Play Now Login Create Account
illyriad
  New Posts New Posts RSS Feed - Battle Calculation with Equipment
  FAQ FAQ  Forum Search   Register Register  Login Login

Battle Calculation with Equipment

 Post Reply Post Reply Page  <1 345
Author
Sene View Drop Down
Greenhorn
Greenhorn


Joined: 19 Oct 2016
Status: Offline
Points: 56
Post Options Post Options   Thanks (0) Thanks(0)   Quote Sene Quote  Post ReplyReply Direct Link To This Post Posted: 06 Jun 2018 at 20:58
My sincearest apologies, man, but i didn'tunderstand a word of what you said. Would you please clarify the following:

corner case scenarios
fractional troop rounding
effect of multiple divisions

Cheers in advance

Originally posted by Ten Kulch Ten Kulch wrote:

[QUOTE=Sene]
These factors disappear in any battle of meaningful size, but they will serve to muddy your results data on the micro scale..

We talk hunting, you talk PvP. I guess it goes without saying those are different matters, arn't they? While the basic algorithm, formulas and stuff are the same, those modes demand completely different approaches. Hunting is single player, while PvP is a multi-player campaign. i can't really get your point here
Back to Top
Ten Kulch View Drop Down
Postmaster
Postmaster


Joined: 20 Jan 2017
Location: Fellandire
Status: Offline
Points: 668
Post Options Post Options   Thanks (0) Thanks(0)   Quote Ten Kulch Quote  Post ReplyReply Direct Link To This Post Posted: 07 Jun 2018 at 00:01
The engine is the same, regardless of hunting or PvP.

If your intention is to demonstrate the algorithm for interactions between equipment and terrain (for example), then a proper experiment attempts to isolate those two factors in a predictable way so that they can be measured, in a manner that removes all other potential influences from the picture.

Many divisions (A or D) will result in fractional casualties, per Stormcrow and Thundercat on this forum. In a tiny hunting army of 61 troops, a single fractional casualty is a significant percent deflection in results (~1.7% per casualty). A tiny hunting army also disproportionately involves factors like commander Heroism, elite division status, etc. in a way that can completely swamp the quantities you are trying to measure. Your experiment construction also lacks a proper control, and the way that external factors dominate the equation on your tiny sample size basically invalidates the experiment.
Check out my blog, Warmongering in Illyriad for self-defense techniques, military city construction, and PvP strategies.
Back to Top
Sene View Drop Down
Greenhorn
Greenhorn


Joined: 19 Oct 2016
Status: Offline
Points: 56
Post Options Post Options   Thanks (0) Thanks(0)   Quote Sene Quote  Post ReplyReply Direct Link To This Post Posted: 07 Jun 2018 at 06:51
P.S (PreScriptum in this case:)

I'm still struggling to get your point, Ten Kulch. Since I'm not a born English speaker, I am asking you to be as 'plain words' as possible. Please imagine you are talking to your 8 year son and he is not a young einstein (I'm times older than 13 in case devs are reading).

In previous post you stated that (plz correct me if I'm wrong) proper experiment should involve:

1) both armies being of reasonable size;
2) one of them should be several times bigger than the other.

While I completely agree with 1), that 2) is totally escaping me. 

Now I'm not sure If you are familiar with that algorithm hunt. Hucbold precisely defined the basics in his 1-5 points of his last post. 

Originally posted by Ten Kulch Ten Kulch wrote:

If your intention is to demonstrate the algorithm for interactions between equipment and terrain

The intention - the HolyGrail of Illy attackers - (the way I see it) has always been to find the connection between all the factors influencing attack/defense score. Innoble in his post 6 years ago presented a clear formula that included terrain and divison bonuses - everything that had been introduced by the moment. Since then biome, pres and crafted were added on and the task was to squeeze them all in without loosing the formula's working algorithm.

In the battle outcome/attack score/def score equation the latter part is easy to calculate - well, at least when it comes to hunting. No crafted, biome, pres or divisional bonuses for the poor NPC fellows, they are only ifluenced by terain. Hence all the experiments first aimed to work the actual attack score back from the def/kill rate, and we are brothers in arms with Hucbold here. Just like him or many other guys I spent years attacking NPCs over and over again trying to extract attack score form the def one. Hundreds of email with guys like Naigoizati who is sadly abandoned few months back were exchanged discussing how to do it the proper way.

The problem there was that casualties rate is shown with integer number, whilst calculations regard it as float. The fractional part isn't shown and it has been quite a headache for it produces what I call errand range (just checked the dict and it seems i musunderstood the word, it should be just 'error range'. My apologies here).

That's the way the story goes for those who arn't too familiar with the issue. The reson I'm telling you all that is to clarify the following 2 points: the way I see it to narrow that error range one has to 1) use big armies - the bigger, the better and 2) keep the attack as close to def possible,  the kill ratio will be very close to 1 and def/kill ratio is as close to deff and hence to attack as possible. And yes, 1 division is a better case but it would mean less commanders hence less attack score hence more trrops to compensate hence much more expensive experiments troops-wise, which is a serious issue for single 'researchers' like me.

So I agree on 'big armies' point Ten Kulch but can't see the "different size". You say many divisions result in fractional casualties, and that is partially true. Partially firstly cause all troops are added up when outcome is being calculeted, secondly this factor is diminished when using big troops numbers. Now that was my main point in the discussion with Hucbold: I used 2 scenarios with 1901 troops / 100% casualties rate and 837 troops  100% - 1/837 casualtie rate. While Hubold presented 100 troops with 10 CR% and 80 troops with 9% CR.

