General noob question grab bag

[elitetroops]

Series of random numbers will NATURALLY show streaks. A series of coin random coin flips won't be HTHTHTHTHT on into infinity; it will naturally have streaks.

Not sure if this was aimed at me, but this was pretty much my point with the post you quoted anyway. With "built in streakiness" as a flaw, which many have mentioned, I assumed that would mean a flaw in the code that makes a low roll more likely after another low roll than after a high roll. Didn't see any signs of this in my tests though.

But yes, if you ask your buddy to simulate 100 coin flips by writing a random sequence of H or T on a paper, then you actually flip a coin 100 times and write down the sequence on another paper, you can usually recognize the sequence derived from the real coin flips based on the fact that it shows more streaks. A person trying to simulate a random series will compensate when he feels there is too many in a row of one or the other result. A coin doesn't care.

 

[FlorbFnarb]

Precisely. It's the sort of thing that causes our natural pattern-detection algorithms to detect false positives. The streak seems to mean something when in reality it doesn't.

 

[Pangaea]

Not sure if this was aimed at me, but this was pretty much my point with the post you quoted anyway. With "built in streakiness" as a flaw, which many have mentioned, I assumed that would mean a flaw in the code that makes a low roll more likely after another low roll than after a high roll. Didn't see any signs of this in my tests though.

Precisely. It's the sort of thing that causes our natural pattern-detection algorithms to detect false positives. The streak seems to mean something when in reality it doesn't.

That could well be (false positives), but in this case we happen to know that there is an issue with the game's random number generator, as explained by Lennier just a few posts above.

When so many combats are simulated these issues won't show, as over time it will even itself out. But if you happen to experience a 'streak' when using 7 units to attack a city early on, and you lose lets say 6 of 7 units at 50-60% odds, then it clearly doesn't even itself out. Use 100 units and it probably will.

When several different code divers say there is a problem with the RNG, I think there is reason to trust them.

It's not a huge problem, it's not like if you lost a 90% combat you're doomed to lose the next 10 such combats because the RNG is so dependent on the previous number drawn, but there does appear to be an issue, and a different RNG algorithm would have been better.

 

[elitetroops]

When so many combats are simulated these issues won't show, as over time it will even itself out. But if you happen to experience a 'streak' when using 7 units to attack a city early on, and you lose lets say 6 of 7 units at 50-60% odds, then it clearly doesn't even itself out.

But this is not a problem with the RNG, as long as it doesn't happen more often than expected. The odds of losing 6/7 consecutive battles with 55% odds would be about 3,6% or 1/28. Now if you consider that 35 battles include 29 different sequences of 7 consecutive battles, you will see that this isn't as uncommon as you think it should be.

Sure, if you happen to lose 6 out of your first 7 battles you're screwed. But if this didn't happen every now and then, the RNG would really be broken.

And I'm not sure if Lennier said anything about the civ4 RNG. He just explained what a bad random number generator means. Keep in mind that a RNG that avoids producing results such as losing 6/7 is also correlated and therefore broken.

 

[Pangaea]

I recall from when this was discussed earlier that the RNG algorithm was one that had the feature Lennier spoke about, whether that is the exact method used in Civ 4 or not. Quite simply, it doesn't draw completely random numbers.

Of course, if the RNG made you lose a 50% battle after winning one, it wouldn't be a true RNG either, so that's not what I'm talking about.

 

[elitetroops]

No computet algorithm can draw completely random numbers, it all has to be calculated base on something. But they can get close enough... Of course there are better and worse pseudo random number generators.

In civ the RNG is called several times over the course of a battle, once each round, so any minor flaws should not affect the outcome of consecutive battles. It would lead to the percentages being off at very high odds, like Fippy mentioned, but my 10000 battle test did not support that theory...

 

[FlorbFnarb]

Random numbers move to the mean as the sample size increases. If you fight a series of seven 50% chance battles, there's not only no reason to expect them to even out half and half, they CAN'T even out like that; you can't win 3.5 matches, after all.

 

[Pangaea]

I'm not sure if people refuse to understand, or you think I'm arguing that the RNG is broken, but you don't have to take my word for the problems with the RNG. Take the code divers' instead. I've looked at the code myself, but it's a little tricky to understand. But what God Emperor writes in this thread (see page 2 and 3) fits pretty well with what I recall being discussed before. Don't recall who that was, if it was Tachy, DanF or somebody else, but it was somebody with good code knowledge.

That there are problems with the selection of RNG algorithm doesn't mean the RNG is crap. It simply means it could, and perhaps should, have been better than it is.

 

[drewisfat]

Do battle dynamics change significantly with different units? I.e. more rounds / hp / damage?

My thought was that maybe odds for battle results were shortcutted to be mostly accurate but not using the correct probability equation, in part because it might require a lot of minor tweaks if units are injured / different units / withdrawal chance, etc.
There's also other ways there could be issues, i.e. the streakiness mentioned that may crop up in the RNG, such that rolling 8 heads in a row isn't quite as uncommon as it should be.

Brute testing battle calculations seems like it would alleviate all the concern. If the results held up at very high numbers what would code diving rly prove?
My question is whether your specific test conditions matter. I.e. do certain promos / injured units / stronger units change anything. I never look at the battle results, so I'm mostly talking out of my you know what here.

