Invincible Spearmen

[Zachriel]

Originally posted by History_Buff
Wheather, what the hell, im definately missing something.

Tee hee. Weather is not represented in Civ3. However individual battles are resolved with the use of the randomizer. This accounts for the lost nail in the horse's shoe.

For want of a nail
the shoe was lost.
For want of a shoe
the horse was lost.
For want of a horse
the rider was lost.
For want of a rider
the battle was lost.
For want of a battle
the kingdom was lost.
And all for the want
of a horseshoe nail.

The randomizer also accounts for the weather at Agincourt, and the Divine Wind that stopped the Khan's invasion of Japan. It accounts for the misinterpretation of orders in the Crimean War. It accounts for an exploded topedo on your submarine, and the suspicious explosion on the U.S.S. Maine. It accounts for the unexpected fortitude of the 300 Spartans, and the extraordinary good luck of Sgt. York.

 

[TheNiceOne]

Originally posted by Tassadar
Artificial random generator just doest produce random number. Anyone saying that civ 3 random generator work ok is wrong, because informatic program are not able to produce perfect random number. Why do you think the lotery use numeroted ball for the winning combination ? because a.i. random generator are not truly random. The programm try to produce random number but its not the case, talk to a proffesional in informatic-statistic and he ll tell you same thing.

The civ3 random generator doesn't produce "perfect" random numbers, that's true, but stating that it doesn't work ok is much too harsh. If that's true, then no computer random generator works ok.
A computer random generator is not truly random, but the work on this has come so far that the numbers generated simulates real random numbers very closely.
Since it has been tested that the civ3 generates results close to real random numbers I don't see why you mean it doesn't work ok?
Or do you think that we all should have 100 numerated balls and a lottery machine and write the results into the computer?

 

[TheNiceOne]

Originally posted by Zachriel

The randomizer also accounts for the weather at Agincourt, and the Divine Wind that stopped the Khan's invasion of Japan. It accounts for the misinterpretation of orders in the Crimean War. It accounts for an exploded topedo on your submarine, and the suspicious explosion on the U.S.S. Maine. It accounts for the unexpected fortitude of the 300 Spartans, and the extraordinary good luck of Sgt. York.

 

[etj4Eagle]

Originally posted by TheNiceOne

The civ3 random generator doesn't produce "perfect" random numbers, that's true, but stating that it doesn't work ok is much too harsh. If that's true, then no computer random generator works ok.
A computer random generator is not truly random, but the work on this has come so far that the numbers generated simulates real random numbers very closely.
Since it has been tested that the civ3 generates results close to real random numbers I don't see why you mean it doesn't work ok?
Or do you think that we all should have 100 numerated balls and a lottery machine and write the results into the computer?

Yep there was also a thread a while back where some one had posted results of hacking into the game and doing calls and the random number generator or something to that effect. And from this he was able to run some stat anal routines on the generator and concluded that it was passing all the standard validity tests. Which means that it was showing the proper distribution and all.

My tests were just to verify that the AI got no special hidden bonuses in combat. Which it does not.

 

[Lt. 'Killer' M.]

Originally posted by Zachriel


Sorry, not trying to be smug, but after thousands of combats, I just don't see the problem with the randomizer. I usually know in advance of combat whether I will win easily, have a slugfest with lots of casualties, or lose.

When you fight at 2-1 odds, expect a lot of casualties, i.e. you will need to have lots of reinforcements to cover your losses. Try to avoid these situations to better your chances, or use bombardment where appropriate. In earlier war games, they had three combat results, win, lose and exchange. Exchange is where each side would lose like numbers of units. The one left standing would win the combat, but of course, would be substantially reduced in strength.

Hollywood! Cool. Every once in a while, a Sgt. York unit will win repeatedly against all odds. Sgt. York was a Hollywood movie, with Gary Cooper -- but then again, he was a real soldier, too.

thanx for the wise ans smug advice, Zachriel, but how the hell do you think I win????? Still, it sometimes sucks!

 

[Zachriel]

Originally posted by Killer
thanx for the wise ans smug advice, Zachriel, but how the hell do you think I win????? Still, it sometimes sucks!

Yup, "War is Hell."