So please let me repeat the question: why would you propose clashing armies with troops numbers differentiting greatly? Wouldn't that increase the error/deviation?

P.S. (proper 'post scriptum' now) I get a bit graphomaniac sometimes, please let me know if do. Cheers.
Back to Top
Ten Kulch View Drop Down
Postmaster
Postmaster


Joined: 20 Jan 2017
Location: Fellandire
Status: Offline
Points: 668
Post Options Post Options   Thanks (0) Thanks(0)   Quote Ten Kulch Quote  Post ReplyReply Direct Link To This Post Posted: 07 Jun 2018 at 15:19
Originally posted by Sene Sene wrote:


In previous post you stated that (plz correct me if I'm wrong) proper experiment should involve:

1) both armies being of reasonable size;
2) one of them should be several times bigger than the other.

While I completely agree with 1), that 2) is totally escaping me.

If you use a single division, and both armies are of good size, then the second point matters less. If the armies are very small (ex. 61 knights vs 400 rats) then you again create situations where a single fractional kill will significantly affect the data. That error appears to increase with Heroism, powerful commander gear, elite divisions, multiple divisions.

Regarding experiment cost, I think the best configuration is: alt account, a settlement with no walls, defenders equipped, 5000-10000 defenders, attackers not equipped, 500-1000 attackers, single divisions. One city with neutral modifiers for both armies. A second city with a large modifier for the defender, and not the attacker (ex. defending chariots vs. attacking sentinels or fangs on plains). That way you can run a control with no equipment. When you run the comparison, the defenders will recover all equipment. I would analyze both terrain gear and defense-specific gear. With enough equipped troops, you should be able to clearly see whether the equipment adds to the terrain modifier, or multiplies it.

Army size would have to be adjusted to the point where it is easy to detect (and repeat) a different result for 1.25*2.2 vs. 2.2+0.25. That's only 2.55 vs 2.45, so my 10000 vs 1000 might not be enough. Potentially it might be better to use equipment where the modifier is closer to the terrain modifier, causing the numbers to diverge more.

Innoble probably used less troops, but he was approaching a single factor in isolation, not the interaction of two factors.
Check out my blog, Warmongering in Illyriad for self-defense techniques, military city construction, and PvP strategies.
Back to Top
Sene View Drop Down
Greenhorn
Greenhorn


Joined: 19 Oct 2016
Status: Offline
Points: 56
Post Options Post Options   Thanks (0) Thanks(0)   Quote Sene Quote  Post ReplyReply Direct Link To This Post Posted: 08 Jun 2018 at 07:47
It looks like you overestimate the influence of the divisional error. In fact, there is no any as far as I'm concerned. The final casualties number will solely depend on attack/defence ratio as the algorithm doesn't compute individual divisional losses. When number of casualties has been calculated via "Total army troops deployed x Looser's score / Winner's score", formula, they are then distributed evenly between divisions. If that's is impossible because of great difference in troops numbers, last "error" casualties are ditributed between random divisions evenly. 

So even when say classic 61 elite army is deployed with 5 commanders  and 57-1-1-1-1 troops allocation, and they win the battle with 45.5% or 27,76 rounded up to 28 casaualties, the divisions will suffer 26-0-1-1-0, with 2 single lost unit divisions being chosen at random. I'm not too sure if It could be 27-1-0-0-0 or even 28-0-0-0-0 either (this one is unlikely I'd say from my experience), but in any case that will not alter the total casulaties namber and rate. That's why I would upvote you 'big armies' point to eliminate fractional error, but thumbs down for single division. Or thumbs sidewards rather: while it does make life a little easier computing-wise, it makes no impact on the outcome.

You do have a good point, Ten Kulch, in proposing 'suicidal' (for attacking side) battles, hypothetically that is the best way to take computational error away. I did use failed battle reports to get the precise attack score, but in practice the bigger issue was not finding attack itself, but rather finding calculation model that gives the same result. Since I did found it and I'm sure beyond any reasonable doubts its true, I ceased any experiments and focused on precalculating battle reports ranging from 61 to thousands of units deployed. I think my maximum was around 3,5k stalwarts and 2,5k runeriders deployed, with 1 or 0 survivors left, can't remember for sure. Not that you have to believe my words, guys.

Lucks,

S.
Back to Top
Sene View Drop Down
Greenhorn
Greenhorn


Joined: 19 Oct 2016
Status: Offline
Points: 56
Post Options Post Options   Thanks (0) Thanks(0)   Quote Sene Quote  Post ReplyReply Direct Link To This Post Posted: 08 Jun 2018 at 08:21
Originally posted by Sene Sene wrote:

Not that you have to believe my words, guys.

Ah, it must've been "... but I'm ready to provide battle reps if ncessary".

PS Checked some battle reps, seems like "26-1-1-0-0" allocation in my last post's example is the only real case. 1-1 still randomly chosen. The same goes to survivors, actually. When casualties ratio is close to 100%, some divisions will loose all guys and few left troops will be randomly awarded to the remaining ones.
Back to Top
 Post Reply Post Reply Page  <1 345
  Share Topic   

Forum Jump Forum Permissions View Drop Down

Forum Software by Web Wiz Forums® version 11.10
Copyright ©2001-2017 Web Wiz Ltd.