RNG - Civ's Detractor?

[CellarDweller22]

To clarify, it seems that this is not about the RNG, but about the application.....

If we start placing results that are with "two dice", so to speak, then that would change the battle outcome percentages in a parabolic fashion (or something like that! ), leading to percentages that are almost impossible to calculate. I'd HATE to have to use an odds calculator while playing civ.....

Hasn't anyone used REAL dice? How about those aa guns in Axis and Allies? People always used to complain that the computer's numbers were skewed, yet in my last real life game, I threw FIVE ONES (they only hit on a one)..... and my opponent thereby lost the game.

I also know that in key battles in Risk, I probably threw sixes on defense half the time....

Remember, this game isn't chess..... it's more akin to backgammon. Not only must you plan out your moves, but must also calculate the odds of certain things happening several turns in advance. The great backgammon players always know what the odds are, and when to accept or reject the double (and thereby "rack em up" again).

In my current civ game, I saw three samurai take out 3 tanks (AI vs AI). It happens. How often?? well, if it's a 5% result, it's statistically insignificant. If your odds of taking a city are 90%, then one in ten times you fail. Are we saying that you fail 3 in 10 times because of the dice generator? I would believe that someone failed to take into account some insignificant factor (fortified, city of size 7, terrain, etc.) before I would believe that the numbers are skewed. Perhaps someone could perform a test?!

-- From The Cellar

 

[carlosMM]

cellardweller: test have been done to the fare-thee-well.

the problem is 'strings' - first doing the tests as that requires a far larger databse, then deciding what is 'normal'...

I personally feel that the strings of good and bad luck are too long, too. but testing it in a meaningfull way is tough.

 

[TheNiceOne]

Interesting discussion.

My view is that Firaxis should do absolutely nothing with the RNG. The RNG gives (close to) perfectly distributed random numbers with a flat distribution, and that's how it should be. With a bell-shaped distribution it would be damned hard to know the chance for even a 1 HP attack 6 unit vs a 1 HP defense 3 unit.

The RNG is only the input for the battle calculation though, and whether the battle calculation in itself should be changed is another matter.

One simple change is to increase the number of HP for all units. This will not alter the result for any single battle round, but will still make the total result more centered around the middle, i.e. the strongest unit will win more often. The problem with doing this is that it makes bombardment less effective, but a doubling of HP could be combined with a doubling of the fire rate of bombardment as well.

One possibility that requires more programming by Firaxis could be to require 'x' number of battle round victories for each HP. X=1 would be as today: win one battle round and the enemy looses one HP. X=2 would mean that you had to win two battle rounds before the opponent won two battle rounds to take a HP from the enemy. In effect, 2 or 3 RNG numbers would be used for each HP. This would be computed behind the scene, but the effect would be close to a doubling of HP. X could be increased further to get deterministic results. X should of course be adjustable in the editor, with 1 (as today) as default.

Whether this is needed is a matter of taste. Personally, I think slightly less randomness in the ancient age only would be preferable. In the ancient age, wars are conducted with relatively few units, so a string of special results means quite much. In the later ages the current randomness is perfectly fine (IMO).

 

[TheNiceOne]

Originally posted by carlosMM
the problem is 'strings' - first doing the tests as that requires a far larger databse, then deciding what is 'normal'...

Well, extensive tests of the CIV3 RNG has proved that the strings produced matches truly random numbers, and there should be little doubt that the definition of 'normal' should be truly random numbers, not some people's expectation of how random strings should look like.

I personally feel that the strings of good and bad luck are too long, too. but testing it in a meaningfull way is tough.

This has been extensively tested and proven to be as it should for truly random numbers... These tests can be found if the search engine is currently available (I'm too lazy to check myself)...

 

[anarres]

Originally posted by TheNiceOne
...there should be little doubt that the definition of 'normal' should be truly random numbers, not some people's expectation of how random strings should look like.

Hmm, I agree with everything else you said, but why are you confusing the normal/abnormal form of the word 'normal' and the normal distribution?

What you (obviously) mean when you say the RNG is "truly random" is that the numbers are randomly generated form a uniform distribution. You can call this 'normal' if you want, but that is just confusing the matter.

It is certainly possible to draw a set of random numbers from a normal distribution (as in bell curve, not normal/abnormal), and to use that in place of the current uniform RNG. The fact that they chose to use a uniform distribution is probably to make it easier, but as someone else pointed out you can sum two random numbers from a uniform distribution to get a normalised result.

 

[TheNiceOne]

Originally posted by anarres
Hmm, I agree with everything else you said, but why are you confusing the normal/abnormal form of the word 'normal' and the normal distribution?

What you (obviously) mean when you say the RNG is "truly random" is that the numbers are randomly generated form a uniform distribution. You can call this 'normal' if you want, but that is just confusing the matter.

