The New Sudoku Players' Forum

[Pat]

you would need to convince the players of the Patterns Game

if they vote for change,
g.r.emlin will accept that decision

[dobrichev]

In 2012 they didn't

[tarek]

Looks to me more of a case of people not visiting the poll stations than them rejecting a very good idea. In that case why not adopt the changes unless there are objections ?

I can't see an issue currently except that of the rarity bonus. Some players with programming abilities may have their own setup that is based on v1.2.1 which explains a reluctance to change which can be an issue for g.r.emlin too.

The advantage of using the current version is to add another trump option through using the name of the techniques (the %S/%T/%U) formatting options. this is not to mention the added speed and efficiency of rating due to improved coding efficiency.

It is still in Java and 1to9only insisted on keeping it compatible with JDK8 which means it will be compatible and cross-platform to a great degree. I tried to keep the versioning as semantic as possible . As the current version is 1.4.0 this should indicate that it is still backward compatible with 1.2.1

The GUI's presence and improvements are a bonus

The ability to handle Pencilmark grids and Sukakus is a bonus

[Pat]

Looks to me more of a case of people not visiting the poll stations than them rejecting a very good idea.
In that case why not adopt the changes unless there are objections ?

we could run a Poll --

remain as is
change to your new software
change to SKFR

to add another trump option
through using the name of the techniques

would need a volunteer to change g.r.emlin

the added speed and efficiency of rating

and SKFR is even faster.

my guess is,
if the players of the Patterns Game vote for change --
they'll go with SKFR.

[m_b_metcalf]

I'd be happy to start a new series with the new SukakuExplainer, and live with any small discrepancies that might crop up with the rarity ratings.

Can we be sure that g.r.emlin can handle a new series? It needs to keep track of all the patterns ever played in previous series as well as the rarity statistics.

I would be very unhappy with a new trumping option (and it would be yet another hurdle for any new player).

Thanks to all for the effort you're putting in.

Mike

[tarek]

I guess part of me will be disappointed (especially after my impassioned speech) if skfr is picked over sukaku explainer

I guess a poll is going to be a good idea. I'm not against skfr and I didn't follow skfr development closely. I'm not sure if it inherited SE bugs or if it has bugs of its own. Certainly Sukaku explainer inherited everything good and bad that SE had.

With all of our options available: Patterns games SE, Sukaku explainer 1.4.0 and skfr, is it worth doing some benchmark testing/comparisons?!

tarek

[dobrichev]

skfr is in 2 to 3 orders of magnitude faster compared to SE.
On the other hand skfr has 2 to 3 orders of magnitude more unnecessary dependencies compared to SE and its source code is difficult to follow.
It is possible to merge these 2 projects as well as latest rating code kindly published by champagne but it would take much much efforts, especially taking into account that the final goal is unclear and acceptance of the "new rating" will always be problematic.

From the posts in this forum I see that some tried writing his own rating processor but finally were stuck at the high-end techniques.

Champagne is the only contributor who shared his rating code. (considering closer to SE rating system, else www has plenty of original rating systems)

[tarek]

skfr is in 2 to 3 orders of magnitude faster compared to SE.
On the other hand skfr has 2 to 3 orders of magnitude more unnecessary dependencies compared to SE and its source code is difficult to follow.

You and champagne are better at identifying issues with skfr as I've never used it nor was I involved in the project.

I've located the repositories and what I think are the latest releases and downloaded the binaries as I'm hopeful that I can avoid the need to compile. I'm forcing myself to explore these so that I can do the comparison myself if I can. a simple set of puzzles can be used then (like my pearly6000 or the 17clue vanilla collection)

[champagne]

I guess part of me will be disappointed (especially after my impassioned speech) if skfr is picked over sukaku explainer
tarek

Frankly, I would never try to replace Serate by Skfr.
Skfr has been produced using the frame of an old solver, heavily oriented on the "full tagging" process.
Serate frame is "pm" oriented. I used a similar frame for the last clones but, for several reasons, halted the implementation around the "nested levels".
The last clones are nx times faster than skfr, but I still use skfr in the tiny area where the fresh code is not ready (and serate at the end).

Anyway, whatever are the sluggishness of serate, it's part of the game for the players to find alternative ways to be competitive. Changing the referee of the game would be IMO a bad decision.

[tarek]

From the initial tests that I've done there is no comparison between skfr and whatever improved version of Sukaku explainer we have. skfr is much much quicker.

The known 49157 17 clue puzzles were rated in 10s with skfr compared with the regular solving Sukaku explainer's 7 minutes. This goes down with batch solving to less than 4m40s with B2 mode faster than B1.

skfr ratings are slightly off the true ratings though which I'm sure it is how it was intended to work. The change to skfr appears to be more radical than compared to sukaku explainer. A mix of the 2 therefore will be our ultimate wish but short of achieving that I will stick with sukaku explainer as the potential replacement for SE 1.2.1 serate modification

[Edit: corrected typo]

[blue]

The known 49158 17 clue puzzles ...

Just so I'm clear: Was that a typo, or have you seen a #49158 ?

[tarek]

The known 49158 17 clue puzzles ...

Just so I'm clear: Was that a typo, or have you seen a #49158 ?

The last puzzle was on line 49158. 1st puzzle was on line 2 as I had line 1 with parameters used.
therefore I can confirm that there were only 49157 puzzles. Thanks to you and rjamil for pointing that out.

tarek

[creint]

My solver: Single thread performance currently also 4:40
and multithread performance 1:30 where bottleneck is memory speed.
While skfr_win32_v_2_0_1.exe -i"17puz49157" runs in 5.15 sec.
While great part of the set were easy puzzles a dedicated solver for this part is obviously better than a variant solver which collect all steps of one level before applying.
How well would they scale on larger + harder puzzles like 16x16 or samurai?

[dobrichev]

How well would they scale on larger + harder puzzles like 16x16 or samurai?

So far SE can't solve tarek's pencilmark-only 9x9 puzzles. Skfr uses the same set of techniques and wouldn't solve them too.
What about your solver (unless it only solves but doesn't calculate SE-compatible or similar rating)?

[1to9only]

How well would they scale on larger + harder puzzles like 16x16 or samurai?

SE for 16x16 sudokus is possible! Many moons ago, I had a modified SE16 up and running - it rated one easy puzzle:

Code: Select all

.F..G..O.P.M.A..
L..ACE...B.DF.O.
.O...N.BF.......
..JP..M...C..EBI
.AC.PDO..JH...E.
GMEN......B.OC..
...F....LI..HNP.
P.....N.D.K....J
A....O.M.L.....P
.CNM..HD....I...
..PD.I......CHLK
.I...CK..DPN.BJ.
NKL..M...O..BJ..
.......CI.M...N.
.J.OI.F...LKP..G
..M.O.L.H..J..C. ED=2.3/1.2/1.2

Analyzing Sudoku #1
145 Hidden Single
3 Direct Pointing
4 Direct Hidden Pair
2 Naked Single
Hardest technique: Naked Single
Difficulty: 2.3


The GUI needs more work, and it failed to solve 2 other puzzles; to solve more difficult puzzles some code changes will be needed, and new techniques added, e.g quintuple, sextuple, etc.

Edit: The ED=2.3/1.2/1.2 puzzle was posted by monstersudoku here.
The 1st of the 2 unrated puzzles mentioned above is m_b_metcalf's modification of the monstersudoku puzzle here, is rated ED=9.2/1.2/1.2.