Reply With Quote
How confident do you want to be that it gives the correct answer at least 99.999% of the time?
Hi all - probably a simple question but I'm not understanding the various articles and such that I've looked up.
Here's the problem ... I have to test some software and verify that it gives the correct answer 99.999% of the time. We have a truth model that can generate as many samples are require and the comparison is effectively binary - either its wrong or its right.
So how do I go about figuring out how many samples we need to run? Where should I start?
Thanks!
How confident do you want to be that it gives the correct answer at least 99.999% of the time?
"His programming is malfunctioning. It begins! Get your weapons, he's going to become a killbot!!!" - bryangoodrich
It isn't actually specified in the requirement - let's say 95% confident and see where it takes us.
Did some more reading last night and I'm not sure that I'm finding the right reference material for what I'm trying to do - or more likely, I'm lost in the terminology. The stuff I'm finding all seems to reference sample size WRT population size. What is my population size in this case? I need to test a certain number of samples, but the pool of possible samples is not feasibly countable - ie., it's not like a public opinion survey (which alot of the stuff I've read appears to be aimed at) where you need to sample X people out of the finite population of New Jersey to understand their opinion to a certain confidence.
Last edited by astrogeek; 10-10-2012 at 09:01 AM.
Posting Permissions
|
|