Oleander Ardens,
I don't know under which conditions the ai skips volleys. I only recall seeing it when the ai is shooting at a ranged unit, and I did see it happen even at short range.
I did save the original html thread in which LongJohn commented on accuracy increase with valor. The thread is no longer available here at the org since it's from Oct 2002. Although it's long, I'm pasting the text of entire thread here since it goes into detail about the testing of the ranged units and analysis of the data. Just for the record, I did have fatigue enabled in my tests, but archers do use all their arrows without incurring much fatigue. Of course if you test slow firing weapons, fatigue does become a big factor in kills/volley. It's interesting that Kraxis noticed the ai archers not firing on every volley even back then. Keep in mind that the projectile.txt file was not available at this time since we were still using mtw v1.0, and that file wasn't made available until mtw v1.1 which does come in at the end of the thread in Nov 2002. Later on, a theory that the uneven spacing of Treb archers was postulated as the reason they did a little better in some of these tests. Most units do suffer loss of accuracy in archers who are in the 3rd row or further back when in close formation, but not if in loose formation. I don't remember at what row loss of accuracy starts when in loose formation. I didn't edit any of the text, but did remove some spaces and the words "IP logged" to make it a little easier to read. LongJohn's is the 5th post, and the only one he made to this thread.
--------------------------------------------------------------------------------
baz
Patron
Posts: 136
From:essex
Registered: Aug 2002
posted 10-29-2002 07:23 PM
--------------------------------------------------------------------------------
Can somebody please inform me of the following, or let me know where i can find the following.
1.The cost of upgrades : what percentage it costs to add valour to a unit, and also on top of valour each weapon and armour values.
2.The increase in unit stats : what unit stats are changed by which upgrades and by how much.
------------------
previously BarryNoDachi - KenchiBND
Clan Kenchikuka
Puzz3D
Clan Takiyama
Posts: 1362
From:Boston, MA, USA
Registered: Oct 2000
posted 10-29-2002 08:40 PM
--------------------------------------------------------------------------------
In MTW currently:
Valor 50%, attack +1, defend +1, morale +2
Weapn 30%, attack +1
Armor 35%, defend +1, armor +1
According to LongJohn after the patch:
Valor will go to 70% except for ranged units. Ranged units will have the component of their cost relating to the ranged weapon removed from the cost of upgrades since upgrades primarily improve hth combat.
Note: The cost of upgrades is cumulative and is a percentage of the unit's current value which you can see by holding the cursor over the unit during army selection. The tax on more than 4 of a single unit type is 20% of the base unit cost. It's cumulative for additional units, but is not compounded.
Kraxis
Patron
Posts: 1146
From:Lyngby, Denmark
Registered: Feb 2002
posted 10-30-2002 04:24 PM
--------------------------------------------------------------------------------
quote:
--------------------------------------------------------------------------------
Originally posted by Puzz3D:
According to LongJohn after the patch:
Valor will go to 70% except for ranged units. Ranged units will have the component of their cost relating to the ranged weapon removed from the cost of upgrades since upgrades primarily improve hth combat.
--------------------------------------------------------------------------------
Aha.... So he just informally confirmed that Valour does not increase accuracy?
I wonder why the old STW manual said it did (at least I think it did)?
------------------
BTW, Danish Crusades are true to history.
You may not care about war, but war cares about you
Perec_Dojo
Patron
Posts: 41
From:New York, NY, USA
Registered: Aug 2001
posted 10-30-2002 05:33 PM
--------------------------------------------------------------------------------
STW honor upgrades increased accuracy for missile units, but the upgrading did not work properly. Under testing it was found that an appreciable increase in kill rates did not occur until honor 9 (). Why this issue has never been addressed in MTW mystifies me, but likely we are up against the same thing here.
longjohn2
Programmer
Posts: 1060
From:UK
Registered: Nov 2000
posted 10-30-2002 07:38 PM
--------------------------------------------------------------------------------
Valour does increase accuracy, but not by much. The accuracy was increased in MI rendering the valour upgrade relatively minor.
Never quite found time to address it in MTW
Kraxis
Patron
Posts: 1146
From:Lyngby, Denmark
Registered: Feb 2002
posted 10-30-2002 11:01 PM
--------------------------------------------------------------------------------
quote:
--------------------------------------------------------------------------------
Originally posted by longjohn2:
Valour does increase accuracy, but not by much. The accuracy was increased in MI rendering the valour upgrade relatively minor.