General William Tecumseh Sherman
to the Mayor and Councilmen of Atlanta

http://www.rjgeib.com/thoughts/sherman/sherman-to-burn-atlanta.html

 

[bengood22]

You must not have been playing the Americans. When I was in the U.S. Marine Corps, I would have had no trouble picking off a spearmen from 500 yards with my M-16

Seriously, though, I tend to expect unfavorable results in combat. If I don't see my Marines failing against the AI I tend to think something is wrong with the game! Computers are supposed to cheat, aren't they?

 

[Akka]

Originally posted by bengood22
Computers are supposed to cheat, aren't they?

Don't panic, they do

 

[Sanaz]

When I watch a battle, it seems that some special units sometimes kick into a faster pace after the first strike is made (Immortal's sword slashing and Jaguar Warrior's funny fast punching come to mind). Maybe this is an undocumented advantage? They seem to do better when moving faster.

Also, it seems that whoever strikes first (in the graphics, not just the attacker) almost always lands the first hit successfully, regardless of advantage/disadvantage. Maybe there are a few extra variables in battles? Does anyone know if speed is a variable in the actual fight (not just for movement rate and retreat ability, but in landing more hits in a battle?

BTW, I have never seen particularly strange battle results - I think I'm equally lucky and unlucky in battle, with most results being fairly expected.

 

[cracker]

I am still looking to some sort of round off coding error to account for the anomolous results that we see far too often.

I do believe that the RNG is probably standard computer fodder and that with 100 or 200 engagements the results may begin to approach parity.

But, round off errors and unresolved integer conflicts are everywhere in the game. They are so common that it is hard to believe they would not occur in the combat system as well.

The major integer round off errors that result in undocumented increases in waste, corruption, and food loss begin from moment one in the game:

How much "wealth" is produced per turn in a city that produces 2 shields but is set to "wealth" in the early game by the all knowing governor?

How many shields does it cost to produce 4 workers at 10 sheilds each from a city that is producing 42 shields per turn?

How many food units does it take to grow a city from pop 3 to 6 when the city has a granary and is producing 8 food units per turn due to wheat bonused flood plains?

I have also not seen the stats on the CIV3 RNG and there can be a very big difference between a statisticallt normal RNG where the results are a bell shaped curve versus a uniform random number generator where every outcome has essentially an equal probability.

The round off error would be most significant in match ups that have a more significant difference in the opposing A and D values.

We encounter these round off errors in upwards of 90% of the simulation programs for population genetics and real world event simulations so there is no reason to expect that they would not be built into CIV3 when there is so much evidence to indicate that integer round off issues were totally ignored in almost every aspect of the program development.

 

[Sanaz]

Originally posted by cracker
I have also not seen the stats on the CIV3 RNG and there can be a very big difference between a statisticallt normal RNG where the results are a bell shaped curve versus a uniform random number generator where every outcome has essentially an equal probability.

Is this like the difference between a double dice roll, where the chance for a total value of 1-12 is bell shaped, vs. a 12-sided die roll, where the values from 1-12 each have an equal chance of occurrence? This can leave a lot of room for interpretation for what "random" really means in a battle sequence. There must be ways to weight the results towards a steep curve, or no curve, right?

Thanks for pointing out all of the round-off issues in the game (shields, food, science, etc...). I had never thought much about it in hard numbers - I usually just move my working citizens and science slider around to optimize the best I can for minimal waste.

 

[Loopy]

Originally posted by cracker
I am still looking to some sort of round off coding error to account for the anomolous results that we see far too often.

But, round off errors and unresolved integer conflicts are everywhere in the game. They are so common that it is hard to believe they would not occur in the combat system as well.

You'll be looking a long time 'cause they don't exist. You're confusing round off error with game play design.

The major integer round off errors that result in undocumented increases in waste, corruption, and food loss begin from moment one in the game:

How much "wealth" is produced per turn in a city that produces 2 shields but is set to "wealth" in the early game by the all knowing governor?

The wealth setting is designed to produce much less wealth than previous incarnations. It is definately not a 1:1 relationship. No proof or hint of roundoff error here.

How many shields does it cost to produce 4 workers at 10 sheilds each from a city that is producing 42 shields per turn?

It costs 10 shields. You can only produce one unit a turn. No roundoff error here.

