promotion probabilities

[sumthinelse]

Originally posted by mirrek23


For an attacking unit, the odds seem right, but again, if I attack twice in the same turn with the same unit (obviously not possible until late in the game), it is *always* promoted.

I am seeing something like that in the windows 1.21f version. But I think you mean that you have to kill 2 enemy units in the same turn, not just make 2 attacks. If my tank attacks and wins the first attack but retreats on the 2nd attack there is no promotion (unless it happened on the 1st attack), correct?

 

[mirrek23]

D'oh!

Yes, I meant you have to attack and *win* twice...

 

[Mike B.]

Oh yeah, that's one little tidbit I forgot to mention. If a victorious unit fails to get promoted, it will always be promoted upon surviving any other battle in the same turn. This gives an obvious advantage to defenders as well as an incentive to use the same offensive units for additional battles whenever possible.

 

[kring]

I have had a number of units defending being attacked two or more times in numerous games that have NOT been promoted. It happens infrequently, depending on experience level of the unit.

I still say 100 is too small a number. There are statistical abnormalities all over the place. I have seen games where things went almost all one way or the other (sometimes good for me, sometimes bad), yet in the long run: several thousand combats for me as well as reading the results of others, have me agreeing the odds are close, both for promotion and for leader appearance.

 

[Lt. 'Killer' M.]

Mike, thanx!

 

[Evincar]

Hey, you see, Mike has confirmed what some of us said in a thread called "Easy promotion" or something like this.

 

[Isak]

So what happens if we add in a second wonder with the same effect as the Heroic Epic and build both in a game.
Will the second one have any effect?

 

[kring]

You might be able to find out by looking in the civpedia txt file under Heroic Epic to see if it says anything useful on the subject.
I would do that, but I am not at my Civ 3 computer at the moment.

Interesting question. I wonder what the answer will turn out to be?

 

[Isak]

This is the description from the civilopedia.txt:

Increases the chances of the appearance of leaders from victorious combat.

Not much help there, I'm afraid

 

[kring]

I guess someone will have to mod one and see. I do know on some things, they were designed to be one shot deals. You could mod other things, but only the first counted or the game crashed when you built the second one.

 

[Isak]

Originally posted by kring
I guess someone will have to mod one and see.

We have already done that in the DyP mod - but testing whether they actually work is rather impossible when the results are randomly determined.

 

[sumthinelse]

Originally posted by Isak
We have already done that in the DyP mod - but testing whether they actually work is rather impossible when the results are randomly determined.

Not impossible. It's easy to create a scenario with 25 modern armor (modded to have 100 moves) attacking 100 AI units (warriors, chariots, horsemen, archers). The tricky part is to create the 2 buildings in your city and make the armor units all elite. This you can do with the civ3 multitool http://forums.civfanatics.com/showthread.php?s=&threadid=7787. If you make sure the option "preserve random seed" is off you will have 100 trials every time you load this game. Of course you have to remember to disband a leader every time one is created and stop using an elite modern armor unit every time it creates a leader.

 

[sumthinelse]

Originally posted by kring
I have had a number of units defending being attacked two or more times in numerous games that have NOT been promoted. It happens infrequently, depending on experience level of the unit.

I still say 100 is too small a number. There are statistical abnormalities all over the place. I have seen games where things went almost all one way or the other (sometimes good for me, sometimes bad), yet in the long run: several thousand combats for me as well as reading the results of others, have me agreeing the odds are close, both for promotion and for leader appearance.

Anybody who can post a game to prove that Mike B. is wrong about this? The probabilities and exceptions about multiple combats on a single turn in MIke's posts describe the way the game is supposed to work, so I would be interested in seeing a game posted that exhibits behaviour contrary to what Mike B. says.

Of course, programmers are human, and mistakes are possible.

If I try an event that has a .5 probability 100 times, what are the odds that I will get a distribution of 65/35 or 35/65, that is, where the results differ by 30 or more? I don't know the answer but I don't think I have ever seen that happen.

 

[Lt. 'Killer' M.]

sumthinelse: I do get the feeling that some things are sytem dependend that are not supposed to be. Remember the resource thing? i get strange results all the time. Maybe, the RNG for this IS buggy, after all!

On the other hand, I think it is far more probable that people mix up the usually several defenders per city. After all, why should No. 2 be promoted 'out of probability' of No.1 won a fight?????

 

[sumthinelse]

Originally posted by Lt. 'Killer' M.
On the other hand, I think it is far more probable that people mix up the usually several defenders per city. After all, why should No. 2 be promoted 'out of probability' of No.1 won a fight?????

Good point. There is no message when the combat switches to another defender, no good way to see exactly what is happening unless you have the source code and a debugger.

Isn't it late at night in Deutschland? You don't have to go to work tomorrow????

 

[Lt. 'Killer' M.]

Hell, i handed in my thesis last Friday, now it is somwhere in the Himalaya along with the professor who corrects it - so what the hell

and I do want to get the SG out of the way so i can get to 'my' resource test map - after yours got me all confused

 

[sumthinelse]

Yes, that iron is a slippery critter!

 

[kring]

Originally posted by sumthinelse


Anybody who can post a game to prove that Mike B. is wrong about this? The probabilities and exceptions about multiple combats on a single turn in MIke's posts describe the way the game is supposed to work, so I would be interested in seeing a game posted that exhibits behaviour contrary to what Mike B. says.

Of course, programmers are human, and mistakes are possible.

If I try an event that has a .5 probability 100 times, what are the odds that I will get a distribution of 65/35 or 35/65, that is, where the results differ by 30 or more? I don't know the answer but I don't think I have ever seen that happen.

I don't think Mike B. is wrong.

On the other point, I have seen it happen a lot, first and second hand reports. Example: the number of Great Leaders based on 1/12 -1/32 figures haven't happened that often to many people. Even allowing for the ELITE can't create a GL themselves rule now known.

I have rarely gotten a single GL in a game; yet, I believe the odds are as stated by Firaxis. Why? Because other people's games have sometimes gotten them more frequently than the stated odds. It averages out in the long run. We are talking computer random generated seed numbers, much different than rolling the dice. The problem is people get caught up with statistics, probability, and numbers (lest we forget: there are lies, danged lies, and statistics). Statistics aren't all that are hyped as being.

 

[sumthinelse]

Originally posted by kring


Example: the number of Great Leaders based on 1/12 -1/32 figures haven't happened that often to many people. Even allowing for the ELITE can't create a GL themselves rule now known.

I have rarely gotten a single GL in a game; yet, I believe the odds are as stated by Firaxis. Why? Because other people's games have sometimes gotten them more frequently than the stated odds. It averages out in the long run.

OK, I agree. For testing something with a 1/12 probability, which is the best possible odds for a leader, 100 trials does not give you a good picture. Getting only 5 leaders out of 100 tries would not surprise me too much. And if a lot of the combat is on the defense against without the epic (1/32 per trial), well, you need a lot of trials for leader generation! Even with 100 trials you should average 0 leaders 1 in 24 times you do the 100 trials.

And for non-mil. defending against barbs, to test for promotion to
elite (1/16 per trial unless they attack the same defender multiple times) you need more trials also.

 

[Isak]

Originally posted by sumthinelse
Not impossible.

Nope, but rather impossible - I don't think 100 tries are enough either - I wouldn't even believe it after 1000 tries, but I can see your're all starting to agree with that. Like Kring says computer randomness is never really random.

That's why I just thought it would be so much easier if we could just get word from some of the big boys, whether or not the Heroic Epic effect is a one-shot deal or whether it can be used to increase the chances several times in the same game. Anyone with access to the source code could look it up in a fly