 

[elitetroops]

I'm not sure if people refuse to understand, or you think I'm arguing that the RNG is broken, but you don't have to take my word for the problems with the RNG. Take the code divers' instead. I've looked at the code myself, but it's a little tricky to understand. But what God Emperor writes in this thread (see page 2 and 3) fits pretty well with what I recall being discussed before. Don't recall who that was, if it was Tachy, DanF or somebody else, but it was somebody with good code knowledge.

That there are problems with the selection of RNG algorithm doesn't mean the RNG is crap. It simply means it could, and perhaps should, have been better than it is.

Yes, they say it is not the best algorithm available. But I don't see anyone claiming that this would have any noticable effect on the game. On the contrary, everyone seems to agree that the RNG is perfectly fine and fair for the purpose it is used. And everyone in that thread making tests get results that support this.

I don't see the point in blaming the RNG for anything. I general, blaming bad results on bad luck is the best way to make sure you don't improve. That's essentially saying "there was nothing I could do to prevent this, so no point trying to do anything differently". Instead of looking for flaws in a RNG that cannot be improved, look for flaws in your own actions that can be improved. It doesn't matter if the bad results really were a result of bad luck, the time spent thinking and complaining about it is still time away from improving your game.

 

[Pangaea]

It's not something I keep complaining about and thinking about, but this thread kind of took off in that direction. Believe you me, I'm trying to improve my game and am playing the HoF now, so it's not about that. And like I wrote above, saying the RNG could be better and has flaws doesn't mean it's crap, it simply means that it could be better.

 

[elitetroops]

Yeah sorry, that might have come off a bit harsh. Wasn't really directed at you, was more a general comment on how I feel about the issue. I know you are working hard to improve your game and you are doing a very good job at it. You're becoming quite a beast!

Back in the days I was quite active at some poker forums and you wouldn't believe how much complaining there was about the RNG. Every day people would sign up only to complain about their insanely bad luck and/or how the poker sites were rigged against them. All of them convinced that bad luck was the only thing stopping them from becoming rich, because they were clearly much better than everyone else, as they had watched a few episodes of high stakes poker on TV... Since then I guess I'm a bit overly allergic to complaints about RNGs and bad luck.

 

[FlorbFnarb]

If you're allergic to complaints about RNG, go to the forums for some game like World of Tanks or War Thunder. People who are unimaginably bad will spend hours arguing that they're really not bad players, the game is just rigged to make them lose.

Fortunately I haven't seen that here, really.

 

[elmurcis]

If you think you get unlucky, think about real world RNG in some battles when some hurricane can destroy 90% of your fleet in single turn... Or you can get ammunition from wrong caliber and be sitting duck even against archers...or there can be navigation errors and your superpowerfull Modern Armor can get in swamp where few Spearman kill all crew

 

[SeedenGreedens]

If you think you get unlucky, think about real world RNG in some battles when some hurricane can destroy 90% of your fleet in single turn... Or you can get ammunition from wrong caliber and be sitting duck even against archers...or there can be navigation errors and your superpowerfull Modern Armor can get in swamp where few Spearman kill all crew

You must be talking about wooden fleets and smaller modern boats.

 

[FlorbFnarb]

You must be talking about wooden fleets and smaller modern boats.

I think he was referring to three Mongol invasions of Japan failing due to bad sea weather.

 

[reddishrecue]

I think he was referring to three Mongol invasions of Japan failing due to bad sea weather.

What era did the three Mongol invasions occur?

 

[FlorbFnarb]

Long time ago. Medieval times for Japan for sure. Can't say for sure if it was late Middle Ages or early Renaissance in Europe. I'd have to look it up.

Back in Wooden Boat Times for sure. Although, to be honest, truly crap weather is going to make a honest-to-God amphibious invasion essentially impossible even today. Landing boats or hovercraft can't handle rough seas at all. I doubt even SEALs or their equivalent in other services or nations would be particularly keen to be inserted by a sub in monsoon seas...if they could even find a sub skipper willing to surface in such seas.

 

[reddishrecue]

Long time ago. Medieval times for Japan for sure. Can't say for sure if it was late Middle Ages or early Renaissance in Europe. I'd have to look it up.

Back in Wooden Boat Times for sure. Although, to be honest, truly crap weather is going to make a honest-to-God amphibious invasion essentially impossible even today. Landing boats or hovercraft can't handle rough seas at all. I doubt even SEALs or their equivalent in other services or nations would be particularly keen to be inserted by a sub in monsoon seas...if they could even find a sub skipper willing to surface in such seas.

Transports that seals need to land could get destroyed by submarines. What we don't know is if modern day transports get to survive in high typhoon weather at sea. Most engineers will obviously promise you that their transports won't sink.

 

[Lennier]

I think he was referring to three Mongol invasions of Japan failing due to bad sea weather.

Hmm...and I was thinking of the Spanish Armada.

Transports that seals need to land could get destroyed by submarines. What we don't know is if modern day transports get to survive in high typhoon weather at sea. Most engineers will obviously promise you that their transports won't sink.

Well, we're better at forecasting hurricanes and getting ships away from them now than in the age of sail.