Combat cheats

[Gil]

Greetings to all Civ fans.

Don`t know if I am also this game fan now.
I want to start really serious discussion about this game.
After I had learned all main game`s features and rules, I won the game twice on
regent. After that I won it on monarch and relized that this level is rather
easy for me. I wanted hard and enjoyable play and started to play on emperor.
I began games on emperor several times and always ended it being totally
disappointed. The reason is AI cheats really bad in combat!
No patches fix this. Now I play 1.29f.
After I had found this site (few days ago) I researched all docs`n`faqs hear.
In FAQ there is written that AI doesn`t cheat in combat and is done link to
combat calculator.
I want to explain: I know math`s basics rather well, and after I had began to
play Civ3 at spring 2003, I wrote on C prog that does the same calculations of
probabilities by Bayes` scheme as combat calculator does. (I even had plans to
expand prog to calculator of probabilities of mass battles with many units
fighting on a square at a turn and probabilities of results of even mass events
such as wars.) I`m not a dummy, gentlemen.
Already on monarch I had started to notice that AI cheats in combat but monarch
level is rather easy level for me and I could winning with such cheats. But on
emperor cheats are really bad. Last game played (and paused) now I initiate
combats (and build my defense) to have a combats with only >50% probability of
my success. I.e. I attack only if probability of success is >50% and
concentrate my defending forces to have >50%, if AI would realize to attack me.
The results are TOTALLY disappointing. AI nearly always attacks if it has
probability of success <50% and nearly always wins. Moreover it very often wins
if it has even <35%. I.e. I have near 66%! And when I attack with chance <35% I
NEVER win. NEVER!
Moreover AI consiously enters and enters combat having probability <50%, much <
50%. It can do it only if it has confidence in success or equality of chances.
I shall illustrate on example. Current situation in my game:
1. AI`s archer(3HP) attacks my fortified spearman(3HP) on grassland and wins.
AI had only 36.3%.
2. My swordsman(3HP) attacks enemy spearman(3HP) on grassland and loose.
I had 64.2%.
3. The same situation as in the point 1. Ah yes, the result... Result is the
same. AI won had 36.3%.
4. I have a spearman(3HP) and a warrior(3HP) on a square, and AI has an
archer(3HP) and an archer(1HP) on nearest grassland square. I feel I could
attack though AI has 10% bonus on grassland. I have 2HP handicap.
My warrior attacks and looses with no harm to enemy archer. Ha-ha!
I had 45.6%.
5. I have 2 spearmen(3HP) and 1 archer(3HP) on nearest (grassland) square and
attack with spearman. Yeah, spearman is killed.
I had 45.6%.
It was true episode from the game, I didn`t hide any combats in that I had won
with <50% probability, and I could continue it infinitly, but nothing changes.
All the game on emperor goes such the way.
Now I think as handicap VS barbarians, optimal city number and city`s content
people number decrease with level`s growth, so increases AI`s param that affect
probability in combat. After you calculated your and AI`s factors and applied
terrain and other modificators, you also must apply this param.

I don`t like to create maps in games and such a like. I like to play so never
used editors for games. I have known that you, gentlemen, use editor for
research purposes, for example for empirical investigations.
I going to open editor and find more complex proofs of my theory.

 

[Bamspeedy]

You are trying to say the combat is screwy because you have 5 battles not in your favor???? And these aren't even the classic 'tank vs. spearmen' variety, either! These were battles where you had only ~55-60% chance to win. 55-60% is pretty darn close to 50%. You don't think someone could flip a coin and have tails come up five times in a row?

You didn't mention rivers. Was there none, or did you not realize that it can affect odds?

Difficulty level does not affect battle odds at all. The AI may have more advanced and simply a higher number of units, but when comparing an archer vs. spear there is no difference between levels.

 

[troytheface]

In Civ as in life, Luck counts too.

 

[E-Raser]

As well as terrain bonus can count, eg. attacking a unit on a hill across a river.

