mersenneforum.org  

Go Back   mersenneforum.org > Great Internet Mersenne Prime Search > Data

Reply
 
Thread Tools
Old 2021-08-11, 15:50   #3433
slandrum
 
Jan 2021
California

22·5·11 Posts
Default

Quote:
Originally Posted by Zhangrc View Post
M57885161 is now in Category 0 range. We'll achieve the milestone much sooner.
The major factor was when it hit cat 1, because that prevented any new cat 2 assignments with 120 day expirations from occurring. There are assignments that will expire in mid October, which was pretty much determined by when it hit cat 1. As long as things proceed normally, that's usually the major hurdle. All of the cat 3 and cat 4 assignments will have expired out before the last of the cat 2 assignments.
slandrum is online now   Reply With Quote
Old 2021-08-16, 06:58   #3434
greenskull
Xebeche
 
greenskull's Avatar
 
Apr 2019
🌺🙏🌺

40610 Posts
Default

So far, it looks like we'll cross the M57885161 level on November 22, 2021.
That is, all tests below M57885161 will be verified by this date.

Click image for larger version

Name:	Nov22_2021_1.png
Views:	61
Size:	64.6 KB
ID:	25465

We are moving pretty close to the predicted trend.
greenskull is online now   Reply With Quote
Old 2021-08-16, 08:30   #3435
greenskull
Xebeche
 
greenskull's Avatar
 
Apr 2019
🌺🙏🌺

2·7·29 Posts
Default

And my prediction for all tests below M74207281 is ~August 29, 2025.
By this time, all exponents below ~143 million will be tested at least once.

Of course the most intriguing would be the forecast of the time when the next Mersenne prime will be found. Perhaps I will make such a try a little later.

Have you some ideas? :)

Last fiddled with by greenskull on 2021-08-16 at 08:40
greenskull is online now   Reply With Quote
Old 2021-08-16, 17:49   #3436
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

3×1,933 Posts
Default

Quote:
Originally Posted by greenskull View Post
So far, it looks like we'll cross the M57885161 level on November 22, 2021.
That's quite a graph. Axes only labeled x and y. Numerical labeling of axes cropped off. No statement of what is being graphed. Please descriptively label axes and describe the data being graphed.

The projection is an outlier compared to other methods. Here are two, giving ~October 6, based on the history of reaching the 56M DC milestone and remaining counts to Mp#48* versus date, and ~October 26 (detailed on page 5 of PDF attachment here) based on a polynomial fit to the previous ten years of DC milestones.
Attached Files
File Type: pdf gimps progress and rate.pdf (57.8 KB, 26 views)
kriesel is online now   Reply With Quote
Old 2021-08-16, 18:13   #3437
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

132478 Posts
Default

Quote:
Originally Posted by greenskull View Post
But there is a nonzero and non-vanishingly small probability that All Tests will find the missing Mersenne prime let's say between 57885161 and 74207281. Since then there were not all these GEC and PRP innovations with improvements yet.

... Correct me if I am wrong.
A) I don't think there's a missing or missed prime there. The interval 10M-100M is already rather flush with discovered Mersenne primes (13 compared to ~5.9 expected per decade). The ratio 74207281/57885161 ~1.282 is below the mean expected ratio ~1.47576.
B) PRP with GEC was introduced in Gpuowl with Mihai Preda announcing v1.0 on 2017-08-30, with kracker's posting of a v1.0 Windows build occurring 2017-09-04. That's a little over a month before first testing was completed up to 74207281 on 2017-10-05. And admirably quick turnaround after Robert Gerbicz' post of the highly reliable fast PRP error check method on 2017-08-13. Much of the double checking will be performed not with LL DC but with far more reliable PRP/GEC/proof-generation/cert. (Be careful what you ask for ... )

Last fiddled with by kriesel on 2021-08-16 at 18:32
kriesel is online now   Reply With Quote
Old 2021-08-16, 18:50   #3438
Uncwilly
6809 > 6502
 