How many food units does it take to grow a city from pop 3 to 6 when the city has a granary and is producing 8 food units per turn due to wheat bonused flood plains?

Again, you can only increase in size once per turn. After increasing in size, you start from scratch. No roundoff error here.

I have also not seen the stats on the CIV3 RNG and there can be a very big difference between a statisticallt normal RNG where the results are a bell shaped curve versus a uniform random number generator where every outcome has essentially an equal probability.

Actually this isn't the case. You can take a standard normal distribution function and create a uniform distribution by taking the thousandth place from your randomly generated normal.

We encounter these round off errors in upwards of 90% of the simulation programs for population genetics and real world event simulations so there is no reason to expect that they would not be built into CIV3 when there is so much evidence to indicate that integer round off issues were totally ignored in almost every aspect of the program development.

There is reason to expect that they wouldn't be built into Civ3. The combat RNG is relatively simple, dealing spitting out one of 1024 integer. The threshold calculation for combat resolution can be done at the float / double level. There's no reason why rounding takes place at all.

 

[cracker]

Loopy,

I don't think you got any of the test questions correct even though it is obvious you might have some statistical background.

Would you like to use your one "Get out of Jail Free" card to reread the posting and refocus on the issues to see if you can divine the real answers and purpose of the post.

Gameplay Design issues (or at least your interpretation of those issues) aside, building granaries cuts the required food production to support growth in half and allows the city to store excess food production against the risks of interruption by war, pollution, or global warming. So why does excess food production above the requirement for growth just get thrown away instead of stored. You say it a Game design choice, I say its just an oversight and one of many.

I think the track record so far does not support a position of "Oh, things work that way by design" as opposed to "Oh, things were designed that way because we didn't consider all the impacts of the design choices."

The Mid-game Integer round-off waste in production and growth is so high as to account for an average of 15% additional waste above and beyond the designed in corruption levels. Even intensive and expert micro management cannot defeat the waste factors because it is almost impossible to match up the production values with available units and improvements.

I should also add the "Hurry" waste factor to the list:

Example: city produces a net of 11 shields per turn and needs to produce a longbowman for 40 shields. After one turn 11 shields are in the production bin and 29 shields are required to build the unit in the next turn. 11 more shields would be produced by the city in the next turn, so only 18 shields need to be "hurried" or purchased in order to complete the unit in the next turn.

You get the same error, when switching production in midstream. The advisor notifies you that X number of shields will be wasted by switching from a more expensive unit down to a less expensive unit but ignores the full amount of the current turn's production that will be wasted on top of the simple production bin waste.

Game design choice issue --or-- lack of robust game design and coding. You can argue boths sides of the issue, but the large number of well documented actual examples supports the position that these errors are more a result of lack of big picture coding perspective that results in waste and frustration that is 2x to 3x what the game designers are actually perceiving as a valid part of the game.

I have already produced test results that validate this premise by showing that bombardment style engagements will currently target units, civilians, and improvements that cannot be hit or destroyed and thus artificially inflate the "bombardment failed" rate by as much as 25%.

There are just too many of these issues that become obvious when finally explored without dogmatic acceptance, for us not to have deep seeded suspicions that similar errors would exist in the combat resolution system when long strings of anomolous results are produced under seemingly improbable circumstances.

 

[Zachriel]

Originally posted by cracker


So why does excess food production above the requirement for growth just get thrown away instead of stored. You say it a Game design choice, I say its just an oversight and one of many.

I think the track record so far does not support a position of "Oh, things work that way by design" as opposed to "Oh, things were designed that way because we didn't consider all the impacts of the design choices."

The "lost" food and shields is the same as in Civ2, so that would suggest that it is a game design decision.

 

[Loopy]

Furthermore, all your examples deal with excess production. Since production is handled as an integer (as opposed to a float or double), there is no integer rounding. There is a very consistant rule of once the total is reached, any excess doesn't carry over -- that is, after completion you start from zero.

Granaries follow this rule perfectly. Once the granary is filled and the city grows, half of the required food production is stored (lowering the risk against pollution, lower production, etc). Consistant with other production events, here is no food above the granary stored food.

I see it as all being very consistant. You (cracker) have yet to point out any actual errors in coding. Your supposed bombardment validation is fraught with invalid assumptions. Instead of a "get out jail free card", might I suggest you take something for your logorrhea?

