The Azure Heights Forum


Post New Topic  New Poll  Post A Reply
my profile | search | faq | forum home
  next oldest topic   next newest topic
» The Azure Heights Forum   » Main Campus   » Research Lab   » 99.6% not 100% ??

UBBFriend: Email this page to someone!    
Author Topic: 99.6% not 100% ??
borg_5x8
Orangutan Spouse
Member # 549

Rate Member
posted 06-25-2000 04:38 PM      Profile for borg_5x8   Author's Homepage   Email borg_5x8   Send New Private Message      Edit/Delete Post  Reply With Quote 
...I'm guessing this is the a confidence interval, based on the assumption that the chance of an attack succeeding will produce a normal distribution? (And therefore a 100% CI will span from -infinity to infinity...kinda impractical)

If so, does anyone have the actual data that was originally used to determine that the success probability IS random? (Purely for interest) or even some ROM code for a random routine? Producing a truely random outcome isn't incredibly simple on a computer system, but I guess nintendo made sure the gameboy had a built in (approximately) random function? (This is Nintendo we're talking about here... )
again, if so, how accurately do GB emulators model the gameboys random fuction? (being based on PC psuedo-random functions)? A lot of Pokemon statistics are based on emulator data...has anyone ever done any kind of tests to see if any emulators produce biased results compared to the GB itself?

Just wondering (for fun)

Ugh, I posted this at the cheatplanet boards when I first arrived there, and the answers were.. less than satisfying. You guys seem more technically oriented...


From: Surrey, UK | Registered: Jun 2000  |  IP: Logged
Blitzzzzzzz
Farting Nudist
Member # 586

Member Rated:
posted 06-25-2000 06:23 PM      Profile for Blitzzzzzzz   Email Blitzzzzzzz   Send New Private Message      Edit/Delete Post  Reply With Quote 
ok...yeah each move except swift i believe has a chance of missing.

well, i don't know if u people make hey i'm new topics, but i'm just saying that i'm new here now, i guess.

Don't get mad at me! i know azure has a rep. of being really smart about the moveset and raising research...i don't know much about that stuff! lol, I hope I like it around here!

------------------
Hey i'm Blitzzzzzzz from TPM

Best/Improved Team:
Charizard Nidoking Mewtwo Articuno Raichu Gyrados

Ice Pokemon!
fire and ghost rcool2


From: CA, USA | Registered: Jun 2000  |  IP: Logged
borg_5x8
Orangutan Spouse
Member # 549

Rate Member
posted 06-25-2000 06:24 PM      Profile for borg_5x8   Author's Homepage   Email borg_5x8   Send New Private Message      Edit/Delete Post  Reply With Quote 
Oh, I see! There's me, going and over-complicating things again..

But still, IF the number produced is random, the results will tend to cluster around the midrange (128) area...this won't affect something with a high probability of success (surf) coz it's just "success or not", it doesn't matter WHICH of those 256 values is produced... but something with only, say, 30% accuracy can't just do a "if (random value) < (30% of 256)" since the numbers below 30% of 256 wouldn't appear 30% of the time? How are these mechanics done in the actual game code? (ROM fiddlers, the stage is yours)

------------------


From: Surrey, UK | Registered: Jun 2000  |  IP: Logged
White Cat
Nobody knows why I'm an admin.
Member # 42

Member Rated:
posted 06-26-2000 03:54 AM      Profile for White Cat   Author's Homepage   Email White Cat      Edit/Delete Post  Reply With Quote 
If the random number generator was more likely to pick number at the middle of the range, then the chance of missing with Swift (or whatever) would be considerably less than 0.390625%. AFAIK, the game generate random numbers on a uniform distribution (flat line), not a normal distribution (bell curve). It's just as likely to pick 255 as it is to pick 128.
From: Calgary | Registered: Feb 2000  |  IP: Logged
borg_5x8
Orangutan Spouse
Member # 549

Rate Member
posted 06-26-2000 11:54 AM      Profile for borg_5x8   Author's Homepage   Email borg_5x8   Send New Private Message      Edit/Delete Post  Reply With Quote 
Hmm, Psuedo-random then? You would have thought they'd build in a least ONE method of generating a truely random number into the gameboy...? I wonder if the other consoles use this method too?