Its possible I misunderstood carlosMM, but I used the word 'normal' in the same way as I though he meant it, i.e. the normal as opposed to abnormal meaning. I understood him to imply that the CIV3 RNG gives abnormally many/long string of special results, and I therefore answered that it doesn't.

It is certainly possible to draw a set of random numbers from a normal distribution (as in bell curve, not normal/abnormal), and to use that in place of the current uniform RNG. The fact that they chose to use a uniform distribution is probably to make it easier, but as someone else pointed out you can sum two random numbers from a uniform distribution to get a normalised result.

Yes, that is certainly possible, but that would make it very hard to calculate the odds.

 

[anarres]

OK, my bad TNO. I guess I was confusing myself. I didn't see carlos use the word in quotes.

Regarding the predictability of results: would you not prefer it if the combat results were harder to work out for us mortals? I think I would, even if I do like to know of all the in-game mechanics.

 

[carlosMM]

Originally posted by TheNiceOne

Well, extensive tests of the CIV3 RNG has proved that the strings produced matches truly random numbers, and there should be little doubt that the definition of 'normal' should be truly random numbers, not some people's expectation of how random strings should look like.

This has been extensively tested and proven to be as it should for truly random numbers... These tests can be found if the search engine is currently available (I'm too lazy to check myself)...

sorry TNO, you misunderstood me as I (again) didn't properly distinguish between rng and battle results.

I agree with everything you said, what I emant was: what is the 'normal' (= wanted for civ) outcome in battle results. As you said, HP changes will do, the other system you describe sounds even better. It would be hard to define mathematically what amounts to 'too stringy' for civ and what not. After all, civ is not real life (and even there humans then to try and avoid randomness!) but also isn't chess - where exactly is middle gorund and how do you get that into a formula?
I tried statistics on starts with early war but quickly gave it up - even if I just take MY personal view as the basis, the tests get too complicated!

 

[TheNiceOne]

Originally posted by anarres
Regarding the predictability of results: would you not prefer it if the combat results were harder to work out for us mortals? I think I would, even if I do like to know of all the in-game mechanics.

I'm not sure... In one way, I like to know what I'm doing, and at least have a good idea of my chances, but on the other hand, I want to have the feeling that I'm playing a game where I'm leading a nation - not that I'm performing some mathematical exercise. I guess I will be content with a vague knowledge of the actual statistics, but what is important to know is how to improve my chances (not attacking across rivers, bombing a city to below 13 or below 7 etc.).

 

[Dr Alimentado]

I would definitely prefer combat results that were harder to work out... but I'm the kind of player who only grudgingly learns the mechanics, I'd prefer it if the nitty gritty of the mechanics were totally inscrutable to all

What I actually want a better combat model all together. Without turning civ into a wargame something like CTP2, make all stacks into combined arms 'armies' that fight as a single entity.

 

[BomberEscort]

People have commented that the battle calculation may be the problem (by this I assume you mean the A/(A+D) calc)... What would we replace it with???

 

[carlosMM]

Bomber Escort: A and D is fine I think it is just that there is no correlation between the COST of a lost battle and the commonness of unexpected losses. If you lose 5% of your per turn production 5 time in a row (25%) despite a 80% probability of losing only one unit (i.e. not 25% but 5%) that is tolerable. If you lose 60% four times (240%) in a row despite a 80% probability of not losing any battle at all (i.e. 20% of losing 60% or more but maybe only 1% of losing 240%) that hurts the gameplay beyond the point of fun!

So random numbers is fine and the principle of probabilities is fine - it is just that in my experience the ture random strings hit too hard especially in the early stages of the game. Replaying the current system is tough work unless you simply double HP.........

 

[RougeKing]

In the long list of stupid game outcomes, calvary defeated several of my tanks. If this worked in the REAL WORLD, Germany would have never survived in Poland and WWII would have been a dust up.

RNG is OK, but unit to unit combat should compare unit capabilities. Knights in shining armour, men with spears should loose outright to modern units.

 

[BomberEscort]

If the cavalry were attacking this is very possible...

 

[Melchior]

I am not concerned about how random generator operates inside the game. I only think about more factors from real life that normaly interfere with the outcome of any battle. The game is , in fact, the simulation of history line, and I would be more satisfied if it could be more realistic. But not like chess game. I'd like to calculate with more elements before I get into a war conflict (element of surprise, weather conditions, etc.), but I have no problem with random numbers - they have to be used because it's too difficult to predict all factors. There is no formula that can describe any real life event. So, randomness have to be accepted as necessity. (I hope this will be slightly changed in the next expansion).

 

[JavalTigar]

I've always said "When you start screaming at the game it ceases to be fun."

I thought the first Civ had way to many one sided battles (ie battleship sunk by spearchucker)

Civ 2 concepts of hit poins and fire power really helped.

Civ 3 systems are a slide backwards from Civ 2. But they are not that bad.

As for the case at hand. You should not attack spears with spears! Shock troops are needed to break the line.