Combat cheats

[Gil]

2Ribannah:
It must be not very slight auto-correlation, I think. In the whole,
auto-correlation may be the result of any AI`s manipulations with odds.
I think, they could create a well working random generator in the game. They
must aim to this. If a random generator gave perceivable auto-correlation, it
would be a bad random generator. Can it be that the random generator in Civ3 is
a so bad one?
And, please, tell about deviation from the uniform on the edges. Why is it so?

 

[Yeti]

Issues with the random number generator could cause bad streakiness and tests with bad std. dev., etc..., but that shouldn't change with difficulty.

I'd like to see your first and last (chieftain and diety) tests repeated a couple more times each.

There's a fair amount of variation in the results - you're looking at a small variation that is within the apparent noise in your stats - so it would be interesting to see if the AI's victories on diety always tended towards the high end.

I don't think the AI cheats in combat, I believe it has been stated that the AI doesn't cheat in combat by Firaxis (meaning that if a discrepency really exists it is a bug, not design), but I'm willing to believe numbers if they are sufficiently convincing.

 

[Padma]

Welcome to CFC, Gil.

(I missed it when you first posted a couple of weeks ago. )

I am glad to see another member here who is willing to get into the mathematical processes involved. But be forewarned, when you say the AI cheats in combat, you are going up against a staggering amount of testing and evidence compiled by other mathematically-inclined players here, which show the exact opposite, as well as the word of the developers themselves, who claim they did not put such a cheat in the game.

 

[IglooDame]

Gil, you might also create some tests in which the AI has a few hundred tanks back in it's home country, and some tests in which you have a few hundred tanks in your home country and then compare the winning percentages of your spearmen vs archer type tests, so as to rule out a possible "cheat algorithm" being present but on or off.

 

[Longasc]

I am studying Statistics, but Gil really got it...

I recommend for everyone that is interested in root mean square and so on:

http://www.amazon.com/exec/obidos/t...103-2517951-0548605?v=glance&s=books&n=507846

That's what I am learning right now. I am really excited to see someone using it for a real purpose...

Well, I saw no mistake in his elaborations... too long, but nothing wrong afaik.

Now that I see that the AI cheats, and I do not have the time to do so many tests myself and prove his results... I simply cheat, too. *reload*

Well, perhaps could Gil take a look at the fact that always Incans are spawned when chosing random civ?

 

[Hurricane]

Gil, here is some reading for you. Just so you don't do any unnecessary work...

http://apolyton.net/forums/showthread.php?postid=1251043#post1251043

http://forums.civfanatics.com/showthread.php?postid=264498#post264498

http://forums.civfanatics.com/showthread.php?postid=171590#post171590

http://forums.civfanatics.com/showthread.php?s=&threadid=44046

 

[Ribannah]

Originally posted by Gil
2Ribannah:
It must be not very slight auto-correlation, I think. In the whole,
auto-correlation may be the result of any AI`s manipulations with odds.
I think, they could create a well working random generator in the game. They
must aim to this. If a random generator gave perceivable auto-correlation, it
would be a bad random generator. Can it be that the random generator in Civ3 is
a so bad one?
And, please, tell about deviation from the uniform on the edges. Why is it so?

There is no need for AI manipulation to cause autocorrelation.
It's simply a case of whether Firaxis chose a good random number generator or not.

Remember that there is no such thing as a true random number generated by a computer (in fact, there is no such thing as a true random number, period!), it's always a formula.

Most random number generators satisfy first- and second-order conditions (mean and variance), as well as symmetry, but many do not satisfy higher-order conditions, percentiles and perhaps least of all absence of autocorrelation. This is a genuine problem in statistical research, and many mistakes are made by researchers who are not aware of this.

It's likely that in a game, any game, not the most sophisticated random number generators are used.

I would therefore not be totally surprised if Spearmen do indeed have better chances against Tanks than they should have, or if heroics (favourable results by one defender against many attackers) were more common than they should be. It could also be that the opposite is true, however.