what are Reliability and Confidence?

[dragonbud20]

I was looking at the specific setting for my computer on the website and I noticed these two values as well as CPU rolling average. I believe I've seen Reliability and Confidence mentioned before and I was wondering what they all meant.

[MattcAnderson]

Hi Math People,

That is a good question. Although I am not intimately familiar with the inner workings of Prime95, I offer two Wikipedia articles.

Reliability - https://en.wikipedia.org/wiki/Reliability

Confidence - https://en.wikipedia.org/wiki/Confidence

In My Humble Opinion (IMHO), Wikipedia is a great free encyclopedia.

Thanks again for the question.

Regards,
Matt

[Mini-Geek]

Reliability is a measure of how likely your computer is to produce a correct result. Confidence is how sure PrimeNet is of that reliability figure.

For example, when you start out, you'll probably have a good reliability measure, but with very low confidence (i.e. we can only guess you're good). If you consistently return error-free results (either because of a matching DC or just from the error tracking during the test), the reliability and confidence get higher (i.e. we know you're good). If you consistently return bad results, the reliability gets lower and the confidence goes up (i.e. we know you're bad).

[dragonbud20]

Do you have any idea what the ranges of the values are? my reliability is .98 which I presume is a percentage and the confidence is 10.0 my first assumption would be statistical confidence but that doesn't really make sense in this context.

[Mini-Geek]

Originally Posted by dragonbud20

Do you have any idea what the ranges of the values are? my reliability is .98 which I presume is a percentage and the confidence is 10.0 my first assumption would be statistical confidence but that doesn't really make sense in this context.

I found some details on it.

Reliability is a rolling average, and is a percentage like you thought and so would range from 0-1: yours means that it thinks there's a 98% probability that a result you give is correct (probably uncoincidentally, 0.98 is the score given for an unverified but no-errors-reported LL test, like most first-time LLs). Confidence is just a count of how many LL tests you've returned, so yours means you've done 10 tests (this is more simplistic than confidence intervals). A more advanced confidence test should include the estimated time of the LL test, but that level of precision isn't really needed for GIMPS's purposes.

See also the thresholds page that has some reliability/confidence levels (in short, a confidence of 2.0 or above and a reliability of 0.95 or higher is good enough for anything).

[dragonbud20]

Thanks for the help in tracking that info down

[chalsall]

Originally Posted by dragonbud20

Do you have any idea what the ranges of the values are? my reliability is .98 which I presume is a percentage and the confidence is 10.0 my first assumption would be statistical confidence but that doesn't really make sense in this context.

Some empirical data: One of my absolutely trustworthy machines (no errors, only does Double Checks, thousands of results) has a reliability of .98 and a confidence of 10.0.

It would be, indeed, interesting to have an explication as to the maths behind these numbers...

[kladner]

Originally Posted by chalsall

Some empirical data: One of my absolutely trustworthy machines (no errors, only does Double Checks, thousands of results) has a reliability of .98 and a confidence of 10.0.

It would be, indeed, interesting to have an explication as to the maths behind these numbers...

My only system, which has committed a variety of errors, though not recently, also has .98 reliability and confidence 10. I don't believe confidence goes above 10. I assume that my system has done enough good work, since the last bad spell, for the rating system to max out.

[snme2pm1]

My confidence took a tumble after 4 of 4 cores returned bad results from double check 34M work on almost the same day not long ago.
That has been the only known problem for such work.
As I look now after a little more DC work, I see Reliability, Confidence 0.98, 10.0.
P.S. I was referring to my personal confidence, I didn't note to what extent PrimeNet blinked.

[LaurV]

To contribute some data, all computers in my list have a 0.98, 10.0, except two.

From the two, one is 0.99, 10.0 (Yaaarrrr! )

The other is a "split", or "clone" which means (quote from the web page) "The server may erroneously create a duplicate entry for one of your computers. This isn't harmful, but can result in some confusion having two entries for the same computer in the table above. This "cloning" can occur if you upgrade a computer's CPU, upgrade or re-install Windows, or run into a program bug. You can merge the two entries by checking two computers above, checking the "I am sure" box, and clicking the merge button."

This "split" shows a 0.98, 2.0 initially, but then I "merged" it (luckily, the computer is still there, and the "clone" didn't report anything since end of June, I assume the server got confused by my July holiday). After merging it with the original, the score of the original is still 0.98, 10.0 (no "average" was done, I was a bit afraid it will do that).

[kladner]

Originally Posted by snme2pm1

.....
P.S. I was referring to my personal confidence, I didn't note to what extent PrimeNet blinked.