Never quite found time to address it in MTW
--------------------------------------------------------------------------------
Ahhh... so we are still at the MI accuracy table? Meaning the Valour would not be worth upgrading for MP if I want better accuracy???
Ciould you perhaps say how much it is?
------------------
BTW, Danish Crusades are true to history.
You may not care about war, but war cares about you
tootee
Patron
Posts: 1228
From:Singapore
Registered: May 2001
posted 10-31-2002 12:59 AM
--------------------------------------------------------------------------------
Even in STW-MI, upgrade improve the archers little in accuracy, and only the 1st few upgrades for musk show observable improvement.
------------------
tootee the goldfish,
headmaster of Shogun-Academy
loyal roach of Clan S.G.
'Pa Si Buay Chao Si Liao Ka Song'
------------------
baz
Patron
Posts: 136
From:essex
Registered: Aug 2002
posted 10-31-2002 05:53 AM
--------------------------------------------------------------------------------
I guess that why LJ hasn't increased the cost of ranged unit upgrades
Thnx Puzz i was pretty sure but just wanted some confirmation honest am i right in that the weapon and armour upgrades are based on the cost of a V0 unit?
Can someone answer my next question of how these stats are used to calculate a units effectiveness in battle?
------------------
previously BarryNoDachi - KenchiBND
Clan Kenchikuka
baz
Patron
Posts: 136
From:essex
Registered: Aug 2002
posted 10-31-2002 06:18 AM
--------------------------------------------------------------------------------
also, are these stats available from a file within the game?
------------------
previously BarryNoDachi - KenchiBND
Clan Kenchikuka
Kraxis
Patron
Posts: 1146
From:Lyngby, Denmark
Registered: Feb 2002
posted 10-31-2002 12:23 PM
--------------------------------------------------------------------------------
This makes me think...
What about hybrids? Such as Ottomans, Nizaris, Futtuwa, Bulgarian Brigands and most importantly Jannisary Infantry... Or what about the good melee archers such as Trebizond Archers and Longbows?
------------------
BTW, Danish Crusades are true to history.
You may not care about war, but war cares about you
CeeBod
Patron
Posts: 22
From:Manchester, UK
Registered: Sep 2002
posted 10-31-2002 12:36 PM
--------------------------------------------------------------------------------
If valour increases don't actually improve accuracy, then what exactly is the point of training Welsh Longbowmen?
Puzz3D
Clan Takiyama
Posts: 1362
From:Boston, MA, USA
Registered: Oct 2000
posted 10-31-2002 02:31 PM
--------------------------------------------------------------------------------
Kraxis,
I guess the hybrids could become among the best units you can buy in MP depending upon how many upgrades you can slap on them. The shifting sands of the MP upgrade system are still with us. I thought you didn't play MP?
BTW, I tested v0 and v4 archers and I don't see any difference in the effectiveness of the ranged weapon. I'm not saying there is no difference because I only ran a few tests. There is so much statistical noise in the test results that I would probably have to do 100 runs on each to see the difference. In practical terms, I make my ranged unit upgrade decisions as though there is no difference.
Baz,
df = attack - defend + bonus
chance to kill = 1.9% * 1.2 ^ df
df is the difference factor of the striking man's attack value minus the defend value of the man he is striking at plus any situational combat bonus. The unit parameters are in the file crusaders_unit_prod11.txt, but the upgrade costs and, I believe, almost all the bonuses are inside the main exe and not accessable.
Kraxis
Patron
Posts: 1146
From:Lyngby, Denmark
Registered: Feb 2002
posted 10-31-2002 02:53 PM
--------------------------------------------------------------------------------
quote:
--------------------------------------------------------------------------------
Originally posted by Puzz3D:
Kraxis,
I guess the hybrids could become among the best units you can buy in MP depending upon how many upgrades you can slap on them. The shifting sands of the MP upgrade system are still with us. I thought you didn't play MP?
BTW, I tested v0 and v4 archers and I don't see any difference in the effectiveness of the ranged weapon. I'm not saying there is no difference because I only ran a few tests. There is so much statistical noise in the test results that I would probably have to do 100 runs on each to see the difference. In practical terms, I make my ranged unit upgrade decisions as though there is no difference.
--------------------------------------------------------------------------------
Hehe... Well, I'm not a vivid MP'er, but I certainly play it. Mostly against/with Shadewolf and the others from Shades and the OOOO (which I'm part of).