 

[wilbill]

Originally posted by Gil
I going to open editor and find more complex proofs of my theory.

Please let us know what you find out.

 

[Mano3]

Those rivers and mountains really affect combat outcome. So do those big cities over size 12.

 

[Yeti]

Something with "only" a 36% chance should happen slightly more than one in three times, not never happen.

Odds are one (or a combination) of three things is going on here:

- Game mechanics (As mentioned by others, attacking across rivers and attacking units in towns make a big difference, as well as terrain, which you seem to know since you make a point of mentioning it in some cases).

- Luck (Things are random, 36% is actually a pretty decent chance of success, and it looks like you have a limited number of sample points).

- Human nature (The bad things stick out more and get remembered longer. We all remember the "How the heck did that warrior beat my knight !?" incidents, but quickly forget the "Phew, I can't believe my warrior held off two horsemen." instances.)

 

[casual_moose]

in example number 5 what was the unit you were attacking and what was the probability? because you will usually lose if you attack with a "defensive" unit such as a spearman.

 

[ybbor]

it's just luck, 5 wierd combat results don't mean a game is realistic, and just because you have a 64% chance of winninng doesn't mean you'll win, and believe us, difficulty has no effect on combat calculations

 

[CIVPhilzilla]

I have lost a Modern Armor Army to a spearman before. But I have also had a injured warrior take out 3 attacking calvary. Sure these were rare occurances, but even if that combat calc sayyou have a 99.9% chance of winning, you still have a .1% chance of losing.

 

[Catt]

In response to a similar claim made on Apolyton not long ago, I threw together a quick test scenario that you can use to your heart's content to test whether difficulty level plays a role in combat outcomes.

The attached zip file contains a test scenario that removes the terrain defensive value of grassland and places 50 warriors, for each of you and the AI, right next to each other. Each warrior (and each experience level) has 10 HPs. The scenario automatically puts "preserve random seed" OFF, and allows you to play at any difficulty level. Feel free to play at Chieftan, save the game at 4000 BC, and then "stack attack" the AI stack by using "J" and goto. Just total the HPs lost be each side, reload, and repeat until you have a what you think is a sufficient number of trials (probably 4 or 5 reloads is sufficient). Repeat the process at Sid level and compare results. Please share the outcome of the tests!

Knock yourself out!

 

[mrmitchell]

The difficulty level of a game does not affect combat results. You're just having a run of poor luck.

HOWEVER, 35% chance is not a 0% chance, and a 65% chance is not a 100% chance. It's just a 1 in 3 (or 2 in 3) chance of victory, not a gauranteed loss/win.

 

[CellarDweller22]

As with playing Axis & Allies, the whole concept of random battle calculations comes down to risk management. If you enter a battle with 65% odds, you will lose 1/3 times. Will this battle destroy your plans if you lose? If not, it's probably worth the risk.

On the other hand, if you have 95% odds of winning a battle sequence, but will LOSE the game if it doesn't turn out in your favor, then you are giving yourself a 5% chance of LOSING the game.... if possible, you should wait until you have a 1% of losing the battle in that situation.

I've heard all the randomizer arguments here just like everyone else. I once was playing a player who left only 2 infantry in Russia (once again, Axis & Allies). I had 1 infantry and FIVE fighters to attack the capital. A NO BRAINER!!

Anti-aircraft fire....... (they hit only on a 1 on a 6-sided die)

1,1,1,1,1

And I lost the battle, and eventually the game....

-- From The Cellar

P.S. Was the randomizer fair there? Well..... the odds of that roll were 1/7776. If I roll the dice 10,000 times, I should expect to see such a roll once or twice.....

Did I mention that those were REAL dice???

 

[cgannon64]

The RNG doesn't cheat, for the last time.

Get over it. It is a chance-based system. Which means that sometimes you lose when the odds are in your favor...

