F3 Strong/Average/Weak - Page 2

BigFurryMonster · May 16, 2003, 13:41

I am currently doing some extensive testing on this subject.

The tests I do are:

X Warriors -vs- Y Warriors (to determine the ratios)
X Spearmen -vs- Y Warriors
X Archers -vs- Y Warriors (these two are used to compute the effect of attack and defense)
X Jag Warriors -vs- Y Warriors (to compute the effect of movement points)
X Vet Warriors -vs- Y Reg Warriors (to compute the effect of combat experience)

A test should be done with wounded units, but that is rather difficult to set up.
I already found out that non-attacking units (workers, settlers, ...) have no impact on Military Strength.

From Test 1, I get a few different results than quoted above. Maybe we used different Patches.
For example, I get:
X = 10, Y = 8, Average
X = 10, Y = 13, Weak

From Test 1, I get:

0% <= Relative Army Strength < 80% : Weak
80% <= Relative Army Strength <= 125% : Average
125% < Relative Army Strength : Stronger

The Relative Army Strength is only calculated from the number of warriors, of course.
From Tests 2 and 3, I get roughly the following result:

Value of Warrior = 1 (assumed)
Value of Spearman = 1.52
Value of Archer = 1.57

edit: update below

Army Strength = Sum (Unit Values)

I'll keep you posted..!

Jaybe · May 16, 2003, 14:41

BFM, you could test for Conscript-Elite units instead of for damaged units. Also, could you please investigate fast units to see if they have any affect?

badams52 · May 16, 2003, 23:26

go BFM! It will be great to see your results

BigFurryMonster · May 17, 2003, 08:13

More test results coming in:

From Test 3:
(update) 1 Archer = 1.60 Warrior.
From Test 4:
1 Jaguar Warrior = 1 Warrior.
==> Movement points have no effect on Unit Value.

Hypothesis:

Unit Value = 1 + 0.60 * (Attack Value - 1) + 0.52 * (Defense Value - 1)

...testing will continue...

(update) Formula correction on next page.

BigFurryMonster · May 17, 2003, 09:29

From Test 5:
1 Vet Warrior = 1.45 Warrior Updated to 1.33 in a later test!

I noticed one strange thing here, though:

6 Vet Warriors -vs- 7 Reg Warriors: Average
18 Vet Warriors -vs- 21 Reg Warriors: Strong

...so, tripling the units on both sides gives a different result..!

update: this can be accounted for by the extra Army Support Cost caused by the greater increase in Reg Warriors (see next page)

BigFurryMonster · May 17, 2003, 09:32

Next tests coming up:

...
...
...
...
...
X Hoplites -vs- Y Warriors (to test the effect of the Defense Value)
X Bowmen -vs- Y Warriors (to test the combined effect of the Attack and Defense Values)
X Vet Spearmen -vs- Y Reg Warriors (to test the combined effect of Experience and the Defense Value)

Nor Me · May 17, 2003, 11:27

I did a short test on this closer to when this thread first came up. I remember that archers were worth more than spearmen and didn't come up with enough detail.

But I ruled out cost as a factor.

nbarclay · May 17, 2003, 18:04

Quote:

Originally posted by BigFurryMonster
I noticed one strange thing here, though:

6 Vet Warriors -vs- 7 Reg Warriors: Average
18 Vet Warriors -vs- 21 Reg Warriors: Strong

...so, tripling the units on both sides gives a different result..!

I wonder if there's some kind of rounding anomaly involved.

E_T · May 17, 2003, 19:50

BFM, how many warriors did you use for the Y (what your testing against)? I would use 100 (or even 1000), as that will help you to define your formula abit more.

No matter what, good job

E_T

Panzer32 · May 17, 2003, 22:52

Quote:

Originally posted by BigFurryMonster
I noticed one strange thing here, though:

6 Vet Warriors -vs- 7 Reg Warriors: Average
18 Vet Warriors -vs- 21 Reg Warriors: Strong

...so, tripling the units on both sides gives a different result..!

