Question about win % in 1.52

[Apocalypse Cow]

In 1.52, when you go to attack, it not only shows the numbers, but the probability of win percentage. I'm not going to get into the whole "This shouldn't happen"/"Superunits"/"gunship vs. knight" debate, nor am I getting into the statistics quagmire that is going on in the other recent thread.

This is purely a question about numbers shown on the screen. Does anyone know if the odds show "100.0%" does that REALLY mean 100%, or are the odds simply greater than 99.95%, and are rounded up to 100.0?

[Provost Harrison]

That's a very early Longbowman

[Aro]

Indeed.

[Veroporo]

Try it out. A few thousand battles should do it.

[Arrian]

My understanding is that there is something funky in the formula as regards first strikes. So in the case of units w/first strikes, the odds may not be quite right. It is also my understanding that this is going to be corrected. Still, I'd say you're gonna win that battle

-Arrian

[DeepO]

Originally posted by Arrian
My understanding is that there is something funky in the formula as regards first strikes. So in the case of units w/first strikes, the odds may not be quite right. It is also my understanding that this is going to be corrected. Still, I'd say you're gonna win that battle

-Arrian

Woaw, one second. There is a bug indeed, but it happens so rarely I would certainly not say all first strikes calculations are off! The bug only happens when both units have first strikes, and have chance of first strikes (so they display not '1' like here, but both display '1-2'). This is so rare, you will nearly never run into the situation where the bug happens. I have not seen it during regular play, and you can bet I'm looking for it...

And yes, it will be corrected. It probably already is.

As to how rounding works: if it says 100.0%, it means the result is rounded from 99.95% upwards. It does not say your unit will win in all cases, just that you will have a 1999 out of 2000 chance of winning.

To be 100% correct: internally, some minor roundings occur as well. This might lead to 99.93% being rounded to 100.0% as well. The rounding can't introduce faults higher than 0.1% though, so when CIV displays 100.0%, it always is at least 99.9%

I have not seen 100.1% chances yet, however in theory it is possible. 100.2% should never be spotted.

[Apocalypse Cow]

Originally posted by DeepO
As to how rounding works: if it says 100.0%, it means the result is rounded from 99.95% upwards. It does not say your unit will win in all cases, just that you will have a 1999 out of 2000 chance of winning.

Thank you, this was exactly what I was looking for. I didn't know if they made a change in 1.52 to appease the "this shouldn't be happening" camp by actually making it impossible to lose certain battles if the combat percentage was high enough.


Hmm, that actually might not be a bad mod.

[Arrian]

Doh, sorry DeepO! I must've misunderstood... I'm not a math wiz so I tend to "skim" when reading those math-heavy threads about combat odds.

I stand corrected.

-Arrian

[DeepO]

Originally posted by Apocalypse Cow


Thank you, this was exactly what I was looking for. I didn't know if they made a change in 1.52 to appease the "this shouldn't be happening" camp by actually making it impossible to lose certain battles if the combat percentage was high enough.

I would strongly oppose such a change to the system. It will never end: everybody has a different limit on where they want to get 100% sure, instead of 99.9% sure. Spear beats tank is a very serious concern for Firaxis, but the current system should make it extremely unlikely to offend people, without hardcoding countermeasures.

There is one thing, though... nearly all code uses integer math. I couldn't for all of the combat calculator code, but it comes close. But the RNGs use it as well. There is a good reason why the result is capped at 1 decimal (well, 3 to be precise), anything lower is not really valid. There is no point of analysing the RNGs, and using their roundings in the formula as well as they would only start to show after the 4th or 5th decimal, but in theory this formula could still be improved upon. When the odds are so bad it dips below this RNG-rounding-threshold, the results might be an exact 0%, instead of a rounded 0.0%. This calculator will not take that into account... not much point to it.

DeepO

[DeepO]

Originally posted by Arrian
Doh, sorry DeepO! I must've misunderstood... I'm not a math wiz so I tend to "skim" when reading those math-heavy threads about combat odds.

I stand corrected.

-Arrian

Arrian, it's not about correcting you, on the contrary. However, I thought you might have misunderstood my explanation, and decided not to use the calculator with first strikes... that would be a shame. I mean it that you will very rarely see the bug happening, you have to be spear-beats-tank lucky to spot it

For those more puzzled by this: if you see a result of e.g. 120%, you've found the bug. Anything else, and assume it's correct.

DeepO

[Arrian]

I'll always use the odds calculator (hey, I even found the older, more flawed one useful).

I apologized b/c I don't like spreading misinformation. I like to give people the proper info.

-Arrian

[DeepO]

Originally posted by Arrian
I apologized b/c I don't like spreading misinformation. I like to give people the proper info.

Me too... which is why I proposed to Soren to code the calculator for him. Which is also why this is a bit touchy: as a former programmer I know all code is susceptible to bugs, but the one thing I add to CIV should have been perfect

Well... one thing might not be totally correct... I'm enjoying a lot of the smaller things I lobbied to get in. But this calculator certainly is the most visible one. It should have been perfect from the get-go, instead of needing a patch to fix.

DeepO

[Enigma_Nova]

The odds seem to be working well.
Once in my Stack of Doom rampages across the world, a Catapult won with a 2% chance of victory. I had probably attacked at least 25 times, but not 100 times, with catapults by that point in time - so 1 cat victory seemed reasonable.

[Jonny]

Originally posted by DeepO
For those more puzzled by this: if you see a result of e.g. 120%, you've found the bug. Anything else, and assume it's correct.

DeepO

So that explains why I saw a 128.8% yesterday. Forgot to grab a screenie though. (But at least I won the battle; I would have been thoroughly annoyed if I hadn't.)

[seoman]

Originally posted by DeepO

As to how rounding works: if it says 100.0%, it means the result is rounded from 99.95% upwards. It does not say your unit will win in all cases, just that you will have a 1999 out of 2000 chance of winning.

Personally, I don't think an odds calculator should ever display 100.0% chance of winning (similarly for 0%), as no combat in this game is certain (to my knowledge). I would suggest adding displayed bounds of 99.9 and 0.1. Of course, having those would have hidden the 120% error, so your milage may vary.