 

[ainwood]

I personally tend to attack only if the odds are more like 80%....

 

[dragontooth182]

i dont take time to calculate the percentage of me winning. i just attack and see if I win or not. there are more importent things out there than combat.

 

[slothman]

Actually the difficulty level does affect battles but only to barbarians.

 

[Grille]

Originally posted by CellarDweller22


Anti-aircraft fire....... (they hit only on a 1 on a 6-sided die)

1,1,1,1,1

And I lost the battle, and eventually the game....

Did I mention that those were REAL dice???

Did your friend mention that, possibly with a special emphasis?

 

[Ribannah]

The AI doesn't get more favourable odds than the human player. However, there may be a slight auto-correlation in the battle results, or there could be a slight deviation from the uniform distribution near the edges (near 0% and 100% odds). Both could account for upsetting results (positive and negative, mind you!) happening somewhat more often than one would expect.

 

[Gil]

After the delay referring to my NY-vacation I`m here once again.
First, I thank all the subscribers who answered me.
Special regard to those gentlemen who posted messages such as "Do you
understand that terrain and rivers also affect odds?" And there were several
messages like that.
OK, I`ll answer. Yes, I know that terrain and rivers affect odds, too.
And I said that while playing the warlord level yet I wrote the prog that
calculates odds with such factors as terrain, rivers etc. This prog gives the
same results as Combat Calculator v2.0. In the last message I commited a
mistake: I called the math method of calculating odds "Bayes` scheme". It`s
wrong. I mistake Bayes with other mathematic have been researching the theory
of probabilities - Jacob Bernully. The name of the method has to be Bernully`s
scheme. Hope, that in the USA it`s called so, as in Russia.
Second, I must proclaim that I have finished the serie of tests to research
affection of a level on odds.
For that I created the scenario that consists of three grassland tiles in the
ocean. On the one side there are my town and my army and there are enemy`s on
the other side. My and enemy forces are equal and contain 25 spearmen and 25
archers. Spearmen are fortified. First, my archers attack enemy spearmen 25
times. Next, enemy archers attack my spearmen 25 times. I count the number of
my alive archers and spearmen after. I check the number of attacks and events
of rank-up that can affect the number of alive fighters after the battle also
(i.e. 4HP spearmen wins against 3HP archer but has only 1HP after the battle,
he would be dead if he had 3HP) and correct the numbers. If AI stopped the
attack due to the great deal of losses I would wait until he would attack
again. My aim is to get the sample of 25 events (attacks).
I got 10 such samples for every level have repeated the process 10 times.
Due to the laws of math statistics it`s the same thing as 250 events for every
level.
All that is up to me is to calculate mean value for every sample consisting
from 250 events and to compare it with theoretical value. I calculated the
differences between received and theoretical values and compared the
differences with the theoretical mean-square runaround.
Here the results:
results mean value theoretical value difference mean-square
runaround
Chieftain my attack
11 39,2 36,3 2,9 3,04
8
11
9
8
10
8
10
11
12

