Perft(14) Weekly Status Reports for 2016

Discussion of chess software programming and technical issues.

Moderators: Harvey Williamson, bob, hgm

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Rochester
Posts: 55
Joined: Sat Feb 20, 2016 5:11 am

Re: Perft(14) Weekly Status 2016-03-06.

Post by Rochester » Sun Mar 06, 2016 5:32 pm

Hello Spain! I think you can also make the project slip chart in excel. Then see the predict is always wronged.

User avatar
Ajedrecista
Posts: 1483
Joined: Wed Jul 13, 2011 7:04 pm
Location: Madrid, Spain.
Contact:

Re: Perft(14) Weekly Status 2016-03-06.

Post by Ajedrecista » Sun Mar 06, 2016 7:03 pm

Hello again:

I have used Excel with the data provided by Steven, that is, elapsed days (515, 522, 529, ..., 571, 578) and the number of calculated results. Then I used my proposed formula and I obtained exactly the same results than Steven.

I think you are confused by the fact that expected remaining times reported weekly have not decreased at a constant pace of 7 days, this is, I think you expect expected_time(week_i) - expected_time(week_i + 1 week) = 7 days. Please re-read my example of downloads and speeds. The download speed plays the role of results/week.

Other example could be driving a car from point A to point B. Please imagine you are at M (the midpoint of A and B) and your logical prediction is that you will arrive to B in the same time that you went from A to M... IF you maintain your speed. Then imagine that it is not enough for you because you are late for a job interview (for example) so you increase your speed. There are not traffic jams so time(M -> B) < time(A -> M). With numbers:

Code: Select all

Distance&#40;A -> B&#41; = 20 miles.
Distance&#40;A -> M&#41; = Distance&#40;M -> B&#41; = 20/2 = 10 miles.

Speed&#40;A -> M&#41; = 50 mph.
time&#40;A -> M&#41; = Distance&#40;A -> M&#41;/Speed&#40;A -> M&#41; = 10/50 hours = 12 minutes.

/* I think the following but it has not happen yet&#58;

Prediction&#58; if I maintain Speed&#40;M -> B&#41; = Speed&#40;A -> M&#41; = 50 mph&#58;

time&#40;M -> B&#41; = Distance&#40;M -> B&#41;/Speed&#40;M -> B&#41; = 10/50 hours = 12 minutes.

I will be late! I must hurry.

End of my thought. */

Speed&#40;M -> B&#41; = 60 mph.
time&#40;M -> B&#41; = Distance&#40;M -> B&#41;/Speed&#40;M -> B&#41; = 10/60 hours = 10 minutes.
You thought that you will spend 24 minutes from A to B when you were at M and you accelerated, stopping the clock at 22 minutes < 24 minutes. These 2 minutes were enough to arrive on time to the job interview. Congratulations! Well done. :)

The speed of your car can be compared with Steven's resources. The increase of speed of your car can be explained, compared with Steven's run of Perft(14), with the fact that he has dedicated more machines to run the computations (hardware related) plus the improvement of some algorithms that are more efficient now (software related). At the start of 2016, he thought that there were 1168 days remaining but in a period of nine weeks, those upgrades and enhacements have shown up and now he predicts 940 days remaining. Of course 1168 - 940 = 228 > 63.

I would ask you to write your methodology on how you calculate the remaining time based on the data provided by Steven. I need three numbers: number of total results (96.4M, which is constant by the way), elapsed days and calculated results. My formula [(96,400,068)/(calculated results) - 1]*(elapsed time) = (expected remaining time) matches perfectly with Steven's forecasts of the end of his run. What is your formula or methodology?

Regards from Spain.

Ajedrecista.

Rochester
Posts: 55
Joined: Sat Feb 20, 2016 5:11 am

Re: Perft(14) Weekly Status 2016-03-06.

Post by Rochester » Sun Mar 06, 2016 7:51 pm

Ajedrecista wrote:I think you are confused by the fact that expected remaining times reported weekly have not decreased at a constant pace of 7 days, this is, I think you expect expected_time(week_i) - expected_time(week_i + 1 week) = 7 days.
I say the good predict go down 7 days in week. If not so no good use. The download can also be bad in windows!

Remained days = remained work / current rate.

The bad is

Remained days = remained work / total rate.

Rochester
Posts: 55
Joined: Sat Feb 20, 2016 5:11 am

Re: Perft(14) Weekly Status 2016-03-06.

