When ulf-ing a defenceless unit, you get 0% inflicted

[cmonyiman]

(I don't know if this has been posted before, apologies if it is the case).

It came to my mind that it would be more fair, if, when you kill a defenceless unit (an adept for example) with an ulfserker or berserker, you would recieve 0% inflicted for it. Indeed, I do not complain at all about the luck system of Wesnoth, I just say it would be more fair, as the luck system, from what I've seen , tends to get back to 0%. So if you kill an adept, how many hits missed or taken doesn't affect your "overall" luck , thus it would not be reasonable to recieve "luck" (think about +30% or -30%) for it , because you would have killed the unit and that wouldn't be nor lucky nor unlucky. It would not be fair that you hit more (or less) later on in the game because of that - or that one would complain about his inflicted or recieved damage while ulf had a part of that % too.

There is also the case of ulf not killing adept ( around 0.3 or 0.4% chance if I'm not mistaken), in that case you should recieve the - inflicted for it - for you would have been unlucky and you would deserve that - inflicted.

Please post back what you think.

Cheers !

[turin]

This just goes back to the idea of EV being done on the basis of individual combats, not strikes, which was proposed a while ago and makes a lot of sense to me.

So, in any Ulf vs. Adept combat, the EV is going to be 19 damage if the adept has 19 HP left, 25 damage if the Adept has 25 HP left, etc, and if the combat ends with 19 damage being dealt, or 25 damage being dealt, then the EV was dealt, and the stats should reflect that. It doesn't matter how that damage got dealt - if the Ulf missed 10 times while dealing the strikes needed to kill the Adept, or if he got all hits until the Adept died.

And good job - the fact that Ulf v. Adept combats would mess up the EV so much when calculating EV-per-strike, but be perfectly accurate to "real life" good/bad luck when calculating EV-per-combat, is probably the best argument I've seen so far for doing it per-combat not per-strike...

[JW]

Indeed....I am quite amazed I had not thought of such a thing myself actually. Cheers!!

[TL]

I'm... pretty sure that the RNG does not have any sort of compensatory function. The fact that damage statistics trend towards the expected value over time is just the law of large numbers at work.

That said, I still support this idea as it would give more meaningful statistics.

[Lorbi]

There is also the case of ulf not killing adept ( around 0.3 or 0.4% chance if I'm not mistaken), in that case you should recieve the - inflicted for it - for you would have been unlucky and you would deserve that - inflicted.

the adepts chance to survive is way below 0.3% ... so much that this is a thing you really do not need to worry about ..
its more like 0.0000000000000000000000000000000000000001% ( and there is still a lot of zeros to add )

[JW]

There is also the case of ulf not killing adept ( around 0.3 or 0.4% chance if I'm not mistaken), in that case you should recieve the - inflicted for it - for you would have been unlucky and you would deserve that - inflicted.

the adepts chance to survive is way below 0.3% ... so much that this is a thing you really do not need to worry about ..
its more like 0.0000000000000000000000000000000000000001% ( and there is still a lot of zeros to add )

Imagine 4-4 against 33hp on 50% def. 9 hits to kill with 120 strikes. Now it's just a matter of doing the math.

[Dave]

The chance is truly miniscule. Even if the ulfserker only has 20% chance to hit it's a very small chance of hitting less than 9/120. With 10% chance to hit the chance would be reasonably high.

However, from what I remember of probability theory, actually calculating it is kinda complicated.

David

[Cheradenine]

Oooh! An opportunity for unrestrained math geekery!

[math geekery]

The probability of getting exactly k hits out of 120 should be given by:

(120Ck) * (.4)^k * (.6)^(120-k)

where 120Ck is the number of combinations of 120 objects taken k at a time, that is, 120!/(k!(120-k)!). Essentially, 120Ck gives the number of ways the ulf can hit k times, and the (.4)^k * (.6)^(120-k) gives the probability of any one of those ways actually happening.

Adding up the probabilities for 0 to 8 hits, I get a probability of about 8.75 X 10^-17 (or .0000000000000000875) of the adept surviving.

[/math geekery]

For context, this means that if the ulf attacked one adept every second, it would take on average more than 362 million years before an adept survived. So yeah, not very likely.

As Dave says, at 10% cth, it becomes a reasonable probability -- about 14%. So all those dark adepts with 90% defense can breathe easier.

[ilor]

Bah, if you want to really show off, use the Poisson distribution to approximate, it's quicker this way (and maybe implement it in wesnoth so calculating probablilities for a long berserk fight doesn't take long)

[cmonyiman]

Indeed, I see ulf "always" kills adept, though my 0.3% was taken from Wesnoth itself (99.97% to kill, .3% to leave 3hp -adept at 40%), and it seemed strange to me too though I did not calculate exact chance of killing.

So if I understand well, making an exception in RNG for ulf v adept can't be programmed, pity because these fights are rather often played in Knalgans v Undead matches.

As for calculating per-combat, I haven't read the threads, but I don't see why not, it would as well give more meaningful statistics. For, in my opinion the player would care more about exact "luck" in the game than RNG, or combat per strike (which sometimes differ from "luck" - think about killing a low hp unit).

[uzytkownik]

Am I missing something or ulf had berserk and adept no attacs?
Therefore chances of death for ulf equals 0%.

Let's assume that ulf needs k strikes to kill and chances of hitting are p. Therefore chances of killing the adept in n strikes are (from binomial distribution):

Code: Select all

\sum_{i=k}^{n} \binom{n}{i} p^{i}(1-p)^{n-i}

However for simplisity of calculation let's assume that the ulf need to strike k times which are placed as one event (i.e. kn+1, kn + 2, ... k(n+1) hits are sucessfull). The chances for killing the adept in kn tries are:

Code: Select all

\sum_{i=1}^{kn} \binom{n}{i} p^{ki}(1-p^{k})^{kn-i}

Chances for adept to survive:

Code: Select all

\binom{n}{0}p^{k0}(1-p^{k})^{kn-0} = (1-p^{k})^{kn}

which tends to 0 as n tends to infinty (it does as chances for ulf to survive are 100%) provided that p is non-0 (it tends to 1 - or rather is equivalent to 1 if p equals 0).

However - we made the additional cause for ulf to kill adept. Therfore the chances for ulf to kill are greater and chances for adept to survive are smaller (chances of all events has to be equal 1).

Therefore the chances for adept are smaller or equal 0. But the chances has to be in rance [0, 1]. Therefore the chances for adept to survive are 0 (unless p = 0 - in this case they are equal 1).

Of course the wesnoth engine have to take into account the case where chances for ulf to survive are not = 1. Or there is finite number of strikes.

[turin]

The Ulf can't die, but berserk != "unlimited attacks", berserk = "30x attacks", which is effectively unlimited in most cases, but it is possible to miss 120 times and thus not kill the adept.

[JW]

The Ulf can't die, but berserk != "unlimited attacks", berserk = "30x attacks", which is effectively unlimited in most cases, but it is possible to miss 120 times and thus not kill the adept.

Or even 120-(#hits to kill)+1