I too believe they will be very powerful, at least some of them. Jannisary Inf do need it, and this will perhaps make archers better overall. Trebz. will certainly benefit a lot from this.
Actually back when I tried to test Jan. Inf and Longbows vs. Pikemen and SAP, I noticed an interesting thing.
The Jannisaries didn't stay put and fired until the quivers were empty against the Pikemen, so I tried to find another ranged unit that was equally efficient against the SAP as the Jannisaries.
I found that Jannisary Archers were indeed better archers than Jan Inf and more interestingly better than normal archers. While the JI averaged at 24 kills, the JA averaged at 32 kills and archers at 30 kills.
So hybrids are not archers that can fight, but rather infantry with bows. I did not test Ottomans or the other hybrids.
------------------
BTW, Danish Crusades are true to history.
You may not care about war, but war cares about you
tootee
Patron
Posts: 1228
From:Singapore
Registered: May 2001
posted 10-31-2002 03:37 PM
--------------------------------------------------------------------------------
quote:
--------------------------------------------------------------------------------
Originally posted by Puzz3D:
df = attack - defend + bonus
chance to kill = 1.9% * 1.2 ^ df
--------------------------------------------------------------------------------
And LJ once said for STW-MI df cap at [-20, 20], which give chance2kill range of [0.05%, 72.8%].
------------------
tootee the goldfish,
headmaster of Shogun-Academy
loyal roach of Clan S.G.
'Pa Si Buay Chao Si Liao Ka Song'
------------------
Puzz3D
Clan Takiyama
Posts: 1362
From:Boston, MA, USA
Registered: Oct 2000
posted 10-31-2002 03:43 PM
--------------------------------------------------------------------------------
I also see a difference between v0 Janissary infantry and v0 Janissary archers in my standard test on Pikemen. I made two runs on each, and strangely got exactly the same result each time. I got 72 kills with the JanInf and 83 kills with the JanArc. I can't account for the difference because they both use the SBOW. I'm not sure the difference is statistically significant because I only made two runs on each. I observed anywhere from 0 to 7 kills in a single volley of 28 arrows from either unit which is a lot of variability.
I think we'd all like to see ranged weapon effectiveness improve more with upgrades rather than this reduced cost approach, but it's better than nothing being done to help ranged units in MP.
baz
Patron
Posts: 136
From:essex
Registered: Aug 2002
posted 10-31-2002 06:05 PM
--------------------------------------------------------------------------------
ok i have been looking the stats availablr at the TWA and found some questions i would like answered if you can?
i took this screen shot from the TWA
http://terazawa.totalwar.org/iB_html/uploads/FMAA stats.jpg
and this screenshot from MTW F1 screen
http://terazawa.totalwar.org/iB_html/uploads/FMAA F1stats.jpg
EDIT: i cant get the pics to work
i was going to show that the stats were diff.
In that the TWA uses Melee (is this the same as attack?) and also the value for armour and the value for Def seem to be conflicting
------------------
previously BarryNoDachi - KenchiBND
Clan Kenchikuka
[This message has been edited by baz (edited 10-31-2002).]
Puzz3D
Clan Takiyama
Posts: 1362
From:Boston, MA, USA
Registered: Oct 2000
posted 10-31-2002 06:28 PM
--------------------------------------------------------------------------------
The unit has a large shield which adds +2 to armor and defend. The F1 screen includes this +2 as indicated by the $. A small shield provides a smaller +1 bonus, and cav only gets a +1 bonus from any shield they may have. Interestingly, the Chiv MAA has a 0.5 modifier on its shiled and thus only gets a +1 bonus.
The melee value is the attack value. The column labelled "morale" is unused. The column labelled "honor" is morale.
[This message has been edited by Puzz3D (edited 10-31-2002).]
baz
Patron
Posts: 136
From:essex
Registered: Aug 2002
posted 10-31-2002 06:38 PM
--------------------------------------------------------------------------------
crusaders_unit_prod11.txt
From this file i have seen where the stats at the TWA have come from.
However, i am a little confused where exactly the values on the F1 screen come from, which also leads me to wonder where exactly the values used in this formula
df = attack - defend + bonus
chance to kill = 1.9% * 1.2 ^ df
come from?
for example, what do i use in this formaula for a FMAA V0 W0 A0 and where do the numbers come from?
