PieceOfMind said:
Agreed. Take it up with me.
The mod is called Advanced Combat Odds and I can assure you all it is correct, including its dealing with first strikes, with a couple of nuances...
The first one being that as opposed to the vanilla Civ odds calculator, in ACO percentages are rounded. This means 100.0% as reported could in fact be between 99.95% and 100.0%. ACO will actually tell you explicitly if you have exactly 0 chance of losing a combat but that is a very very rare circumstance (like attacking severely wounded scouts after nukes with modern armor) so I doubt anyone has even seen it yet.
The second nuance: barbarian odds can be misleading depending on whether barbarian "secret combat mechanics" are set to hidden or not. The default behaviour in ACO is to reveal these but it's possible for a modder to hide them again.
I accept your assurance the odds calculator isn't giving false information.
... 1) These anomalies
are against barbarians
... ... (I'll read up on that nuance shortly).
... 2) The rounding is interesting to note,
... ... ... but it is not large enough to be significant
... ... ... (It could knock another notch off the 2/12 gap making it a coin flip
**,
... ... ... ... but the actual gap was larger).
**coin flip is an idiom here and not a specific reference to 1:1 odds.
Please post a screenshot of the odds when hovering over the unit to attack. Alternatively, post some screenshots of the combat log after the 12 combats you are reporting.
No screenshots.
I remember the combat log was more informative than the odds display.
I find it extremely unlikely you'd lost 12 combats in a row against a single defender with 62% odds, which is why I want to see proof.
Note that as someone else pointed out, reloading and retrying the combat does not count if you are playing under the option "no new random seed on reload".
I don't reload games,
... except between sessions,
... or when I'm studying a new teching path.
In my defense,
... I don't normally waste a stack after a certain percentage of losses,
... but I wanted to understand where the RNG is a factor.
I didn't see any difference in the combat log for these attacks.
... They were data clones of each other.
... This is why I thought it might be a first strike issue.
It is worth mentioning, and I went into this in detail in t
his thread (Combat Odds and RNG), that the RNG and odds calculator are
two completely separate things.
Thanks for the link. I will read it shortly.
RE:
Combat Odds vs. RNG
...
Q: Could there be a mathematical anomaly ...
... ... ?) between Civ IV's combat calculations and ACO
... ... ?) something new in combat calculations with 3.19
I know, I know ...
screenshots.
RD-BH/PieceOfMind said:
Did the RNG roll the right number 12 times in a row? Can you be more clear about the meaning of this question? What is the "right number"?
Darn,
... I was hoping you would clear this up for me.
I'm not sure where the RNG comes into the equation,
... or even how many times a new number is generated.
Sometimes it feels like only one random number is ever associated with a whole stack of combats ... uh, oh.
... I'll have to get back to you on this.
ParanoidDelusion said:
... uh, I have a sickening feeling right now.
What RNG does Civ IV use? If they are using a stock routine in a dll on my machine (either WinXP or C/C++) then I may be the cause of my own problem.
Nah! That's just crazy ... end paranoid delusion.
RD-BH/PieceOfMind said:
I despise RNGs in general, because of their predisposition to crossing magnitude boundaries. What are these "magnitude boundaries"? EDIT...Ok I see what you mean by magnitude boundaries now. I think you could rephrase the problem as - floating point operations (as done by a machine) are more likely to produce errors as you reach more extreme orders of magnitude. That said, I am fairly confident the floating point calculations done in ACO are stable - that is, they are not overly susceptible to things like catastrophic cancellation and the like. The calculations are kinda like Taylor series expansions - as you add to a number, smaller and smaller numbers, the last very small numbers might have less and less effect on the overall sum, but the overall sum is very accurate in its order of magnitude. e.g. 100+10+1+0.1+0.01+... might get truncated as 111.11 (this is a simplification - floating point operations are much more accurate than that) but that is good enough for many uses.
Excellent summation!
... I knew there were better words out there.
I'm too old to learn new math,
... thanks for looking past the words,
... ... and seeing my meaning
... (Okay, Taylor series aren't
new math ... he's been dead a long time).
... Another example:
... ... Infinite sum (Euler?) of cosine and sine
... ... ... Stop calculating at the wrong magnitude and values can vary significantly.
However
... (isn't there always a however with me
),
... I'm not talking about the FPC.
... The shift in magnitude is ...
... ... 1) between the reported odds and the observed odds
... ... 2) in very specific combat situations.
... ... 3) frequent enough I noticed.
... ... 4) not frequent enough I was impelled to write a bug report.
RE:
magnitude boundaries (orders of magnitude, scale, etc)
... They aren't always Log10,
... ... the base depends on the precision expected from the equation/model.
... They don't
prove errors.
... They suggest the
possibility of error.
... Combat percentages (0..100) lend themselves to Log10.
RE:
RNG
With assurance the combat odds are accurate,
... this only leaves the RNG.
There is no upside to pursuing this issue.
... If my paranoid delusion turns out to be real,
... ... I've done this to myself.
... If Civ IV is using its own routines,
... ... then I have to live with the occassional anomaly.
TODO:
... 1) Read
PieceOfMind's article.
... ... I'm sure I've read it before,
... ... ... though not with my current questions in mind.
... 2) Grab a screenshot the next time I see the anomaly.
... ... This may be awhile,
... ... ... I am currently playing Deity games with no barbs and an extra AI.
... ... ... ... (Barbs are too random on Deity,
... ... ... ... ... they might as well be goody huts)
... ... ... IIRC, the last time I noticed this was on immortal.
... ... ... Who knows,
... ... ... ... maybe I'll get a bee in my bonnet and stage an attack or two.
... 3) Reconsider using so many ellipses (...) for text formatting.
... 4)
Have FUN 8)