mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Data > Marin's Mersenne-aries

Reply
 
Thread Tools
Old 2020-09-03, 02:50   #1
petrw1
1976 Toyota Corona years forever!
 
petrw1's Avatar
 
"Wayne"
Nov 2006
Saskatchewan, Canada

43×103 Posts
Default Did the P1 percentage calculation program change recently?

About a year ago I tried to emulate prob.php with an Excel macro and was able to get to within about 0.04% of prob.php.

I rechecked lately and find I am now out by about 0.4% (10 times worse).

I did not {intentionally} change my macro.

Thanks
petrw1 is online now   Reply With Quote
Old 2020-09-03, 04:40   #2
preda
 
preda's Avatar
 
"Mihai Preda"
Apr 2015

22·52·13 Posts
Default

Quote:
Originally Posted by petrw1 View Post
About a year ago I tried to emulate prob.php with an Excel macro
Yes. One implementation of the new calculator is in gpuowl source code https://github.com/preda/gpuowl/tree/master/pm1 in python and C++.

The news is good, the new calculator returns higher percentages :)
preda is offline   Reply With Quote
Old 2020-09-12, 19:51   #3
petrw1
1976 Toyota Corona years forever!
 
petrw1's Avatar
 
"Wayne"
Nov 2006
Saskatchewan, Canada

114D16 Posts
Default FWIW ... this is still below what I am seeing.

Quote:
Originally Posted by preda View Post
Yes. One implementation of the new calculator is in gpuowl source code https://github.com/preda/gpuowl/tree/master/pm1 in python and C++.

The news is good, the new calculator returns higher percentages :)
Over the past 3 years I have been doing extensive P-1 in the 4xM and 5xM ranges where B1=B2. Extensive in this case is over 33,000 tests.

For no other reason than it worked and I like round numbers I mostly used a new B1/B2 of 1,000,000/20,000,000.
The P1 Probability Calculator at mersenne.ca; and what is reported by Prime95 gave me a success rate back then of somewhere close to 4.25%.
Most exponents had existing P1 with B1=B2 such that their expected success rate was about 1.75%
The difference is 2.5%
The new calculator has a success rate of about 4.35% with a difference of 2.6%.

Over my 33,000 tests from the start and consistently continuing to the current date my success rate has been very close to, and often just above 3%.
petrw1 is online now   Reply With Quote
Old 2020-09-12, 22:03   #4
preda
 
preda's Avatar
 
"Mihai Preda"
Apr 2015

22·52·13 Posts
Default

Quote:
Originally Posted by petrw1 View Post
Over the past 3 years I have been doing extensive P-1 in the 4xM and 5xM ranges where B1=B2. Extensive in this case is over 33,000 tests.

For no other reason than it worked and I like round numbers I mostly used a new B1/B2 of 1,000,000/20,000,000.
The P1 Probability Calculator at mersenne.ca; and what is reported by Prime95 gave me a success rate back then of somewhere close to 4.25%.
Most exponents had existing P1 with B1=B2 such that their expected success rate was about 1.75%
The difference is 2.5%
The new calculator has a success rate of about 4.35% with a difference of 2.6%.

Over my 33,000 tests from the start and consistently continuing to the current date my success rate has been very close to, and often just above 3%.
You should also mention the factored-to (TF) value you used, as it affects the probabilities.

mprime uses BS (Brent-Suyama extension) (when E>2, usually E==6), which may also bring a slight benefit, but not exactly known how large.

If you have a big set of results, you could analyze them to filter segregate the cases:
- P-1 found a factor missed by TF (i.e. factored-to value was not correct, too large)
- P-1 found a BS factor (i.e. beyond B2)

Last fiddled with by preda on 2020-09-12 at 22:04
preda is offline   Reply With Quote
Old 2020-09-12, 22:23   #5
petrw1
1976 Toyota Corona years forever!
 
petrw1's Avatar
 
"Wayne"
Nov 2006
Saskatchewan, Canada

10001010011012 Posts
Default

Quote:
Originally Posted by preda View Post
You should also mention the factored-to (TF) value you used, as it affects the probabilities.

mprime uses BS (Brent-Suyama extension) (when E>2, usually E==6), which may also bring a slight benefit, but not exactly known how large.