------------------
previously BarryNoDachi - KenchiBND
Clan Kenchikuka
baz
Patron
Posts: 136
From:essex
Registered: Aug 2002
posted 10-31-2002 06:51 PM
--------------------------------------------------------------------------------
thnx puzz, reason i asked is that ive made a spreadsheet with all the stats in it (just numbers) and was going to make it so i can input the V,W and A of one of the units and it would work out how cost effective the unit would be...eventually making a comparison tool in java to compare units with diff valour...seems a little more complex than i first thought
------------------
previously BarryNoDachi - KenchiBND
Clan Kenchikuka
NinjaKilla
STW Fundamentalist
Posts: 1651
From:London, England
Registered: May 2001
posted 10-31-2002 07:19 PM
--------------------------------------------------------------------------------
I've been saying for a while now how hardcore Kenchi are
------------------
Clan Kenchikuka
Puzz3D
Clan Takiyama
Posts: 1362
From:Boston, MA, USA
Registered: Oct 2000
posted 10-31-2002 08:01 PM
--------------------------------------------------------------------------------
The formula relates to a pair of units. The attack value comes from one unit and the defend value from the other. The formula is computed twice during each combat cycle so that each man gets in his strike against the other man.
On flat ground, head-to-head fighting you have charge, attack, defend, shield, armor piercing, anti-cav, rank bonuses and varying unit sizes to consider. If you add a unit's attack + defend + shield, that gives the basic combat power in hth for each man in that unit. You could scale that by the number of men in the unit, although, that neglects the ability of men in a larger unit to get double attacks on men in a smaller unit. Make a rough assumption that charge cancels or assume that 4 points of charge equals 1 point of attack. Add in the rank bonus which operates against all units, and the anti-cav and armor piercing bonuses if appropriate for that unit.
When you're all done, you still don't know how the unit will work in a battle because morale, rate of fatigue and speed have not been accounted for.
Kraxis
Patron
Posts: 1146
From:Lyngby, Denmark
Registered: Feb 2002
posted 10-31-2002 09:37 PM
--------------------------------------------------------------------------------
quote:
--------------------------------------------------------------------------------
Originally posted by Puzz3D:
I made two runs on each, and strangely got exactly the same result each time. I got 72 kills with the JanInf and 83 kills with the JanArc.
--------------------------------------------------------------------------------
How did you manage to get the JI to use all their arrows? Did you give the Pikemen V4?
I suspect that Archers have better accuracy and/or power than hybrids such as the JI. Good archers such as Trebz. and JA apparently have even better accuracy and/or power, for they seem to kill slightly more than normal Archers.
It is like the Mounted Archers from STW, they had lower accuracy than Samurai Archers.
------------------
BTW, Danish Crusades are true to history.
You may not care about war, but war cares about you
Puzz3D
Clan Takiyama
Posts: 1362
From:Boston, MA, USA
Registered: Oct 2000
posted 10-31-2002 11:23 PM
--------------------------------------------------------------------------------
Kraxis,
I did make the pikemen v4. It appeared that the JanInf used all their arrows.
Kraxis
Patron
Posts: 1146
From:Lyngby, Denmark
Registered: Feb 2002
posted 11-01-2002 02:17 PM
--------------------------------------------------------------------------------
Ok... I got tired of all these speculations and half tests.
So I went and made a large test of ranged units, ten tests on each.
I used Agincourt, summer, hard with me as the attacker and Pikemen. I moved to when I saw the ranged unit changed from standing to reloading. No rain was permitted at all (making it a long test).
First subject was the Jannisary Infantry, they averaged at: 65.6 kills
Jannisary Archers: 83.2 kills
Archers: 81 kills
Trebizond Archers: 80.8 kills (but their range seemed a little longer than the Archers, perhaps about 500)
Bulgarian Brigands: 81.8 kills
Ottoman: 75.4 kills (their range seemed a little shorter than normal Archers)
Nizari: 72.6 kills (but they did not use all their ammo as they closed for melee at around 25 volleys)
Four interesting things I noticed was:
Firstly it does seem like some units have better ranges than others, but it is very slight.
Secondly Nizari makes a little dance when firing at times. They fire, reload, aim, takes down the bows to reload again, aim, fires... Strange...
Thirdly, four rows of archers seem to be of little influence in the tests. Bulgarian Brigands in four rows defeated both Trebz and Archers in three rows, though very slightly (basically there was no difference).
Fourthly, climate seems to have a say in these matters as well.
