Ai / dice rolling combat?

[Blake]

HAX .

I knew it was only a matter of time before I'd lose a 100% battle.

It's a Guerilla2+Drill1 Archer fortified on a forested hill.

[Hauptman]

I took a few SS's of some lost 100% odds fights.

However they apear to be on the house computer, not this one.


This is definatly happening TOO frequently to be just honest RNG luck. Honestly something is not right with 1.54 combat.

[okblacke]

Originally posted by Blake
HAX .

I knew it was only a matter of time before I'd lose a 100% battle.

It's a Guerilla2+Drill1 Archer fortified on a forested hill.

I don't get the math there. You're hit for 12, but in the next line it shows you with 4 damage?

[Blake]

Hmmm, interesting. I've observed it cutting damage to avoid a strong unit 1-shotting a weak unit (all battles seem to be required to go for 2 rounds), but I havn't noticed it happen in "fair" fights. I think it's related to the no 1-shot code anyway.

Edit: Something else is that first strikes mean the combat odds aren't always correct. I've seen things like 105% odds (that was on a riflemen upgraded from a Drill longbow, it might be drill on units with no FS). But I think in this case 8.25 vs 3 is a true 100% battle.

The bizzare thing about such random combat is it would actually be broken if extreme upset combats never happened. I like to call this sort of feature "Broken as Intended" (like "Working as Intended").

[cal_01]

Okay, I tested the following scenario tonight:

Archers in cities with and without the city defense promotion. Warriors attacking.

Strangely enough, my first two battles resulted in victories even when it was 99%+ or so against me.

Moreover, combat seems a bit weird in that streaks occur for both players, though I personally noticed that the AI tends to get a lot more streaks than the player.

Another interesting thing I noticed is that First Strikes for Archers don't always happen; sometimes my warriors would get the first hit.

[weepy]

Look like Unit strength are ignored.

line up some unit of different strength.
Save the game, start attacking from the least powerful,
Reload the game, now start attacking from the most powerful

Will notice the combat result are the same.
even the damage done and taken.

Look like the sequence of attack is important, not the strength of unit.

[Yosh]

Interesting cal...

I wonder if it's something in the FS that is screwing up the calcs?

I took a SS everytime it happened to me in this current game I'm playing -- it happened more than 10 times I would say easily; unfortunately, I've discovered that you can only have 15 SS's in your ScreenShots folder at a time, and it doesn't overwrite them! Consequently, I have 2 SS's, neither are really worth posting up without the others

It'd be interesting to hear some word from someone who is more intimate with innards of the games' take on this -- DeepO (or other beta testers), you listening?

[cal_01]

I'm betting there's a way to redraw the dialog box so that it becomes highlightable. Judging from how many of them are set up, I wouldn't be surprised if they're just standard "windows" with skins.

Or even better, judging from how it loads, I bet it loads the data/cache somewhere in memory (or even hard disk). The python code might show where it puts this data.

edit: I'm not too sure, but I should recheck whether or not the combat odds take into account the difficulty level, and hence the barbarian penalties if applicable. They show up in the combat logs, but I'm not sure if they show in the hover.

[Yosh]

I'll dig around in the python files and see if I can find a way to extract the text... I'm thinking that the combat log file is stored in a temporary XML file or something - I'll post and let ya'all know when/if I find something

[cal_01]

What I noticed about the combat logs is that there's no miss result; it's either attacker or defender that hits.

[Yosh]

hmm yea that's true - I never thought about that before!

Well, I did a lot of looking, and I found the relevant python functions that write the combat log entry, and pending problems, I think it would be possible to write an override for an event that would also output the combat log to a text file, but I'm not that familiar with the CIV python API yet... any volunteers to assist?

[cal_01]

Heh. I'd have to admit that I know nothing in terms of making Python output a text file...

[cgrecu77]

[SIZE=1] After all 0.03% is 1 in 330, that's bound to happen.

0.03% is 1 in 3300, which looks small but it's actually huge. In a regular game you probably have around 100 battles at most ( on a standard map) and sometimes you only have 10-20. Probably the average user has aroud 30-40 battles per game, so in theory this should happen only once every 100 games. Since you can't play more than one (full) game per day and Civ 4 appeared less than 3 months ago ...

Still, 1 in 3300 doesn't mean it takes 3300 to happen. It could be that it happens the first time and it doesn't happen for 10000 more battles, or it could be 3 in a row, although chances of that happening are about as big as winning the lottery ...

[cgrecu77]

Originally posted by cal_01
Has anyone checked to see (or if it's even possible) if the random number seed is random "enough" in a small sample?

a random number seed that is random enough over a large sample is by definition a correct one. The phrase "random "enough" in a small sample?" is mathematically incorrect, statistics only work on large sets. If you mean that the random generator generates ONLY streaks that are biased to one side or the other (like you win 5 in a row, then the computer wins 5 in a row and so on) then you might have a point, but I don't see that happening in my games ...

One thing I noticed though, and I'm probably wrong, is that when attacking with a large stack I'll usually lose the first unit (usually a siege weapon) even if the odds are in my favor. This might indicate that the random generated odds are slightly altered programatically to help the AI (on higher difficulty levels) or the human player (on lower levels).

[cal_01]

Originally posted by cgrecu77


a random number seed that is random enough over a large sample is by definition a correct one. The phrase "random "enough" in a small sample?" is mathematically incorrect, statistics only work on large sets. If you mean that the random generator generates ONLY streaks that are biased to one side or the other (like you win 5 in a row, then the computer wins 5 in a row and so on) then you might have a point, but I don't see that happening in my games ...

One thing I noticed though, and I'm probably wrong, is that when attacking with a large stack I'll usually lose the first unit (usually a siege weapon) even if the odds are in my favor. This might indicate that the random generated odds are slightly altered programatically to help the AI (on higher difficulty levels) or the human player (on lower levels).

The point I was getting at is that combat tends to be too "streaky", and hence the unrandomness.