA made it so that it was more physical, that the animation represented something real. That it would hit the neck, or the torso, or the legs, with varying chances let's say. But that's more work (I should know!)

[Brave Brave Sir Robin]

Please go discuss your maths elsewhere. The numbers have me confused