26-01-2003 00:33									#31
	Support Apolyton, buy Call to Power 2

i wonder how the ai gets to be so good, its all thecheats it has me thinks....

silly game.....

	26-01-2003 01:01									#32
	Support Apolyton, buy Civilization: The Boardgame

quote: Originally posted by War4ever not likely, i am known to have a high sense of wit and to understand the understandable

In that case you can explain Heisenbergs's uncertainty principle (and hence my joke) to BM.

	26-01-2003 01:23									#33
	Increase Your PM Length

quote: Originally posted by SCG I tried a quick test and after making peace in the first contact, tried breaking the treaty at the start of the next turn - i lost 35 of 40 attacks warrior on warrior, attacking with green warrior onto another green warrior in the process of fortifying in the desert.

might find this interesting, although sample size of 40 isn't all that big.
I just loaded the same game in my "fresh from the mailbox" version of MGE and my warrior only lost 24 of 40

thats a 2/5 victory rate vs a 1/8 victory rate
I'm too lazy to try and do those 1000 repeats some of the real testers use to stamp out statistical annomolies

	26-01-2003 01:49									#34
	Lose 30 kilos (of popups)

you said you made peace first, you dont need peace to have the surprise sneak attack bonus...usually what happens is someone meets with 1mp left and just attacks, or the next turn he is attacked, we don't do peace in our games

	26-01-2003 04:32									#35
	Avatar Enlargement: We've got the solution

Yeah.. instead of making peace and trying it... experiment with attacking the FIRST time you run across somebody else's unit. That could be why your "tests" have shown nothing

	26-01-2003 04:35									#36
	Support Apolyton, buy GURPS/ Alpha Centauri

now he has to get through the bureaucratic red tape first

	26-01-2003 14:26									#37
	Support Apolyton, buy Alpha Centauri

quote: Originally posted by Bloody Monk I remember your test results, but I don't remember how you set up the tests. Were each of the tests done as a one-off in actual game conditions?? Or did you do reloads, or use cheat mode??

I didn't find a post with my results. Anyway I changed stats in rules.txt to high values so that fortuity is low and so I suppose I didn't need the cheat mode or reloads...

This is a result of Marquis:
http://apolyton.net/forums/showthre...20&pagenumber=6

	26-01-2003 14:29									#38
	Browse Apolyton AD-FREE

Ming,you could test it in MP too. Just change hitpoint values to 10 (100) in rules.txt, firepower to 1 and attack and defense to 20. Fortuity will be very low.

	26-01-2003 15:08									#39
	Inflate your Upload Space

I think we need some definitions of terms here - to we SP players 'sneak attack' is something the AI does to us in violation of a peace treaty and we get the pop-up "Sneak attack by French forces".

It seems to me that what is being discussed by our enlightened MPing brethren is something different - in particular an attack launched on the very first (pre-diplomacy) opportunity to attack someone. As I cannot recall a single instance when the AI has refused to take advantage of such an opportunity to insist upon diplomacy (even if it is my turn!) a doubt any of us have eever seen the phenomena you guys describe - and it would be hard if not impossible to duplicate in a SP laboratory.

SG[1]

	27-01-2003 19:09									#40
	Support Apolyton buy from Amazon

As alluded to by my comrades herem it's probably like the barb bonus at raging level. The barbs get the 50% attacking bonus against human players, but not against the ai. I'm sure the sneak attack bonus is the same. Only good against human players. This would explain the testing results.

	27-01-2003 23:03									#41
	Got spare money?

That should be testable, though, even by one person hot-seating. Could this be related to the bonus human-civ units get against Barb Archers?

	27-01-2003 23:06									#42
	Support Apolyton, buy Alpha Centauri

quote: As alluded to by my comrades herem it's probably like the barb bonus at raging level. The barbs get the 50% attacking bonus against human players, but not against the ai. I'm sure the sneak attack bonus is the same. Only good against human players. This would explain the testing results.

rah, do you want to say the bonus doesn't apply in King and lower levels?