------------------


From: Surrey, UK | Registered: Jun 2000  |  IP: Logged
Givera
Farting Nudist
Member # 234

Member Rated:
posted 06-26-2000 12:27 PM      Profile for Givera   Author's Homepage        Edit/Delete Post  Reply With Quote 
Computers cann't generate TRULY random numbers... they cann't roll a die and read what the result is...

Most computers will collect "noise" (Disturbance generated by analog parts) and then do some magic and "hash" (An algorithm that takes an input of variable length and makes an output of fixed length) the information into a "seed" (A number that starts the psuedo randomization)

Once the "seed" is generated it is run through an algorithm that makes all sort of changes etc. and comes out with a psuedo-random number... the next time it needs to generate a number the psuedo-random number is USUALLY used (In high security routines and computer a new seed is hashed every time)

What I've heard as being the best accuracy test is:

A) Generate a random number (X) between 0 and 255 (Inclusive)
B) If X < Accuracy Then Hit

It's the simplest there is - If the accuracy is 255 (Swift, etc) and if 255 is picked it still gives the chance for missing...

It's just as likely to pick 255 as it is to pick 128.
I doubt that's 100% true with the difficulties of generating random numbers but they most likely wrote an algorithm that TRIES to make those two values equally likely...
-Givera

------------------
"Yea..." -Heero Yuy


From: Plano, Tx | Registered: Apr 2000  |  IP: Logged
Mr. K
Racist
Member # 2

Member Rated:
posted 06-26-2000 07:39 PM      Profile for Mr. K   Author's Homepage   Email Mr. K   Send New Private Message      Edit/Delete Post  Reply With Quote 
quote:
Originally posted by Darryn:
When a attack is done, a random number is selected.....now there are 256 possibilities ( including 0 ) and for an attack that is 'considered' to have 100% accuracy, if the random number is 1 or above the attack hits, but if it's 0, it's a miss.

I thought this was the case too, but cfalcon tells me it's the other way around. If the computer rolls a 255 (FF), it's a miss.

Doesn't really make much difference, tho.

But yes, the 99.6% comes from 255/256 and it also applies to Swift. Swift ignores Accuracy/Evade modifiers, but it still misses roughly .4% of the time.

The distribution of "random" numbers the CPU comes up with is usually pretty good, but we have seen some things that just don't seem to add up.

After a over 1000 trials of some attacks, tho, it was comforting to see that a number of them hit just about exactly 996 out of 1000 times...

------------------
KOFFING!
Azure Heights Pokémon Laboratory


From: Cinnabar Island | Registered: Feb 2000  |  IP: Logged
magmar5
Farting Nudist
Member # 592

Member Rated:
posted 06-27-2000 09:02 PM      Profile for magmar5   Author's Homepage   Email magmar5   Send New Private Message      Edit/Delete Post  Reply With Quote 
It is fair. We already have SWIFT and it is a short chance relief.

------------------
-magmar5

ThePokemasters- magmar5


From: new jersey | Registered: Jun 2000  |  IP: Logged
borg_5x8
Orangutan Spouse
Member # 549

Rate Member
posted 06-28-2000 10:57 AM      Profile for borg_5x8   Author's Homepage   Email borg_5x8   Send New Private Message      Edit/Delete Post  Reply With Quote 
So technically, all the probability data we have is lies.

quote:
is usually pretty good, but we have seen some things that just don't seem to add up.

Pretty much describes EVERYTHING about pokemon mechanics...

------------------


From: Surrey, UK | Registered: Jun 2000  |  IP: Logged


All times are ET (US)  
Post New Topic  New Poll  Post A Reply Close Topic    Move Topic    Delete Topic next oldest topic   next newest topic
Printer-friendly view of this topic
Hop To:


Contact Us | Azure Heights Pokémon Laboratory

© 2000-2016, Maximum Penetration Industries.

Karpe Diem


The views and opinions expressed on this page are strictly those of the author(s). The contents and links have not been reviewed or approved by the University of Miami.