Uncwilly's Avatar
 
"""""""""""""""""""
Aug 2003
101×103 Posts

234368 Posts
Default

Quote:
Originally Posted by greenskull View Post
But there is a nonzero and non-vanishingly small probability that All Tests will find the missing Mersenne prime let's say between 57885161 and 74207281. Since then there were not all these GEC and PRP innovations with improvements yet.
Quote:
Originally Posted by kriesel View Post
A) I don't think there's a missing or missed prime there.
I agree with this point. Between 57M and 74M is well above the 33.2M (10 million decimal digit) range, where we saw a lot of errors. It is well below the 79.3M range (upper limit of traditional GIMPS range), again, testing up near there is more error prone, because the tests ran long on machines that were not suited to those numbers. It is also well away from the 100M range. That range is another "glory" or ego range (just as 332M is, along with its high error rate). Because it is run of the mill I think we can use run of the mill error rate data from the 55M range and the testing that folks have done to investigate the error rate in the range.
Quote:
B) PRP with GEC was introduced in Gpuowl with Mihai Preda....
But, when was it (GEC) being used to turn in say 30% of the FTC results? That would be more important than when a new program became available.
Uncwilly is offline   Reply With Quote
Old 2021-08-16, 20:53   #3439
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

3·1,933 Posts
Default

Quote:
Originally Posted by greenskull View Post
between 57885161 and 74207281. Since then there were not all these GEC and PRP innovations with improvements yet.
Quote:
Originally Posted by Uncwilly View Post
But, when was it (GEC) being used to turn in say 30% of the FTC results? That would be more important than when a new program became available.
The point I was making is the claim greenskull seemed to be making, that PRP/GEC was not available (not possible to use) during the time of completion of first testing up to 74207281 (until 2017-10-05) was false. "then" is a reference to time.

Uptake of PRP was unfortunately very slow initially. Partly because implementation in prime95 came a little later. (The prime95 v29.4b5 win64 executable is dated 2017-11-11.)
Partly because remaining exponents <74207281 and many above there had already been assigned and started as LL. Partly because of inertia and time to change scripting.
There were 86 PRP/GEC first tests submitted before 2017-10-05 (completion of first testing up to 74207281), all by Mihai, all above 75M, and type 4 indicating Gpuowl.

Sampling selected days, for 74M and up, at
https://www.mersenne.org/report_ll/?...dispdate=1&B1=
and https://www.mersenne.org/report_prp/...exfactor=1&B1=
2018-01-01 to 2, PRP 6; LL 289; total 295; PRP 2%
2019-01-01 to 2, PRP 26; LL 212; total 238; PRP 10.9%
2019-07-01 to 2, PRP 69; LL 219; total 288; PRP 24%
2019-08-01 to 2, PRP 48; LL 188; total 236; PRP 20.3%
2019-09-01 to 2, PRP 69; LL 216; total 285; PRP 24.2%
2019-10-01 to 2, PRP 87; LL 222; total 309; PRP 28.2%
2019-10-08 to 9, PRP 95; LL 397; total 492; PRP 19.3%
2019-10-15 to 16, PRP 75; LL 577; total 652; PRP 11.5%
2019-10-22 to 23, PRP 100; LL 477; total 577; PRP 17.3%
2019-10-29 to 30, PRP 92; LL 385; total 477; PRP 19.3%
2019-11-01 to 2, PRP 199; LL 315; total 514; PRP 38.7%
2020-01-01 to 2, PRP 706; LL 152; total 858; PRP 82.3%

and incorporating active assignments data from https://www.mersenneforum.org/showpo...35&postcount=1
2021-01-27 79% PRP
2021-03-28 83% PRP
2021-04-30 98.9% PRP (shortly after cessation of the server issuing LL first tests 2021-04-08)

Last fiddled with by kriesel on 2021-08-16 at 21:10
kriesel is online now   Reply With Quote
Old 2021-08-17, 07:52   #3440
greenskull
Xebeche
 
greenskull's Avatar
 
Apr 2019
🌺🙏🌺

2×7×29 Posts
Default

Quote:
Originally Posted by kriesel View Post
That's quite a graph. Axes only labeled x and y. Numerical labeling of axes cropped off. No statement of what is being graphed. Please descriptively label axes and describe the data being graphed.

The projection is an outlier compared to other methods. Here are two, giving ~October 6, based on the history of reaching the 56M DC milestone and remaining counts to Mp#48* versus date, and ~October 26 (detailed on page 5 of PDF attachment here) based on a polynomial fit to the previous ten years of DC milestones.
Sorry for cropping. But I have already described what is being graphed here:
https://www.mersenneforum.org/showpo...postcount=3417
The axis values are not as interesting as it is interesting how well the current values (red crosses) fit into the trend I predicted. So far, everything looks very good :)
I think that at the moment the dispute over whose forecast is better does not make sense.
Time will tell everything. At the moment we can only place bets.
greenskull is online now   Reply With Quote
Old 2021-08-17, 08:09   #3441
greenskull
Xebeche
 
greenskull's Avatar
 
Apr 2019
🌺🙏🌺

2×7×29 Posts
Default

Quote:
Originally Posted by kriesel View Post
The point I was making is the claim greenskull seemed to be making, that PRP/GEC was not available (not possible to use) during the time of completion of first testing up to 74207281 (until 2017-10-05) was false. "then" is a reference to time.
...
No problem.
This means all this is already embedded in the curve and fine regression analysis is able to reveal all this.