Tired of all the rain in Lush, I attempted to test the Archers in Arid. The result was an average of no less than 87.4, where 6 tests were above 90, in Lush not a single test crossed 90.
------------------
BTW, Danish Crusades are true to history.
You may not care about war, but war cares about you
baz
Patron
Posts: 136
From:essex
Registered: Aug 2002
posted 11-01-2002 02:42 PM
--------------------------------------------------------------------------------
With the rows that they were in, i dont think that matters for archers, because they can fire from behind each other. Where as guns can only fire at front so if they are in ros of four then they will shoot less volleys per time (this is how it was with MI anyway) i think thats right but puzz will knwo for sure
Try shooting with archer against a unit in rows of one, this was interesting in MI.
I beleive crusaders_unit_prod11.txt contains all the ranges in it so have a look about the ranges
The whether gave some interesting results, this may help us in the MP game to have some ranged fighting
Very nice results though
------------------
previously BarryNoDachi - KenchiBND
Clan Kenchikuka
Puzz3D
Clan Takiyama
Posts: 1362
From:Boston, MA, USA
Registered: Oct 2000
posted 11-01-2002 05:27 PM
--------------------------------------------------------------------------------
Kraxis,
Nice test. There might be something to what Baz says about a single line being harder to hit which could affect the results. However, neglecting that for the moment, if you assume each kill is independent of preceeding kills and the random number generator is producing a uniform distribution of numbers, you can say that the statistical error is the square root of the total number of events. There will only be a 5% chance that a result will fall outside twice that error.
Since you have four of the ranged units, Archer, Treb Archers, Jan Archers and Bulgarian Brigands, all falling in a nice cluster, they are probably all the same. They provide a nice solid average of 81.7 with an error of + or - 1.76%. ((square root of 3200) / 3200). That agrees with all the tests I've made of Archers on Pikemen which tended to average out in the low 80's.
It looks like the Ottoman and Nazari fall into another statistically significant group since (square root of 740)/740 is 3.68% which would give a 95% confidence range (+ or - twice the error) of + or - 5.4. That's not enough to drive the Ottoman average of 75.4 up to 81.7. Of course, the 81.7 average has an error of 1.76%. So, it's 95% confidence range is + or - 2.87. Since 81.7 - 2.87 = 78.8, and 75.4 + 5.4 = 80.8 the test isn't quite good enough to show that there is a difference between Archers and Ottoman Infantry with 95% confidence. The test does show a difference to better than a 67% confidence level which is just + or - the error.
Jan Inf is a different story. It's average of only 65.6 kills is well below any possibility of this unit having the same ranged effectiveness of the group which includes the Archers. There is a slight (about 10%) possibility that Jan Inf are in the same group as Ottoman Inf and Nazari, but probably the Jan Inf are in a group by themselves. If the Jan Inf didn't always use all their arrows then you would get an artificially low result. Testing on a LAN or online would confirm the result.
If this analysis is completely wrong, then I claim temporary insanity.
Kraxis
Patron
Posts: 1146
From:Lyngby, Denmark
Registered: Feb 2002
posted 11-02-2002 11:20 AM
--------------------------------------------------------------------------------
No, I'm pretty sure the JI use up all their arrows... It is the Nizaris that stop short of the full ammo, so they seem to fit into the true archer group. And yes I thought pretty much that the Archers, BB, Trebz and JA were the same in ranged when I had done my tests.
GAH Could the patch get already, then we could know for sure.
I still think it is very interesting that the Archers in Arid averaged so high compared to their Lush results.
------------------
BTW, Danish Crusades are true to history.
You may not care about war, but war cares about you
Kraxis
Patron
Posts: 1146
From:Lyngby, Denmark
Registered: Feb 2002
posted 11-02-2002 01:25 PM
--------------------------------------------------------------------------------
Ok... did another set of tests with mounted archers... Same conditions.
Turcoman Horse: 58.8 kills
Horse Archers: 60 kills
Byzantine Cavalry: 64.6 kills
Mamluke HA: 63.8 kills
Boyars: 57.4 kills (they had a fluke of 45 kills in one test, I guess they would average at 60).
Then I did a test I thought would be great. I have long considered Turcoman Foot the ultimate foot-archers, fast, armoured, with shield and not as bad as normal Archers in a fight.
The test resulted in: 84.4 kills... Making them a good deal better than most archers, because I had a hard time stopping them from advancing after having used 25 volleys. When they did use all arrows, they killed respectively 93 and 91.
So now it is confirmed, these guys are the archers in the game.
