The New Sudoku Players' Forum

[Patrice]

::: more :::

Code: Select all

.4...1...7...9.8.........3.....8...4.2.5.3.9.5...6.7...8...9.....5.2...6...7...1.  ED=4.2/4.2/4.2
.3...8...7...2.1.........3.....1...5.9.6.7.8.3...8.2...5...6.....1.4...9...9...6.  ED=4.7/3.8/2.6
.2...8...9...3.7.........1.....8...4.7.5.9.2.3...4.6...5...7.....1.6...3...4...8.  ED=4.8/1.5/1.5
.5...3...3...7.6.........8.....1...4.2.7.8.9.4...6.5...7...9.....6.3...1...5...2.  ED=4.9/1.2/1.2
.2...7...5...6.7.........9.....1...3.6.8.3.7.9...4.6...8...9.....6.5...8...2...3.  ED=5.0/4.0/2.6
.6...3...3...7.1.........8.....5...9.9.2.1.4.7...9.5...8...2.....9.3...5...4...2.  ED=5.2/3.6/2.6
.5...9...2...8.6.........4.....2...1.7.4.5.6.3...1.4...9...7.....8.5...2...3...1.  ED=5.6/4.5/2.6
.9...3...3...6.1.........7.....4...6.2.7.8.5.1...5.4...7...9.....3.8...4...2...6.  ED=5.7/3.8/2.6
.8...6...7...8.6.........3.....4...2.9.5.1.6.8...7.4...1...3.....8.6...4...9...5.  ED=5.8/2.6/2.6
.9...7...1...5.6.........3.....1...9.3.4.6.2.2...7.8...6...9.....8.3...1...8...4.  ED=5.9/3.4/3.4
.4...5...7...2.1.........2.....7...5.9.6.4.1.3...8.2...5...8.....8.6...9...3...7.  ED=6.0/2.0/2.0
.1...7...5...8.9.........3.....6...4.8.1.3.6.6...5.7...2...8.....7.4...9...2...1.  ED=6.5/4.4/2.6
.8...7...5...4.1.........9.....5...2.6.8.3.4.9...1.3...7...4.....6.3...8...2...5.  ED=9.3/9.3/9.3

[Patrice]

::: comment :::
Thanks Joël

I have also these more, but with rating identical to yours

Code: Select all

.5...8...8...2.3.........6.....5...9.2.4.1.7.4...3.6...9...2.....3.8...1...5...9.  ED=4.4/4.4/3.4
.4...2...2...7.8.........9.....6...3.2.9.8.6.1...2.5...8...4.....4.5...6...3...4.  ED=6.2/6.2/2.6

I worked a lot to optimize my generator especialy the brute force solving (validity and minimality of puzzles) and the algorithm to choose the seeds and N for generation {-n+n}.

For this game I generated about 4 millions puzzles.
This pattern didn't give rapidly all the diamonds : as you can see there are 2 diamonds in my more !

But now there too few players ! How can we attract more players ?

[joel64]

::: comment :::


hi, Patrice

Perhaps, are the others players discouraged by your performance?

Personally, I have not found the two "diamonds"

I do a systematic search starting from the highest ranked (010002000300040500000000060000070008090806070800050400060007000008030001000900020)

I have come only to "010002000300040500000000060000070003060201050400050800040003000005090008000800010" with 400000 puzzles

This method is not valid given the power of my computers


Joël

[Patrice]

::: comment :::

Joël,

I tried systematic generation in the past but it was not efficient... so I moved to using generation {-n +n}. The trick is :
- which value to N ?
- which seeds ?

For N : 4 is generally a good choice, in some cases it necessary to use 5 when the rate of generation is tool low

For the seeds : I put all generated puzzles in a database and I choose the best ER rated puzzles or the best ED rated puzzles.
I use also a selection of ER rated puzzles in the range of 4.4 to 6.1 to have enough puzzles in that field.

For the generation : I have a parameter to exclude generation of puzzles beginning with a single : that helps a lot to find rapidly interesting puzzles.

I forgot to say all my software was at the beginning written in Java.
I rewrite all in C++ and I developped a rating engine from scratch that mimics SE (without nested levels). It rates puzzles 10 to 30 times faster than SE. There are very few discrepancies between SE and my rating engine (that explains one bad SE rating in game 176). I have to use SE only for some puzzles above 9.3, that is less than 1/1000 puzzles (in current game 178 : I had to rate less than 200 puzzles with SE on a total above 4 millions).
This rating engine is not SKFR from champagne (even if I helped champagne and dobrichev some time ago).

To help me to find which puzzles to play, I have developped a dashboard that shows what have been played and by whom, what I can play and some informations like diamond/pearl... selftrump... written in Java.

The only part I didn't developped : playing rarities !

For the interest of the game : you are right I must stop playing a while ! the entry level in the game is now too high for newcomers !

Best regards
Patrice

[gsf]

::: comment :::

later this summer I'll have some time to update sudoku code
one of the things on my todo list is to finally replace java SE with a leaner but as-close-as-possible app
that is, for personal use and for the patterns game

a friendly competition between implementations might be a good diversion
faster rating might be one less impediment for new players to join the game

one criteria for using a new rater in the game would be to document all inconsistencies in the game history
between SE and the new rater -- few or no inconsistencies would be a big win, but speed could outweigh that

of course any rater considered would have to be available to the sudoku community

Joël,

I tried systematic generation in the past but it was not efficient... so I moved to using generation {-n +n}. The trick is :
- which value to N ?
- which seeds ?