Chieftain enemy attack
8 39,2 36,3 2,9 3,04
7
6
14
12
9
9
9
9
15

Warlord my attack
9 34 36,3 -2,3 3,04
5
11
9
11
4
9
7
9
11

Warlord enemy attack
6 37,6 36,3 1,3 3,04
11
9
12
9
9
9
13
9
7

Regent my attack
7 34,4 36,3 -1,9 3,04
8
8
11
7
10
8
9
12
6

Regent enemy attack
10 34,4 36,3 -1,9 3,04
5
9
9
4
9
9
9
11
11

Monarch my attack
10 34,8 36,3 -1,5 3,04
9
9
5
9
9
10
7
10
9

Monarch enemy attack
7 38,8 36,3 2,5 3,04
6
8
10
12
15
14
9
10
6

Emperor my attack
12 41,2 36,3 4,9 3,04
8
8
12
8
11
14
11
7
12

Emperor enemy attack
10 36,4 36,3 0,1 3,04
7
10
9
8
10
10
11
6
10

Deity my attack
9 37,2 36,3 0,9 3,04
7
6
8
14
8
9
10
11
11

Deity enemy attack
11 38,4 36,3 2,1 3,04
5
15
10
5
7
10
13
9
11

As is written above, the tables consist of such parts: the first column
contains results (the number of wins) of 10 tests of 25 attacks each, further
there are 4 numbers in percent, they are a mean value, a theoretical value, a
difference between a mean value and a theoretical value, and a theoretical
mean-square runaround.
As a result, I can say that all the tests prove that AI doesn`t cheat for test
conditions. The results are really exact. I didn`t expect that the mean values
wouldn`t differ from the theoretical values no more than theoretical
mean-square runaround for 11 of 12 tests. Only for Emperor_my_attack test the
difference (4.9) exceeded the mean-square runaround (3.04), but it`s normal.
Mean-square runaround is less than a mean runaround also.
In short, the tests proved that AI doesn`t cheat in combat for the test
conditions.

But in the whole, the tests proved nothing. Yes, in simple conditions AI
doesn`t cheat, but could it be that the cheat algorithm was switched off. And
for other conditions (for example, when you had started to win) the algorithm
could be switched on. Assuming this, the problem is much farther from solving.
Yeti told about the third factor affecting odds - psycho factor or human
nature. I thought about this. I cannot say if this factor is really significant
for my perception of the game. May be and may be not. But I can post the
results of all my combats since I had started to play Civ3 on Emperor last time
before I became very disappointed and wrote here two weeks ago.
The form of the results is easy - there are two columns, the first column
contains the probabilities of all my won combats having happend, and the latter
contains the probabilities of all my lost combats having happened.
I didn`t used to enter combat if the probability of my victory was <50%. And if
I entered with the probability >50% and lost, I would put those probability
that AI had in the combat. And if AI attacked me and won, I would put his
probability, or my probability if AI lost. All values are in percent.

I won AI won
85.6 36.3
63.8 35.8
78.4 36.3
63.8 54.5
63.8 79.4
87.4 39.4
60.6 46.1
79.0 36.8
60.6 39.4
95.1 39.4
------------------------------------------------------------------------
mean value = 73,81 mean value = 44,34

I like an answer I received, something like this - You lost several battles and
you are complaining, and it isn`t a classic combat between a tank and a
spearman.
Also I received this one - Luck counts too.

I`m 22 y.o. DSP-student. I don`t hardly believe in luck. In Russia there is a
saying. "H¥ ¡®£¨ £®p誨 ®¡¦¨£_îâ." Or in english "Not gods fire pots."
Yes, if I lost a combat, I wouldn`t complain.
If I lost a classic combat between a tank and a spearman, I wouldn`t complain
too.
But if I lost ten such combats a row, I would think if AI cheats.
There is such thing as the Chebyshev`s law of big numbers. It proclaims that a
sum of some random values divided by their number would have a dispersion less
than a dispersion of one of such random values. Thus, I must expect a mean
value of results of some tests approaching to a theoretical value (or a truth
value, if it isn`t known) and more rapidly with increasing of the number of
such tests.
If you tossed a coin 10 times you could expect that 70% would be of one side.
But if you tossed it 100 000 times you couldn`t expect that 70% would be of one
side. It`s hardly believeble event. So I cannot fully (without some doubt)
believe in such sample like posted upper. And I could never believe if the
number of items in the sample was about 1000. Yes, from the first sight, you
could think that you could toss a coin 1000 times a row of one side. But if you
knew of mathematics and science more, you couldn`t think so, because you must
wait a large amount of time tossing a coin to such an event to occure. A large
so it`s larger than your lifetime.
OK, the volume of my current sample is very small. But I think I would continue
to play the game and to add in the sample results of all battles after I forgot
of my total disappointing in the game.