Post by Rochester » Sun Mar 06, 2016 8:25 pm

The maestro predict:
sje wrote:Estimated remaining day count: 940
That givet Oct 2, 2018.

We can also calculate:

Code: Select all

Estimated remaining day count&#58; 520
That givet Aug 8, 2017. (November is wronged.)

Only God can know the truth. But we have to wait! In the download sometimes I wait very long time with the 99%.

User avatar
sje
Posts: 4675
Joined: Mon Mar 13, 2006 6:43 pm

Perft(14) Weekly Status 2016-03-13

Post by sje » Sun Mar 13, 2016 5:13 am

Perft(14) Weekly Status 2016-03-13

Symbolic has produced more than 37,400,000 perft(7) results so far, about 38.80% of the 96,400,068 needed.

Day count: 585
Estimated remaining day count: 923
Estimated total day count: 1,508

Average throughput: 63,932 results/day
Effective frequency: 134.18 GHz

Work units not yet started (577): 387-963

Sum of perft()s: 874,557,759,012,636,442
Sum of products: 6,840,055,369,516,414,060

User avatar
sje
Posts: 4675
Joined: Mon Mar 13, 2006 6:43 pm

Perft(14) Weekly Status 2016-03-20

Post by sje » Sun Mar 20, 2016 4:02 am

Perft(14) Weekly Status 2016-03-20

Symbolic has produced more than 38,200,000 perft(7) results so far, about 39.63% of the 96,400,068 needed.

Day count: 592
Estimated remaining day count: 902
Estimated total day count: 1,494

Average throughput: 64,527 results/day
Effective frequency: 138.79 GHz

Work units not yet started (569): 395-963

Sum of perft()s: 891,514,207,300,404,108
Sum of products: 7,098,903,525,708,309,498

User avatar
sje
Posts: 4675
Joined: Mon Mar 13, 2006 6:43 pm

Starting the long verification sequence

Post by sje » Sat Mar 26, 2016 6:29 pm

Starting the long verification sequence of work units 400 through 799.

A while back, Ankan Banerjee used his GPU perft() program to process the Perft(14) project work units 400 through 799. This is a tremendous contribution, but it is only now that Symbolic is starting the verification process of those units. It will take about a year to run through all 400 units which total forty million perft(7) calculations.

Because Ankan's program uses a 64 bit hash signature, it is more suspect to false positive matches than is Symbolic with its 128 bit signatures. The verification process should be able to detect any false positives and so give an estimate of their overall frequency.

User avatar
sje
Posts: 4675
Joined: Mon Mar 13, 2006 6:43 pm

Perft(14) Weekly Status 2016-03-27

Post by sje » Sun Mar 27, 2016 4:15 am

Perft(14) Weekly Status 2016-03-27

Symbolic has produced more than 38,900,000 perft(7) results so far, about 40.35% of the 96,400,068 needed.

Day count: 599
Estimated remaining day count: 885
Estimated total day count: 1,484

Average throughput: 64,942 results/day
Effective frequency: 142.55 GHz

Work units not yet started (562): 402-963

Sum of perft()s: 907,864,583,803,992,426
Sum of products: 7,377,649,252,485,977,444

User avatar
sje
Posts: 4675
Joined: Mon Mar 13, 2006 6:43 pm

Perft(14) Weekly Status 2016-04-03

Post by sje » Sun Apr 03, 2016 4:23 am

Perft(14) Weekly Status 2016-04-03

Symbolic has produced more than 39,500,000 perft(7) results so far, about 40.98% of the 96,400,068 needed.

Day count: 606
Estimated remaining day count: 873
Estimated total day count: 1,479

Average throughput: 65,182 results/day
Effective frequency: 144.59 GHz

Work units not yet started (556): 408-963

Sum of perft()s: 918,917,460,576,509,821
Sum of products: 7,570,354,176,779,848,496

User avatar
sje
Posts: 4675
Joined: Mon Mar 13, 2006 6:43 pm

Re: Starting the long verification sequence

Post by sje » Sun Apr 03, 2016 9:58 am

So far, two of the 400 work units in the 400-799 sequence have been verified: units 401 and 403. That's 200,000 perft(7) results in perfect agreement.

It will take about a year to handle units 400-799. As each of these units is completed, the results will be checked against Ankan's earlier results, and any differences will be reported.

Post Reply