If you have a big set of results, you could analyze them to filter segregate the cases:
- P-1 found a factor missed by TF (i.e. factored-to value was not correct, too large)
- P-1 found a BS factor (i.e. beyond B2)
I used the current TF value for both the B1=B2 and my current values (for consistency).

BS has found very few for me (I don't have an exact count though)

I like your idea of an analysis/filter; just a bit of work for me.

Thx
petrw1 is online now   Reply With Quote
Old 2020-09-13, 00:01   #6
storm5510
Random Account
 
storm5510's Avatar
 
Aug 2009
U.S.A.

2×811 Posts
Default

Quote:
Originally Posted by petrw1 View Post
...For no other reason than it worked and I like round numbers I mostly used a new B1/B2 of 1,000,000/20,000,000.

The P1 Probability Calculator at mersenne.ca; and what is reported by Prime95 gave me a success rate back then of somewhere close to 4.25%.
Most exponents had existing P1 with B1=B2 such that their expected success rate was about 1.75%...
I use James' calculator to get a general feel for where I should be. I am running B1 at 1,000,000 and B2 at 25,000,000 on wavefront exponents. James' calculator defaults to 3% probability and I leave it there. My lifetime rate at finding factors is 2.3%. So, that is close.
storm5510 is offline   Reply With Quote
Old 2020-09-13, 00:14   #7
Prime95
P90 years forever!
 
Prime95's Avatar
 
Aug 2002
Yeehaw, FL

112×59 Posts
Default

Quote:
Originally Posted by petrw1 View Post
I used the current TF value for both the B1=B2 and my current values (for consistency).

BS has found very few for me (I don't have an exact count though)

I like your idea of an analysis/filter; just a bit of work for me.

Thx

B-S data from 33,000 tests would be very useful. If you can tell us, for each E value, how many P-1 tests were run and how many factors would have been missed if B-S was not used.
Prime95 is offline   Reply With Quote
Old 2020-09-13, 14:57   #8
axn
 
axn's Avatar
 
Jun 2003

2×7×337 Posts
Default

Does the p-1 probability calculation account for the fact that k from 2kp+1 is more likely to have smaller factors compared to random number of the same size (because 2kp+1 is prime and does not have small factors)?

I had estimated that this is roughly equivalent to a 2-bit reduction in size for k.
axn is offline   Reply With Quote
Old 2020-09-13, 16:07   #9
petrw1
1976 Toyota Corona years forever!
 
petrw1's Avatar
 
"Wayne"
Nov 2006
Saskatchewan, Canada

43·103 Posts
Default

Quote:
Originally Posted by Prime95 View Post
B-S data from 33,000 tests would be very useful. If you can tell us, for each E value, how many P-1 tests were run and how many factors would have been missed if B-S was not used.
Can anyone tell me how I would find such data?
petrw1 is online now   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Mechanics of DC instead of LL percentage Fred PrimeNet 1 2016-02-29 02:19
No results for Recently set up GPU 2 72 computer jschwar313 GPU to 72 3 2016-01-31 16:41
Recently added T.Rex mersennewiki 20 2011-04-09 02:23
Strange happenings recently cheesehead Forum Feedback 1 2008-06-07 08:33
Modifying Percentage-Done Display? Primeinator Software 4 2005-03-15 05:32

All times are UTC. The time now is 04:43.

Thu Oct 29 04:43:37 UTC 2020 up 49 days, 1:54, 1 user, load averages: 1.92, 1.73, 1.59

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.

This forum has received and complied with 0 (zero) government requests for information.

Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.
A copy of the license is included in the FAQ.