	27-01-2003 23:17									#43
	Support Apolyton buy from Amazon

No, what we said was that the bonus maybe wouldn't show up in a game at any SP level.......it could be particular to MP (or at least, only seen *against* humans). As Rah says, there is precedent for bonuses being calculated differently for humans and AI.

	27-01-2003 23:25									#44
	Support Apolyton or Terrorists Win

quote: Originally posted by SlowThinker rah, do you want to say the bonus doesn't apply in King and lower levels?

While that might indeed be true... and we wouldn't know because the majority of our games are played on Deity... Rah was raising an important but different point.

Like the Barb attacking bonus that only is applied to barb attacks on human players and not AI civs... Maybe that is indeed the case here. Maybe only human surprise attacks on human civs have the added advantage. And none of the previous tests would show that, because they were all done against an AI civ.

*and what a pain that would be to test... having to quit the game and restart it MANY MANY times to get a reasonable data set to work with*


added later... a classic example of cross posting... right on Dr Spike

	28-01-2003 01:11									#45
	Help yourself to an AD-FREE life

Even if we tested it in MP and discovered it was true...

Then there would still be the question of whether the AI gets a surprise bonus against Human players, or is it just humans on humans. If you think testing humans on humans is tedious..................well, maybe in wouldn't be that hard because you could test it in sp mode.

RAH
And YES, you never know, just like the barb bonus is only at the upper level of barbs, maybe the attack bonus against humans is only active in Deity games. But as ming says, I don't care since we only play deity.

	28-01-2003 05:33									#46
	Support Apolyton, buy GURPS/ Alpha Centauri

after reading the confusion earlier, i wondered if the ai got a sneak attack bonus again and again against the human player, especially since it almost always wins... not that it picks a tough target usually....

but this isn't the bonus were talking about.....

simply put, first contact , no peace, attack first, get the sneak attack bonus, win against a unit you would not likely beat.... surely this is testable?????

	28-01-2003 18:12									#47
	Browse Apolyton AD-FREE

quote: Originally posted by War4ever after reading the confusion earlier, i wondered if the ai got a sneak attack bonus again and again against the human player, especially since it almost always wins... not that it picks a tough target usually.... but this isn't the bonus were talking about..... simply put, first contact , no peace, attack first, get the sneak attack bonus, win against a unit you would not likely beat.... surely this is testable?????

We are talking about the same bonus. It's just that in our games, we don't make peace then sneak again, so we only experience it once with each player. I would expect to get the bonus every time there is a sneak attack.

	29-01-2003 06:31									#48
	Remove this text

hm, i can't remember the last time we had peace in our games either, i wonder if there were multiple sneak attack bonuses then.....

do you remember the golden days rah?

	29-01-2003 19:01									#49
	Support Apolyton, buy GURPS/ Alpha Centauri

I'm betting that the AI gets the sneak attack bonus every time it breaks a treaty against a human, since it's such a popular thing for it to do.

	30-01-2003 01:33									#50
	Enter the AD-FREE zone

If it is just breaking peace treaties we should have seen it before, unless it is only AI-on-human. If it is "first contact" only, that would limit how often we see it, but I take it from the discussion that others have seen it at later stages. Perhaps it is one of those extra things the AI can pull off in Deity games, like getting a WoW one turn after getting the tech...

	30-01-2003 01:55									#51
	Help yourself to an AD-FREE life

well the ai has the ability to combine all its shields working on wonders IIRC so the one turn wonder isn't so hard for it to do in the later stages....

i also wonder if the ai can buy shilds at a 1-1 pace?

	30-01-2003 02:49									#52
	Support Apolyton, buy Civilization: The Boardgame

War4, since no one seems to be jumping up to volunteer to test it and it take two to test it in MP, I propose that we do it prior to one of our normal weekend games while we're waiting for people to show up. Give us a chance to contribute to the strat community.

	30-01-2003 02:55									#53
	Suffering from ads?

darn nonresponding site, DP Honest.