Maybe the formula is much simpler than we think, and it does not take into account the ratio. Possibly it thinks "only one more warrior, thats average. Three more warriors, thats strong!!"

Of course, this would be disapointing if it were found to be the case; although we need to consider all of the possibilities.

E_T · May 17, 2003, 23:32

BFM, can you check to see how bombardment units effect this score?

E_T

Panag · May 18, 2003, 06:48

hi ,

has anyone been here to take a look , .....

>>>>>> http://www-cs-students.stanford.edu/~soren/

have a nice day

bongo · May 19, 2003, 05:38

Haven't done any tests but from experience it looks like the AI values Attack higher than Defence and fast units over slow ones. A large army of fast attackers is often enough to dicourage the AI from declaring war.

BigFurryMonster · May 19, 2003, 06:51

Quote:

Originally posted by nbarclay
I wonder if there's some kind of rounding anomaly involved.

Very probable.

This is a picture the test method I used. For Test 1:

The dots are "test points". They give either Strong, Average or Weak as a result. I drew the lines in such a way that all Strong points are in the lower-left part, the Weak points are in the upper-right part, and the Average points are in the middle.

The lines happened to correspond to 125% and 80% relative army strength, respectively.

For the other tests, assuming the 125% and 80% remain constant, I multiply one of the axes of the graphs by a number N.
This N needs to be tuned by trial-and-error to get all the test points in the correct area. The slopes of the lines change accordingly (actually, the slopes of the lines remain mathematically constant, but the scale of the graph changes).

For example, I found N = 1.60 for Archers -vs- Warriors: I have to multiply the vertical axis (representing the number of Archers) by 1.60 to get it right. This works pretty well.

I tested no further than, say, 15 -vs- 20 units. But still, it gives a pretty good boundary for the three areas.

Panag - I couldn't find info relating to this subject on the site you mention. Can you direct me to it, if it is present?

BigFurryMonster · May 19, 2003, 06:57

Quote:

Originally posted by E_T
BFM, can you check to see how bombardment units effect this score?

Included!

...
...
...
...
...
X Hoplites -vs- Y Warriors (to test the effect of the Defense Value)
X Bowmen -vs- Y Warriors (to test the combined effect of the Attack and Defense Values)
X Vet Spearmen -vs- Y Reg Warriors (to test the combined effect of Experience and the Defense Value)
X Warriors + Z Catapults -vs- Y Warriors (to test the effect of Bombardment Units)

Testing will resume on Tuesday. Please wear your safety helmet at all times when in the area.

BigFurryMonster · May 21, 2003, 14:57

From Test 6:
1 Hoplite = 2.06 Warrior.

Hypothesis adjusted to:

Unit Value = 1 + 0.60 * (Attack Value - 1) + 0.53 * (Defense Value - 1)
(with some rounding errors)

(updated) Formula correction below.

BigFurryMonster · May 21, 2003, 16:21

From Test 7:
1 Bowman = 2.08 Warrior.

So... a Bowman and a Hoplite would almost be of the same strength...!

The formula is therefore not correct.

(update) Value updated below.

DaveMcW · May 22, 2003, 03:49

Some more data points:

10-attack-Archer = 6.5 or 6.6 Warriors
50-attack-Archer = 30.5 or 30.6 Warriors

BigFurryMonster · May 22, 2003, 04:16

Applying my formula, I'd get 6.4 and 30.4 respectively.

I assume you created (A-D) 10-1 and 50-1 units.

If we assume the difference is due to rounding errors, we can confirm that the unit value increases linearly with the attack value, given that the defense value remains at 1.

DaveMcW · May 22, 2003, 04:22

10-defense-Spearman = 4.6 or 4.7 Warriors
50-defense-Spearman = 20.6 or 20.7 Warriors

This works out to the simple formula:
0.6 * Attack + 0.4 * Defense

Can you run your spearman test with more units to see if the original is accurate? Placing units in the editor lets you work with much bigger numbers than placing them in debug mode.

bongo · May 22, 2003, 04:50

Wow, you're doing some impressive research here. How do you find the time?