Noted a strange weatherforecast. "Clear and getting better later". How can that be?
------------------
BTW, Danish Crusades are true to history.
You may not care about war, but war cares about you
[This message has been edited by Kraxis (edited 11-02-2002).]
Puzz3D
Clan Takiyama
Posts: 1362
From:Boston, MA, USA
Registered: Oct 2000
posted 11-02-2002 02:13 PM
--------------------------------------------------------------------------------
Kraxis,
I ran a quick test of Turcoman Inf on Pikemen in Temperate and Arid conditions, and I don't see any difference. I use a perfectly flat map and always select "fine throughout the day" for weather. All the arrows were used. I got 84 and 87 kills in Temperate, and 82 and 87 kills in Arid.
Kraxis
Patron
Posts: 1146
From:Lyngby, Denmark
Registered: Feb 2002
posted 11-02-2002 02:19 PM
--------------------------------------------------------------------------------
Odd... Maybe those Archers I ran through Arid were just very lucky.
But even your tests confirms that Turcomans are great archers, averaging above most others, and even if not they have other properties that makes them better.
------------------
BTW, Danish Crusades are true to history.
You may not care about war, but war cares about you
tootee
Patron
Posts: 1228
From:Singapore
Registered: May 2001
posted 11-03-2002 01:11 PM
--------------------------------------------------------------------------------
Single row unit (not just missiles) are harder to get hit in MTW as well, and this phenomenon is very obvious. For 4row vs 2row or vs 1row, the difference in casualty rate is quite big.
------------------
tootee the goldfish,
headmaster of Shogun-Academy
loyal roach of Clan S.G.
'Pa Si Buay Chao Si Liao Ka Song'
------------------
[This message has been edited by tootee (edited 11-03-2002).]
Dorkus
Patron
Posts: 110
From:
Registered: Oct 2002
posted 11-03-2002 03:45 PM
--------------------------------------------------------------------------------
This is good info. Thanks guys.
hrvojej
Patron
Posts: 418
From:
Registered: Sep 2002
posted 11-03-2002 07:30 PM
--------------------------------------------------------------------------------
Puzz and Kraxis,
I don't mean this as a critique, but just out of (professional) curiousity, do you guys actually do T-tests etc. on the test data, or are you just looking at the means? That is, are you using the statistical significance to distinguish between the unit effectiveness? For example, a lot of things may not show up on the stats tests because of the small effect size, or small sample size (the smaller the effect, the larger sample size you need to observe it, the power is probably lower as well as a result, plus you might have to fiddle with alpha errors, etc.).
[This message has been edited by hrvojej (edited 11-03-2002).]
Kraxis
Patron
Posts: 1146
From:Lyngby, Denmark
Registered: Feb 2002
posted 11-03-2002 08:37 PM
--------------------------------------------------------------------------------
Of course there can be errors as my test of Archers in Arid showed (or rather Puzz's tests did).
Since archery is less effective against very small units the further the test progresses the greater the fault you can expect. So there might actually be a difference in the effectiveness in the 'archer' group, but we simply can't find it at these times.
Besides actually doing tests like the ones we have made I can't really see how we can gain any knowledge of the various ranged units.
Perhaps using Large unitsize would be better, I don't know.
What I do know is that once the patch arrives (in March 2004) we will be able to see if we have assumed correctly.
------------------
BTW, Danish Crusades are true to history.
You may not care about war, but war cares about you
hrvojej
Patron
Posts: 418
From:
Registered: Sep 2002
posted 11-03-2002 08:53 PM
--------------------------------------------------------------------------------
Kraxis,
I wasn't saying that the test themselves are not good, I was just asking about the data analysis. I use a lot of statistical data analysis in my, hm, profession to be, and was just curious about how you do your analysis. I meant no offense, really.
[This message has been edited by hrvojej (edited 11-03-2002).]
Kraxis
Patron
Posts: 1146
From:Lyngby, Denmark
Registered: Feb 2002
posted 11-04-2002 01:06 AM
--------------------------------------------------------------------------------
Well, it was a little hard understanding what you meant.
With the info one gets from such tests there is hardly anything else that can be done than make it statistical.
------------------
BTW, Danish Crusades are true to history.
You may not care about war, but war cares about you
Puzz3D
Clan Takiyama
Posts: 1362
From:Boston, MA, USA
Registered: Oct 2000
posted 11-04-2002 01:28 AM
--------------------------------------------------------------------------------
hrvojej,
I just did the t-test for 12 runs of Archers vs Pikemen with my standard custom battle set up of flat map, fine day, temperate climate, and the pikemen just inside max range.