I want to be understood correctly. My main interest is all sorts of forecasting, regression analysis, extrapolation. And the data that I find here is an excellent raw material for my research. Perhaps I do not take into account all the fundamental factors, but as far as I understood from your comment, they have already been fully reflected in this very data.
greenskull is online now   Reply With Quote
Old 2021-08-17, 09:00   #3442
greenskull
Xebeche
 
greenskull's Avatar
 
Apr 2019
🌺🙏🌺

2·7·29 Posts
Default

I suggest turning the forecasting into a game. Just for fun :)

Any Diviners when they wish, can make their predictions about the moment of the future event -- the Moment of Truth. There can be any number of predictions.

Obviously, the closer we are to the Moment of Truth (the shorter the Distance), the more reliable and accurate it can be predicted (less Error).

Click image for larger version

Name:	DTR.png
Views:	33
Size:	32.3 KB
ID:	25477

Therefore, we should take into account not only the Error, but also the Distance with which the prediction was made.

So I suggest:
1) To calculate the ratio of the Error in Prediction to the Distance from which this Prediction was made (Errorᵢ/Distanceᵢ)
2) To calculate the average of the ratios – to divide the resulting sum by the Total Number of Predictions N made by the Diviner

In order for the Rate to work on the principle of the more talented the Diviner, the higher the value
3) To take the reciprocal value from the received one in 2). Or in other words, to flip the ratio.

The one whose Diviner Talent Rate to the accomplished Moment of Truth turns out to be the highest is the winner.

Click image for larger version

Name:	DivinerTalentRate.png
Views:	18
Size:	48.2 KB
ID:	25478

The Diviner’s Talent Rate in the total number of Diviner’s Predictions encourages more accurate hits from a greater distance.
And no matter how many Predictions the Diviner made, it is important how accurate Diviner’s Predictions were, taking into account the Distances from which they were made.

IMPORTANT:
1) All predictions are in UTC time (it is a kind of GMT with no summertime shift).
Date and time can be specified. If the time is not specified, then we count 12:00 (noon time).
2) If the Diviner leaves their previous prediction unchanged, it is not taken into account.
If the Diviner changes their prediction relative to the previous one, then, of course, it is taken into account. Even if this prediction coincides with any one made before the previous one.

As for the prize, it can be considered :)

PS Kriesel, "gimps progress and rate.pdf" -- nice report :) Thank you!

Last fiddled with by greenskull on 2021-08-17 at 09:15
greenskull is online now   Reply With Quote
Old 2021-08-17, 16:22   #3443
kriesel
 
kriesel's Avatar
 
"TF79LL86GIMPS96gpu17"
Mar 2017
US midwest

3×1,933 Posts
Default

Quote:
Originally Posted by greenskull View Post
Sorry for cropping. But I have already described what is being graphed here:
https://www.mersenneforum.org/showpo...postcount=3417
That post describes red blue and purple, and mentions magenta without describing it.
The post I commented had no description did not link back to that.


Quote:
Originally Posted by greenskull View Post
I suggest turning the forecasting into a game. Just for fun :)

Any Diviners when they wish, can make their predictions about the moment of the future event -- the Moment of Truth. ...

PS Kriesel, "gimps progress and rate.pdf" -- nice report :) Thank you!
It's all just for fun.
Careful, use of lots of capital letters, defining your own terminology, etc, is elevating your crank score.
The pdf has been available in a minor milestones post of the GIMPS progress thread of the statistics section of the reference info compilation, and updated regularly, for years.
kriesel is online now   Reply With Quote
Reply

Thread Tools


Similar Threads
Thread Thread Starter Forum Replies Last Post
Newer X64 build needed Googulator Msieve 73 2020-08-30 07:47
Performance of cuda-ecm on newer hardware? fivemack GMP-ECM 14 2015-02-12 20:10
Cause this don't belong in the milestone thread bcp19 Data 30 2012-09-08 15:09
Newer msieves are slow on Core i7 mklasson Msieve 9 2009-02-18 12:58
Use of large memory pages possible with newer linux kernels Dresdenboy Software 3 2003-12-08 14:47

All times are UTC. The time now is 20:16.


Sun Oct 24 20:16:03 UTC 2021 up 93 days, 14:45, 2 users, load averages: 1.74, 1.64, 1.46

Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2021, 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.