Will you add mobility to the research list?

BigFurryMonster · May 22, 2003, 16:08

Not so fast, DaveMcW! I found the following interesting fact:

18 Warriors -vs- 7 Immortals: Weak
18 Warriors -vs- 7 Artificially Created "4-2" Units: Average

Apparently, the unit value is dependent on a factor we do not know yet.

Oh, and:

1 Immortal = 3.23 Warrior
1 "4-2" Unit = 3.20 Warrior
1 Legionary = 3.00 Warrior
1 Numidian Mercenary = 2.51 Warrior
1 Veteran Numidian Mercenary = 3.34 Warrior

I set the "4-2"'s cost to 30 shields, to rule out that factor.

FrustratedPoet · May 23, 2003, 05:24

Quote:

Originally posted by BigFurryMonster

18 Warriors -vs- 7 Immortals: Weak
18 Warriors -vs- 7 Artificially Created "4-2" Units: Average

Apparently, the unit value is dependent on a factor we do not know yet.

Time to throw some ideas out and see what sticks....

Maybe the unit value includes such variables as "unique unit" and "resource required". A civ that has Immortals must have access to Iron (whereas a generic 4.2 unit would be missing this flag), which implies strength and the ability to produce more powerful units in the future (although I find it hard to believe that 'potential strength' would be included)

Also, because the Immortal has the ability to put the Persians in their GA they could be valued more highly. Again, the generic 4.2 unit wouldn't have this potential. Try running the test again using a Persian Civ that has already used up its GA and whether this changes things.

BigFurryMonster · May 23, 2003, 06:38

I already thought of this, and set the flags for the "4-2" unit to exactly match the flags for the Immortal:

Can trigger GA
Available to one civ only
Upgrade abilities
Resource required

(update) The CivSpecific Ability causes this effect. See below.

FrustratedPoet · May 23, 2003, 07:10

Ah, OK then. Good thinking.

BigFurryMonster · May 23, 2003, 08:33

Finally found something! I was testing under Despotism and Monarchy. Here, it actually makes a difference how many cities you have. The number of cities determines the Army Support Cost.

For example:

13 Vet Warriors -vs- 15 Warriors: Strong (3 cities for the Vet party = 2 gpt* support cost)
13 Vet Warriors -vs- 15 Warriors: Average (1 city for the Vet party = 10 gpt* support cost)

* It also supported a worker.

To rule this effect out, I should be testing under equal circumstances, with zero support cost, for instance.

(update) 1 Vet Warrior = 1.33 Warrior (matches the result found above for NumMercs)
1 Elite Warrior = 1.66 Warrior

BigFurryMonster · May 23, 2003, 17:44

Using the knowledge obtained above, I had to run most tests again.

I got the following:

1 Archer = 1.60 Warrior
1 Spearman = 1.40 Warrior
1 Bowman = 2.02 Warrior
1 Hoplite = 1.82 Warrior
1 Immortal = 3.23 Warrior
1 NumMerc = 2.42 Warrior
Veterans = 1.33 Regular
Elite = 1.66 Regular

...leading to the following formula: (updated)

Unit Strength = (Hitpoints / 3) * (0.6 * Attack Value + 0.4 * Defense Value) * (1.01 if UU and CivSpecificAbilitiesOn)

Army Strength = Sum (Unit Strengths) * (Factor depending on ArmySupportCost)

...I haven't tested that last factor.

DaveMcW · May 23, 2003, 21:56

Great detective work!

One minor detail: ArmySupportCost should be factored in to the final calculation, not the one for individual units.

BigFurryMonster · May 24, 2003, 04:03

Absolutely right. Edited above.

nbarclay · May 24, 2003, 04:14

Is it times the support cost factor, or minus the support cost factor? That, too, needs to be investigated before the formula can be considered pinned down completely.