I believe combat is rigged in this game and it ruins it for me

[Mr. President]

I thought all the experiment reports in this thread were quite interesting.

I also thought it was interesting that the usual chorus of "38 people agree with me, less than that agree with you, STFU noob" came out right on schedule.

[marvinkosh]

Tell me something - given the nature of round-based combat, which is going to demand at least one random number from the generator each round, is it possible that you can overwhelm the ability of a computer to create pseudo-random numbers and thusly obtain results which favour certain circumstances?

What I guess I'm asking is, how can you be sure you're not seeing a pattern arising from the RNG itself? How good is the randomness before you even start to set units against each other?

[Blake]

Tell me something - given the nature of round-based combat, which is going to demand at least one random number from the generator each round, is it possible that you can overwhelm the ability of a computer to create pseudo-random numbers and thusly obtain results which favour certain circumstances?

In a word, no. The cycle of numbers will repeat, with some period somewhere between seconds in a year, to seconds in the lifetime of the universe, to atoms in the universe, depending on the quality of the RNG.

What I guess I'm asking is, how can you be sure you're not seeing a pattern arising from the RNG itself? How good is the randomness before you even start to set units against each other?

In a sense, all patterns arise from the RNG itself. Take a perserved random seed game.
Imagine the next sequence of random numbers (each from 0-99) is something like this:
8, 4, 9, 3, 7, 14, 11, 16, 83

It wont matter what unit you attack with, as it suffers from a series of very small numbers that result in it losing. This is why players will have a complaint where whatever unit they use loses - it's because by sheer chance it hits a series of very small numbers. A good RNG will have such series of small numbers, it is inevitable.

What people actually want is a non-good RNG, one that instead produces more regular, predictable random numbers.

[marvinkosh]

What I mean is, let's say you have ten rounds of combat for a pair of units using those numbers - or less than ten, if one of the units expires before reaching the tenth round - and then you have ten rounds of combat for another pair of units. Is it possible that by choosing to discard (or not discard as the case may be) the numbers generated for post-victory rounds, you can inadvertantly affect the outcome of the second combat?

[xxFlukexx]

Originally posted by Blake
What people actually want is a non-good RNG, one that instead produces more regular, predictable random numbers.

Yes, this is what I want, hehe. My problems with combat and the RNG are a little different from the OP so I guess I should make another thread but mehhhhh I hate making new threads.

I think that randomness is a great thing to add in Civ, but the degree of randomness currently in Civ4 just kills some of the strategy. I've had a lot of games just ruined and lost simply because of bad combat results, I've had streams of pikemen pour out only to lose repeatedly to Knights, etc, etc. Now in Civ2 I would occasionally "reload-cheat" because I really didn't bring enough troops and I was really fickle, but in Civ4 I find myself losing my whole stack vs a city, then reloading, hitting end turn and then completely demolishing the enemy city with the same units. I don't want the results where I horribly win or horribly lose, I want something inbetween.

I just think it is time for Civ to lessen the amount of randomness in it's Combat System. Again, if you played Starcraft: in a game where you and your opponent did a Zealot rush, 8 vs 8 fight and you lost the game because of dumb luck, Starcraft would never have been popular.

Just want a fair fight. Not losing troops repeatedly when I have over a 99% chance to win. I don't see how this is such a crazy thing to want, surely a streak-breaker code could be put in fairly easily, maybe a "fair combat" option in Custom game?

[Blake]

OMG. I'm playing as Cyrus right, I'm the AI player, obviously.

This (stuipd ****ing cheating) human player, called grater or something stupid, he parked this infantry on his 2 workers. Unfair! He KNOWS I can't resist workers! I threw 3! THREE! TWEEEEEE! infantry at his cheating infantry and they all died and it had nearly all it's health left!!!! I can't believe this!!!@#

**** this game, I'm giving up. All the humans cheat and get combat advantages. Oh yeah, I'm playing on Monarch and I'm getting like +10% production and stuff dunno what that's about probably another bug.

[Blake]

Just want a fair fight. Not losing troops repeatedly when I have over a 99% chance to win. I don't see how this is such a crazy thing to want, surely a streak-breaker code could be put in fairly easily, maybe a "fair combat" option in Custom game?

There's an excellent chance that streak-buster can be modified in when the SDK comes out. A firaxis dude has already said we'll be able to examine the combat/RNG code, which means we can edit to too.

[TheHateMale]

Originally posted by DrSpike


Ok that's enough, and I'm not going to respond to whatever nonsense UR comes up with next.

I would like to reiterate what he said then - I originally made the same mistake. If all units had one hitpoint (and this was after 1.52 had been installed, obviously) then it would be a binomial distibution.

Because combat takes multiple rounds, this isn't a binomial distribution.

Combats (ignoring retreat) end as soon as one unit has done enough to kill the other. Assuming this takes 5 succesful attacks - the other unit could land 0, 1, 2, 3 or 4 attacks. The total number of attack rounds could therefore be 5 to 9. You unfortunately can't ignore this when you base a probablity distribution on the results or you'll have a very inaccurate model.

To expand further, only counting the number of surviving units effectively ignores any succesful combats from a unit that doesn't win. This appears to be modelled on the US voting system (You're throwing your vote away!).

\Please don't focus on the vote analogy.

[Aro]

Originally posted by Blake

There's an excellent chance that streak-buster can be modified in when the SDK comes out. A firaxis dude has already said we'll be able to examine the combat/RNG code, which means we can edit to too.

Barry Caudill, maybe? According to him you can edit not almost everything, but everything… Or close enough.

