Stats and stuffs

Users who are viewing this thread

Hexnibbler

Recruit
Just a thread for me to post random plots of stats I do when getting curious about battle mechanism and effectiveness.

some here about hit localisation : https://forums.taleworlds.com/index.php?threads/sturgias-unplayable.447415/page-3#post-9755798
(not mentionned : right side are hit 55% of the time, left side 45% on the imperial units with shield because the shield blocks on the left.)
others about AI targetting in melee here : https://forums.taleworlds.com/index.php?threads/ai-battle-agent.447477/

and here some about javelins :
BEziUVB.png


Sturgia bandit vs imperial recruits (200v200)
comparing the effect of ordering the javelins to "charge" vs letting them wait
2 upper plots = charging
2 lower plots = waiting

1) In "charge" it takes 5seconds before melee, in wait it takes 15 seconds (ticks are 16ms)
2) in wait, 300 ticks allow for 46 hits, in charge 300 ticks allow for 32 (50% more in wait)
3) the slope of hit landed decreases after 300 in "wait" (see before & after point 300,46). Does distance positively impact accuracy ? (relative speed of target lower ?) friendly blocking line of sight ? [edit] It's the results of recruits arrangement going from "line" to "loose".
4) 50% more hit when standing still (46/300 vs 32/300)
 
Last edited:
battania Raiders (throwing axes)

9c0dgV4.png


same idea.
note : the "tick" only starts counting at the first hit.

1) when standing, it takes 845 ticks before melee (which is 50 less than javelins, but still is 13.5 seconds). However when charging it only takes 115 (1.84 seconds !). Does the moving penalty affects throwing axes more ? (making more misses to the point where it doesn't hit anyone at the first volley)
2) 50% of the hits occured after melee started when charging. (in 1 second.)
3) While standing, the TA hits less than the javelins. Empiric observation suggests that the throwing axes trajectory have trouble adjusting to the moving enemy : axes land behind a running enemy (flying above) and ahead of "shield wall" formation.
 
Awesome stats, I'd be interested in see the differences between formation types for jav. My guess is loose formation will net you the best results, but it be great to see it statistical proven.

Also a little advice, comparing the graphs is easier visually if the axis scales are the same. It will of course make it less detailed but right now they look very similar until you take into account the scales.
 
In general if find AI is only at it's best when stationary using ranged weapons against something directly in front of it. Any attempt at using formations and creative tactics has about 1/3 chance of actually making more troops lost because the troops just fail to move how you wished or perform thier combat role. Some things like flanking or spanking (flanking the back?) with HA are good, but I notice that the HA actually aren't doing as well when in a different orientation (less shots landing, more idle troops)and if they need to be moved again often their response is not as clean.
 
Awesome stats, I'd be interested in see the differences between formation types for jav. My guess is loose formation will net you the best results, but it be great to see it statistical proven.

Also a little advice, comparing the graphs is easier visually if the axis scales are the same. It will of course make it less detailed but right now they look very similar until you take into account the scales.
I tried to use loose formation hoping that there would be more hit, but I haven't found evidence of it. There is the issue of range (people on the flank don't throw) and if forming a "loose square", it doesn't increase either. You are right for the axis but to be frank I should write tools to get better control over the plots instead of using internet free stuff :p

would be nice to see some tests regarding throwing weapon accuracy at different ranges.
I did some without generating plots to get a "feeling" of hit, trying to hold fire from 70 to 30 meters by incrementing 10 meters at a time. And the casualties of my troops only went worse if delaying the fire so I havent't plotted the numbers.

In general if find AI is only at it's best when stationary using ranged weapons against something directly in front of it. Any attempt at using formations and creative tactics has about 1/3 chance of actually making more troops lost because the troops just fail to move how you wished or perform thier combat role. Some things like flanking or spanking (flanking the back?) with HA are good, but I notice that the HA actually aren't doing as well when in a different orientation (less shots landing, more idle troops)and if they need to be moved again often their response is not as clean.

for HA It may be because they have to have their targets on their right or in front. I.E. same reason they ride CCW and not CW if I'm not mistakening (may be the opposite). infantry ranged can turn themselves, but I think horse archers won't turn their horses when standing still.

--
aside from positioning any kind of ranged in a way that disable the effect of enemy shield, positioning and tactics are relatively irrelevant. Horses are not good enough against ranged (I should check the difference of crossbow & bow effective rate of fire ), are very bad versus infantry, infantry die way too quickly in infantry melee and ranged quickly become "machine guns" when upgraded to higher tiers.

I like the fact that the AI individual have their own priority and don't follow orders blindly but there is some work to be done there and as well as on the front of orders/formations and tactics that isn't strictly speaking a problem of "unit balance" (there also are issues there, but it doesn't make sense to balance stuff in a given AI context if that context requires adjustments)
 
Did some stats to get an idea of how many hits are blocked by friendlies.

Empire T2 : 1401 hits, 131 Friendly block => 131/1401 = 9.35%
Sturgia T2 : 1185 hits, 112 Friendly block => 9.45%
Legionnaries : 1492 hits, 76 friendly block => 5.09%
Linebreakers : 1131 hits, 28 friendly block => 2.47%
Vlandia Sergeant : 1650 hits, 96 friendly block => 5.8%
Sturgia heavy spearman : 2593 hits, 292 friendly block => 11.26%
Battania picked warrior : 1853 hits, 167 friendly block => 9%
Battania volunteers : 938 hits, 9 friendly block => 0,95% (less than 1%). Note : they have either 2H or a short 1H.
.
Hypothesis :
with short 1H weapons you reduce the risks of losing hits because of friendlies proximity (see legionaries, sergeants battania volunteers)
with 2H weapon the fighters can start to attack while being further from their target, reducing the risk of being close to a friendly. (see linebreakers, battania volunteer)
with medium/long 1H weapons you have the max risks
 
just to highlight a stat mentioned in the sturgya thread : 44.4% unblocked hits are on the left side vs 55.5% hitting right because shield is on the left hand for those empire dude. [edit to clarify : this is not due to "blocked" by friendlies stats. It comes from stats of overall hits, counting how many hits lands on right should/right arm versus left shoulder/left arm. ]
 
Last edited:
Back
Top Bottom