Last edited by rah on 30-01-2003 at 04:23

	30-01-2003 12:57									#54
	Support Apolyton buy from Amazon

quote: Originally posted by rah War4, since no one seems to be jumping up to volunteer to test it and it take two to test it in MP, I propose that we do it prior to one of our normal weekend games while we're waiting for people to show up. Give us a chance to contribute to the strat community.

That would be great, rah -- finally lay one of the urban myths (or features of the game) to rest...

SG[1]

	30-01-2003 18:34									#55
	Suffering from ads?

Yep. And considering how much effort others have done while I sat back and kibitzed, it's probably time that I stepped up to the plate and participated in the drugde of testing. Due to the nature of the amount of time it takes to restart an MP game, I probably won't run hundreds of tests, but I'll check with my stat boys to find out how many I need to do to prove it at a high confidence level. (95%)

But I will only test human against human, I'll leave AI to human to someone else, even though I believe the results would be the same.

RAH

	30-01-2003 19:28									#56
	Support Apolyton buy from Amazon

quote: Originally posted by rah I'll check with my stat boys to find out how many I need to do to prove it at a high confidence level. (95%) RAH

Perhaps I can help. You are essentially testing a null of 'no bonus' against a alternative of 'there is a bonus'. The exact calculations are tricky, because combat results are a hypergeometric binomial random variable, and because the exact 'power' of the test depends on an exact specification for the alternative hypothesis. However for the first problem the exact formula is in the combat info thread. The second I explain below.

You see your question is badly formed. The probability of rejecting the null when it is true will be fixed by the experimenter......it tells nothing. The probability of accepting the null when it is false depends on the exact specification of the alternative hypothesis.

Since the proportion of 'successes' in repeated binomial trials is normally distributed (and the variance can be derived as a function of the probability of individual success and the number of repititions) you can derive a decision rule consistent with a probability of incorrectly rejecting the null of 5%.

Of course when people do these civ tests usually they overkill the problem, because the number of tests is explicitly included in the decision rule.

The conclusion is you just need enough repititions to yield an acceptable power......and ensure that the proportion is normally distributed as assumed. Use 50.......you definitely need more than 30. Any more than 50 is overkill.

	30-01-2003 20:57									#57
	Support Apolyton buy from Amazon

Hypergeometric Binomial Random Variate - I thought I was the mathematician around here - but in fairness not a statistician...

SG[1]

	30-01-2003 22:12									#58
	Support Apolyton buy from Amazon

I didn't make it up honest.

*whispers* all it means it that combat is the result of several rounds of fighting.......but they aren't independent repititions.

Everyone uses the shortcut instead, but if Rah wants to do the test properly (and hence save himself a few hundred repititions) he will need to know the exact probabilities.

	31-01-2003 11:54									#59
	Support Apolyton, buy Galactic Civilizations: Deluxe Edition

quote: Originally posted by DrSpike I didn't make it up honest. *whispers* all it means it that combat is the result of several rounds of fighting.......but they aren't independent repititions.

Surely the rounds are independent. The probability of winning a particular round depends on the attacker's attack strength and the defender's defence strength (with any modifications - veteran, terrain, etc). Neither of these change during the contest. Similarly the damage does not change during the battle. So the probability of a given attacker beating a given defender is fixed and can be calculated by using a binomial distribution of round results and summing those that give a victory. Once the probability of winning a contest is known, the number of wins in a trial of a known length is again binomial.

For as few as 25 trials, the 95% confidence intervals around the expected number number of wins with a zero and a 50% bonus don't overlap. So the 50% sneak attack bonus can be tested relatively quickly. Incidentally, my tests using SP confirm other people's results that there is no 50% bonus for an attack by the human player that breaks a treaty compared with the same attack during a war.

RJM at Sleepers

	31-01-2003 14:13									#60
	Support Apolyton or Terrorists Win

quote: Originally posted by rjmatsleepers Once the probability of winning a contest is known, the number of wins in a trial of a known length is again binomial.

Sorry but it isn't.

I am not going to prove it here (it requires moment generating functions, which are some rather nasty albeit incredibly useful mathematical tools) but I can post a reference or PM you if you have any questions.