Caudill: The first big step in that direction will be the release of the Civ IV SDK in late January/early February. The SDK will open up the game code for hardcore modders and allow them to change the game on a very fundamental level. The current tool set is very powerful in its own right. Mod makers can already make maps, add units, civilizations, or buildings, and have scripted sequences or random events. The SDK will allow experienced programmers to change the AI and the game to create completely new experiences outside of the norm for Civ IV. Want to make underwater cities? How about adding a new kind of great leader for military uses? Want to bring back armies like in Civ III? Interested in getting rid of workers in favor of a broader public works system? All of this is possible -- and that's just the tip of the iceberg!

Interview with Soren and Caudill (gameSpy)

[LordShiva]

Cyrus... I mean Blake.

[DrSpike]

Originally posted by TheHateMale


I would like to reiterate what he said then - I originally made the same mistake. If all units had one hitpoint (and this was after 1.52 had been installed, obviously) then it would be a binomial distibution.

Because combat takes multiple rounds, this isn't a binomial distribution.

Combats (ignoring retreat) end as soon as one unit has done enough to kill the other. Assuming this takes 5 succesful attacks - the other unit could land 0, 1, 2, 3 or 4 attacks. The total number of attack rounds could therefore be 5 to 9. You unfortunately can't ignore this when you base a probablity distribution on the results or you'll have a very inaccurate model.

To expand further, only counting the number of surviving units effectively ignores any succesful combats from a unit that doesn't win. This appears to be modelled on the US voting system (You're throwing your vote away!).

\Please don't focus on the vote analogy.

I'm sorry, but like UR you are wrong on this. I'll respond since you were reasonable.

Others have actually already posted to explain why you are wrong, with varying degrees of sophistication. Let's go back to the coin tossing example. We could make our basic model more complex by making the result of each coin toss the best of 3. Under this rule if there are 2 heads and 1 tail it counts as one head, because heads won the best of 3.

Thus now we have rounds of coin tossing. Does this change the probability that heads win a given best of 3? No, its still 0.5, as the extra combinations within the round don't change the essence of what we are doing.

Draw the probability tree for the simple experiment. You will see that it's 0.25 heads win 2-0, 0.25 tails win 2-0, 0.25 heads win 2-1 and 0.25 tails win 2-1.

Or, 0.5 that heads win and 0.5 that tails win the best of 3.

I could add hitpoints and more rounds. It wouldn't change a thing, as the essence of the problem will remain the same.

I hope that clears it up.

[JackRudd]

To be strictly accurate, *when the relevant probability is 0, 0.5 or 1*, adding in extra rounds or (equal) extra hitpoints alters the relevant probability not a bid. If the relevant probability is something else, the extra rounds/hitpoints will tend to skew results in favour of whichever side already had the advantage.

[TheHateMale]

True, but being beaten 50 by units that don't take any damage is a different situation to being beaten by 50 units that are all left almost dead.

I'd be far less likely to think battles rigged if on equal odds the winner scraped by rather than if they walked all over me.

What I'm trying to get at is a binomial distribution treats combats as if one side was destroyed and the other was undamaged. There are more than two outcomes though - any amount of hitpoints can be knocked out. For the same amount of combat successes one side could have a few undamaged units, or several slightly damaged units, or many heavily damaged units. This is not recorded in the 'Win 30, Lose 20' level of detail.

The same combat round results, shuffled around between individual fights will give a different number of wins and losses. If this wasn't true, you could access the next 100 or so random numbers that will be used (assuming that they're generated in advance) put them in any order you want and still have the same end results. When I flip a coin 100 times, if the first 50 are all heads, and the second 50 are all tails, it makes no difference. It's not the same with combat.

I know I've said it before, but it would only be binomial if every fight in the sample lasted only one round.

I have no idea why I've put this much effort in when I have no problem with the battle system.

[DrSpike]

Originally posted by JackRudd
To be strictly accurate, *when the relevant probability is 0, 0.5 or 1*, adding in extra rounds or (equal) extra hitpoints alters the relevant probability not a bid. If the relevant probability is something else, the extra rounds/hitpoints will tend to skew results in favour of whichever side already had the advantage.

True, but we are talking about even fights.

[DrSpike]

Originally posted by TheHateMale
True, but being beaten 50 by units that don't take any damage is a different situation to being beaten by 50 units that are all left almost dead.

I'd be far less likely to think battles rigged if on equal odds the winner scraped by rather than if they walked all over me.

What I'm trying to get at is a binomial distribution treats combats as if one side was destroyed and the other was undamaged. There are more than two outcomes though - any amount of hitpoints can be knocked out. For the same amount of combat successes one side could have a few undamaged units, or several slightly damaged units, or many heavily damaged units. This is not recorded in the 'Win 30, Lose 20' level of detail.

The same combat round results, shuffled around between individual fights will give a different number of wins and losses. If this wasn't true, you could access the next 100 or so random numbers that will be used (assuming that they're generated in advance) put them in any order you want and still have the same end results. When I flip a coin 100 times, if the first 50 are all heads, and the second 50 are all tails, it makes no difference. It's not the same with combat.

I know I've said it before, but it would only be binomial if every fight in the sample lasted only one round.

I have no idea why I've put this much effort in when I have no problem with the battle system.

Hatemale, of course there is a difference between winning easily and winning with few hp left. However it is entirely irrelevant to the debate, as the debate is about winning.

The binomial techniques myself and others have described PERFECTLY model what we say they model, the proportion who win out of 100 repetitions or whatever.

The calculations about HP remaining can be derived from a similar (but much more complex) technique, and these will depend on the nature of combat resolution.

Is that clear?