Archers 20x3 vs Pikemen 20x5 kills:
70, 83, 87, 87, 79, 85, 88, 80, 81, 81, 72, 82
Mean = 81.25
Standard deviation = 5.6
Standard error = 1.6
t (.05) = 2.2
df = 11
0.95 Confidence Interval = 81.25 + or - 3.57
Archers 20x5 vs Pikemen 50x2 kills:
61, 70, 64, 77, 64, 55, 68, 63, 53, 64, 67, 66
Mean = 64.33
Standard deviation = 6.36
Standard error = 1.8
t (.05) = 2.2
df = 11
0.95 Confidence Interval = 64.33 + or - 4.04
So Archer kills at max range on Pikemen in 20x5 are without a doubt different from Pikemen in 50x2. Basically, you have to get about 1000 events to achieve about 4% accuracy, and that's assuming you have no systematic errors. The test on Pikemen 20x5 has a problem because the distribution of kills will not be symetrical about the mean due to the reduced kills per volley as the unit becomes thinner from taking casualties. Depending on how the trajectories are computed, another effect which may be greater when the target has very few men left is multiple hits on a single man because there are fewer targets.
hrvojej
Patron
Posts: 418
From:
Registered: Sep 2002
posted 11-04-2002 01:33 AM
--------------------------------------------------------------------------------
Kraxis,
Sorry for the misunderstanding. I'll try to clarify it further.
What I meant was, once you do the tests in the field, what do you do with data you obtain? Do you just look at the means (averages), or do you do some statistical analysis on them to see whether there is a statistically significant difference between the groups?
I'm doing quite a lot of stats for my dissertation at the moment, so I guess I'm a bit spaced out with it, and see statistical data everywhere; therefore, I began to wonder how you guys derive your conclusions, and what statistical methods and approaches can be applied to this kind of dataset.
edit: Puzz,
Yes, that's what I meant. Thanks for satisfying my curiousity.
[This message has been edited by hrvojej (edited 11-04-2002).]
Kraxis
Patron
Posts: 1146
From:Lyngby, Denmark
Registered: Feb 2002
posted 11-04-2002 11:47 PM
--------------------------------------------------------------------------------
Well actually, I have begun to remove the highest two results and the lowest two results out of the ten tests. That should remove any flukes and even the results to something we can use.
It doesn't look good if a unit kills only 50 enemies in one test but kills about 80 in the others, it will even out at far less than the others, eventhough it is actually just as good.
------------------
BTW, Danish Crusades are true to history.
You may not care about war, but war cares about you
[This message has been edited by Kraxis (edited 11-04-2002).]
Puzz3D
Clan Takiyama
Posts: 1362
From:Boston, MA, USA
Registered: Oct 2000
posted 11-05-2002 02:30 AM
--------------------------------------------------------------------------------
Tossing out the high and low runs is not going to make the answer any more accurate. You do get a clustering of most results around the mean in these tests, so you can usually tell after only a few runs if you're looking at a big effect. If you're looking for a small effect, there is no way around doing a lot of runs to see it. Ironically, small effects are much more work to pin down, but are less important in a practical sense.
[This message has been edited by Puzz3D (edited 11-05-2002).]
tootee
Patron
Posts: 1228
From:Singapore
Registered: May 2001
posted 11-05-2002 03:51 AM
--------------------------------------------------------------------------------
Yes, the big effects are the more impt one, like putting unit in single row when under missile fires.. the difference compared to a standard 4/5 rows is quite obvious that I know is always reproducible.
My only answer for this result is that the game engine computes missile casualty mainly for men near the centre of the formation (or that range units tend to fire at the core of the formation), thus having less men per column reduces casualty due to missile fire (not sure about arty).
------------------
tootee the goldfish,
headmaster of Shogun-Academy
loyal roach of Clan S.G.
'Pa Si Buay Chao Si Liao Ka Song'
------------------
[This message has been edited by tootee (edited 11-05-2002).]
youssof_Toda
Patron
Posts: 430
From:Limbabwe
Registered: Sep 2001
posted 11-05-2002 06:58 AM
--------------------------------------------------------------------------------
This is like reading a math exam
Puzz3D
Clan Takiyama
Posts: 1362
From:Boston, MA, USA
Registered: Oct 2000
posted 11-05-2002 01:30 PM
--------------------------------------------------------------------------------
I ran my Archer vs Pikemen test on Pikemen in 20x5 loose formation. The results were:
59, 66, 72, 69, 80, 63, 77, 59, 67, 71, 69, 83.
