Discussion about scoring results/primes

[BlisteringSheep]

I know I'm kind of an outsider, but just wanted to say that I like this scoring idea. It's simple to explain and understand, and seems fair.

[IronBits]

AMDave says he got it, so we'll see how long it takes him to whip it up, once he's finished with the formatting of the progress tables, or before, his choice

[Flatlander]

Sounds good. And simple enough for even me to understand. lol

The only tweak I can think of is to allow for the different size k. i.e. someone testing running on "Team drive #5: k=400-600 n=600K-1M" will/might be running at a higher FFTlen than someone running on "Team drive #7 k=800-1001 n=600K-1M".

Probably not worth the hassle.

[Mini-Geek]

Originally Posted by Flatlander

Sounds good. And simple enough for even me to understand. lol

The only tweak I can think of is to allow for the different size k. i.e. someone testing running on "Team drive #5: k=400-600 n=600K-1M" will/might be running at a higher FFTlen than someone running on "Team drive #7 k=800-1001 n=600K-1M".

Probably not worth the hassle.

I think you mean lower.
I thought about this, but I don't think it would cause a large enough difference to be something we need to worry about. NPLB works in a relatively small k range (depending on your definition of small...I think it is), so it probably won't really have a big effect. (i.e. I agree that it's probably not worth the hassle) For reference, here's the FFT lengths with max n's for the full Drive 5-7 range (with new values for every k=200 difference, which is one drive's ):

k = 401
n(min) = 600000
n(max) = 1000000

The following FFT lengths would be used:

    fftlen       nmax
-----------------------
     40960     658091
     49152     782971
     57344     910051
     65536    1041631


k = 601
n(min) = 600000
n(max) = 1000000

The following FFT lengths would be used:

    fftlen       nmax
-----------------------
     40960     646135
     49152     768624
     57344     893313
     65536    1022502


k = 801
n(min) = 600000
n(max) = 1000000

The following FFT lengths would be used:

    fftlen       nmax
-----------------------
     40960     637647
     49152     758438
     57344     881430
     65536    1008921


k = 1001
n(min) = 600000
n(max) = 1000000

The following FFT lengths would be used:

    fftlen       nmax
-----------------------
     40960     631061
     49152     750535
     57344     872209
     65536     998383
     81920    1239732

Drive 5 will be in:
40K from 600000-646135
40K or 48K from 646136-658091
48K from 658092-768624
48K or 56K from 768625-782971
56K from 782972-893313
56K or 64K from 893314-910051
64K from 910052-1000000

Drive 6 will be in:
40K from 600000-637647
40K or 48K from 637648-646135
48K from 646136-758438
48K or 56K from 758439-768624
56K from 768625-881430
56K or 64K from 881431-893313
64K from 893314-1000000

Drive 7 will be in:
40K from 600000-631061
40K or 48K from 631062-637647
48K from 637648-750535
48K or 56K from 750536-758438
56K from 758439-872209
56K or 64K from 872210-881430
64K from 881431-998383
64K or 80K from 998384-1000000

[gd_barnes]

Many months ago, I had thought of the same thing for the top-5000 site, which has k's ranging from 1 to some ridiculously high k's. Also, I think they should score far more for non-powers-of-2 bases. Alas, the complexity of doing it is why they haven't and why it would be difficult for us to also, even with the easier formulas.

To do it right, we would need to have a database for fftlen's for every k and come up with a "base k" that gives a score of 1 at n=400K. I think it is definitely doable, but let's put it on the backburner for now. As Mini stated, since this project doesn't deal with k's ranging from 1 to 10 gagillion and combines k's over anywhere from a 200k to 1000k range in its drives, the effect on scoring would be minimal.

All of that said, this is a very good idea. Personally, I would like for us to be on the 'cutting edge' of these kinds of improvements in scoring. If we can figure out a reasonable way to implement such a thing, perhaps others, including the top-5000 site, will use it as as example.

In a past life, I created rating systems for board games and sports teams and used such systems for seeding participants in board-game tournaments. So I'm no stranger to rating and ranking systems for different things. If it's got math in it, I'm there! :-)


Gary

[AMDave]

Scoring implemented on the stats site, among other things.

[gd_barnes]

Wow, that was fast! It looks great!

One thing I might suggest: Display both with just 3 decimal places. The results could actually be displayed with no decimals. As long as the internal calculations are carried out to 6 decimals, we're good.

If it's a hassle, no big deal.

For all the Free-DCers, you can still make it sort on total primes or total results.


Gary

[IronBits]

:p

I was read this over on SOB
http://www.free-dc.org/forum/showthr...467#post131467

and found this formula
1-ln(50P)/ln(200P) = 0.0348
1-ln(100P)/ln(200P) = 0.0174
1-ln(150P)/ln(200P) = 0.0072

1-ln(400T)/ln(1P) = 0.0265
1-ln(500T)/ln(1P) = 0.0201
1-ln(400T)/ln(600T) = 0.0119

Thought maybe it was a *scoring* formula ...

[gd_barnes]

Originally Posted by IronBits

:p

I was read this over on SOB
http://www.free-dc.org/forum/showthr...467#post131467

and found this formula
1-ln(50P)/ln(200P) = 0.0348
1-ln(100P)/ln(200P) = 0.0174
1-ln(150P)/ln(200P) = 0.0072

1-ln(400T)/ln(1P) = 0.0265
1-ln(500T)/ln(1P) = 0.0201
1-ln(400T)/ln(600T) = 0.0119

Thought maybe it was a *scoring* formula ...

No, it's some sort of optimal sieve depth calculation...far more complex stuff than scoring.

[AMDave]

Originally Posted by gd_barnes

Wow, that was fast! It looks great!

One thing I might suggest: Display both with just 3 decimal places. The results could actually be displayed with no decimals. As long as the internal calculations are carried out to 6 decimals, we're good.

If it's a hassle, no big deal.

For all the Free-DCers, you can still make it sort on total primes or total results.

3 decimals on the reports - done.

Calculations re happening at 14 decimal places

Bok gets a custom extract independent of the reports for the FDC stats.

BTW - you can re-sort any table by clicking on a column heading that is underlined. Click it again and it will sort in the opposite order.

Since the scoring formula is pertinent to where these primes are going I think the scoring should be prevalent on the project's reports. But that is just me.

As an after thought, I will add the prime score to the 'NPLB Primes List' so that you can see the 'worth' of the prime and we can fire that out with the rest of the details on the mailer (which is now under construction)

[gd_barnes]

Originally Posted by AMDave

Since the scoring formula is pertinent to where these primes are going I think the scoring should be prevalent on the project's reports. But that is just me.

It's amazing that there is SO much stuff there...it's excellent!

Unfortunately I get a little lost in the maze of info. so I'm not quite sure what you mean. Can you point to the report you're referring to and how you think the score should be more prevalent?


Gary