 

[etj4Eagle]

For those believing that the random number generator is invalid, I suggest taking a look at this thread:http://forums.civfanatics.com/showthread.php?s=&threadid=15208&perpage=20&pagenumber=2 it is Dan's combat test map thread.

In particular on the second page there is a post by hwinkels. He has hacked the game and run the DIEHARD suite of tests on the generator, and concluded that the random number generator is working perfectly fine.

At least one concept that is at issue is the number of samples required to acheive a normal distribution. Someone with more stat experience than me will have to answer this one. But one question that needs to be resolved before blaming the generator, is how many combat rounds should be required to acheive the right distribution. Since combat is only a few rounds, often only 3-5 rounds, I would expect many combats to have to occur to acheive a sufficient sample to expect a normal distribution.

Now it is very possible that civ3 should have a narrower than normal distribution or one that is designed to some how converge faster than normal. Or alternavitely the combat code could be modified to treat each round as 3 or 5 comparrisons, since the time delay there is displaying the graphics. Ie instead of taking just one random number and comparing it to the defender's threshold value, you take 5 numbers and only if three are equal to or greater than the threshold does the attacker win the round.


And additionally as some others have mentioned at times, very likely Firaxis is just making a call to a standard random number generator in the library of what ever development langauge they are using.

 

[royfurr]

Sgt. York WAS a soldier, and a very brave and effective one, too.

I think that the possible problem with the RNG and the fact that some see it and some don't may be related to a hardware issue. That is, I speculate that, the way it (the RN generation process) works, some facet of the hardware, with some types of machines, may cause the RNG to be more "stringy" in some case and not in others. Perhaps it is a combination of the chip in the machine, the motherboard integration, etc etc. Probably a combination of several or many things. So Zacherial sees truely random results and Killer doesn't.

I agree that in many cases it is a human poblem as well, with people both remembering only the bad and not the good. Also, with poor expectations of when "bad" is going to happen. This means that the contraness of the combat system is not as bad as many claim. Still, I think there IS at least somewhat of a problem.

Finally, on a different but related issue, I disagree with Firaxis design decision to essentially allow "spearman to defeat tanks" (so to speak) in an attempt to maintain chances for technology outdated civs. Yes yes yes I agree that a tech lead shouldn't automatically mean a victory, (in the game or in a single battle) but I think that Firaxiz went too far with their solution. Somewhere about between those two cases would have worked better, I believe.

This is worse in the naval aspects of the game, but naval warfare and strategy is so totally frelled that poor combat resolution is the least of the problems.

Well, maybe 3 cents worth ...


Civ on.

 

[Loopy]

Actually, IIRC the combat RNG and resolution works as follows. The odds of winning a round of combat are calculated (warrior attacking unfortified warrior on plains: (1/(1+1.1) = 47.62%). The cutoff for victory is calculated (1024*47.62%=487.62). A random integer is drawn from 1-1024 (or possibly 0-1023). If it is below the cutoff, the attacking warrior wins. If above, the defender wins.

Note that while I use rounding in my example, there's no need to do so in code. IF RNG(1,1024) < (1024*A/(A+(D*modifiers))) THEN attacker wins round, ELSE defender wins round.

You'd want to cast the integer from the RNG as a float, but that's trivial. Whether it's actually a "<" or a "<=" is also somewhat trivial.

The problem's not in round off error, nor is it in the RNG. It's just in people's perception of the odds and how "strong" a unit should be. Warrior vs Cavalry is not caveman vs mounted guns when figuring combat resolution -- Warrior vs Cavalry is 1 attack vs 3 defense. The odds of Warrior winning is a lot better than caveman vs mounted guns.

 

[Zachriel]

The odds are stacked against the attacker, due to the defender bonuses. Most players abandon losing positions, so they are normally attacking -- not defending. Therefore combat, as a whole, appears weighted in the AI's favor.

Could this be the problem?

 

[Sanaz]

One other factor may be whether there is a linear chance of getting a random value of 1, 512, or 1024, or if there is a bell shaped distribution, where there is a better chance of receiving the middle numbers (i.e. 512) than the extremes (either 0 or 1024). The distribution could have an effect on the final results of the battle calculations.