Average = 69.6 + or - 4.84 (0.95% confidence)
This is an example of a test that's good enough to show there is a difference between Pikemen in 20x5 close and 20x5 loose formation, but not good enough to show there is a difference between Pikemen in 50x2 close and 20x5 loose formation.
tootee,
It does appear that the archers shoot at the center of the formation if the whole formation is in range. While the horizontal spread of the arrow cluster narrows down to match the width of what they are shooting at, the front-to-back spread seems to have a minimum value that exceeds the depth of targets that are only a few ranks deep. The result is more misses when firing on thin targets, and more hits when firing on deep targets. It's very effective if you can flank and fire enfilade into an enemy ranged unit that's set in a 20x3 formation.
Lord Romulous
Patron
Posts: 152
From:
Registered: Sep 2002
posted 11-05-2002 02:05 PM
--------------------------------------------------------------------------------
quote:
--------------------------------------------------------------------------------
Originally posted by youssof_Toda:
This is like reading a math exam
--------------------------------------------------------------------------------
no wonder i cant understand a word they are saying
seriously though guys.. keep doing your research i will be most interested in the results.. eg what unit is better etc.
i wont understand how you obtained the results but u seem like smart people so ill take your word for it.
Puzz3D
Clan Takiyama
Posts: 1362
From:Boston, MA, USA
Registered: Oct 2000
posted 11-05-2002 02:24 PM
--------------------------------------------------------------------------------
Well, the test I've been doing is really simple. In custom battle, I make the ai defender and give it a ranged unit. I then march my unit close enough to the ai's unit to make it march toward me. It will stop when it gets its front rank in range and start its reload cycle. I march my unit forward a distance of about 3 ranks to make sure all the ai archers will fire. Then I let the ai unit shoot all of its projectiles and record the result. FastCub was good enough to repeat the archer vs pikemen test on his LAN to verify that custom battle (normal difficulty) and LAN (same as online) gave the same results.
Kraxis
Patron
Posts: 1146
From:Lyngby, Denmark
Registered: Feb 2002
posted 11-05-2002 03:03 PM
--------------------------------------------------------------------------------
Ahhh... I always thought that normal gave the player a small advantage, so I always tested on hard (AI small advantage).
From now on I will test on normal.
Basically I do as Puzz, but I run towards the AI (want to do this quick) and stop about half a second after I see the AI change to 'reloading'.
------------------
BTW, Danish Crusades are true to history.
You may not care about war, but war cares about you
baz
Patron
Posts: 136
From:essex
Registered: Aug 2002
posted 11-05-2002 03:17 PM
--------------------------------------------------------------------------------
interesting about firing at a flanked unit, does that bring a lot more casulties?
------------------
previously BarryNoDachi - KenchiBND
Clan Kenchikuka
Kraxis
Patron
Posts: 1146
From:Lyngby, Denmark
Registered: Feb 2002
posted 11-05-2002 05:43 PM
--------------------------------------------------------------------------------
quote:
--------------------------------------------------------------------------------
Originally posted by baz:
interesting about firing at a flanked unit, does that bring a lot more casulties?
--------------------------------------------------------------------------------
I think it does for Arbs and Crossbows, perhaps also for Archers but it must be much less.
But it certainly is better to throw javelins at the flank of a unit, it seems kills will be up to about 20% better.
------------------
BTW, Danish Crusades are true to history.
You may not care about war, but war cares about you
Kraxis
Patron
Posts: 1146
From:Lyngby, Denmark
Registered: Feb 2002
posted 11-05-2002 06:54 PM
--------------------------------------------------------------------------------
Having looked at the projectile file there is no indication of certain units being better than others if they have shortbows.
So in effect there is something not included in the file that makes certain that Jannisary Infantry is not as good at ranged battles as others.
Puzz could you get Cub to do a test of the Jan Inf and make certain all arrows are used?
------------------
BTW, Danish Crusades are true to history.
You may not care about war, but war cares about you
--------------------------------------------------------
Note: I missed a few unwanted spaces, and edited to take them out.
vBulletin® v3.7.1, Copyright ©2000-2025, Jelsoft Enterprises Ltd.