For N : 4 is generally a good choice, in some cases it necessary to use 5 when the rate of generation is tool low

For the seeds : I put all generated puzzles in a database and I choose the best ER rated puzzles or the best ED rated puzzles.
I use also a selection of ER rated puzzles in the range of 4.4 to 6.1 to have enough puzzles in that field.

For the generation : I have a parameter to exclude generation of puzzles beginning with a single : that helps a lot to find rapidly interesting puzzles.

I forgot to say all my software was at the beginning written in Java.
I rewrite all in C++ and I developped a rating engine from scratch that mimics SE (without nested levels). It rates puzzles 10 to 30 times faster than SE. There are very few discrepancies between SE and my rating engine (that explains one bad SE rating in game 176). I have to use SE only for some puzzles above 9.3, that is less than 1/1000 puzzles (in current game 178 : I had to rate less than 200 puzzles with SE on a total above 4 millions).
This rating engine is not SKFR from champagne (even if I helped champagne and dobrichev some time ago).

To help me to find which puzzles to play, I have developped a dashboard that shows what have been played and by whom, what I can play and some informations like diamond/pearl... selftrump... written in Java.

The only part I didn't developped : playing rarities !

For the interest of the game : you are right I must stop playing a while ! the entry level in the game is now too high for newcomers !

Best regards
Patrice

[Patrice]

::: comment :::

Glenn,

There is a big difficulty to replace SE in the range 9.3 to 11.x

SE uses a recursive approch called nested levels : it's a very powerfull way, but very difficult to reproduce to get exactly the same rating as SE. This is why I didn't implement it in my rating engine.

In the game I use my rating engine for all levels except nested levels. I found discrepancies only two times (the first one was a bug and is Ok now - the second : I have to study it). I verify no more my rating to compare to SE.

I think that we can stick to SE rating not including nested levels but we have to shift to a different approch to replace nested levels rating. champagne has developped SKFR to rate very difficult puzzles ... and he can do this very efficiently. eleven tried SKFR and find it very interesting.

I'm open to give my code to the community, but I've first to document it better (and in english) and to correct some bugs.

Best regards
Patrice

[champagne]

some comments regarding skfr

the current version has an identified bug in the nested levels giving an optimistic rating in some specific cases.

I did not try to solve that bug just because I am preparing the basis for a new (likely much faster) design in my own program.

I'll try to play one of the next games using that new version, far from being ready to be included in skfr.

champagne

[dobrichev]

Hi,
I have a version of skfr ported to gnu linux compiler.
Only latest fix by champagne isn't merged, but it is matter of few lines of code.
Skfr is performing well on 24 parallel threads on a shared memory machine. I have some preliminary code for distribution of the rating batch over many tcp/ip connected nodes.
I stopped my activities on skfr project due to the very limited interest from the community.
Now i see that champagne, the author of skfr, has different plans too.

For me the "universal" puzzle rating is still an open issue and i hope that you guys, soon or later, will develop some fresh and much better.

Cheers,
Md

[champagne]

Now i see that champagne, the author of skfr, has different plans too.

Hi mladen,

this is not quite true, I changed significantly the scope , but I am still in line with skfr rating rules and general approach.

I did it in my own program for several reasons, including the freedom to diverge in some points from serate rules, but the normal way ( if as you say some interest from the community appears) is to modify skfr and to produce a new release.

So far, I did not review the code for nested levels, but this is more because I don't need it in my current work, and also because I have still to think more on what to do to improve that level.

In fact, I am much more interested in solving through exotic patterns instead of applying the boring nested nets. My current new code is a mix of skfr up to dynamic plus and of exotic patterns. Next steps if needed will be to reintroduce my previous code. I have no use of the serate nested process.


champagne

[Patrice]

::: comment :::

later this summer I'll have some time to update sudoku code
one of the things on my todo list is to finally replace java SE with a leaner but as-close-as-possible app
that is, for personal use and for the patterns game

a friendly competition between implementations might be a good diversion
faster rating might be one less impediment for new players to join the game

one criteria for using a new rater in the game would be to document all inconsistencies in the game history
between SE and the new rater -- few or no inconsistencies would be a big win, but speed could outweigh that

of course any rater considered would have to be available to the sudoku community

You should open a new thread inviting people to submit proposals !

Regards
Patrice

[m_b_metcalf]

::: comment :::

But now there too few players ! How can we attract more players ?

That I don't know, but maybe some of our spectators and puzzle providers (coloin, tarek) could tell us why they don't participate. But think of the hurdles:

the Game is in an obscure corner of the internet (and needs more publicity?);

some formidable amount of programming effort is required in order to be able to join in;

a lot of time is required in order to follow the Game (or to code a bot);

a powerful computer is required.

I note that interest in sudoku in general has waned over the past few years. Maybe that plays a role.

My own programs seem to work better on some patterns than on others. I played game 176 on a tiny, single-core laptop and still came in at a respectable spot having even been in the lead at some point. So horse-power is not essential to be able to join in, but is necessary to be able to win -- I break down consistently in the final stages of a game.

Regards,

Mike Metcalf

[Pat]

My own programs seem to work better on some patterns than on others

layout 177 reached 10.4/10.4/9.2

any idea why layout 178 would only reach 10.0/10.0/3.4 ?

Code: Select all

. O . . . O : . .
O . . . O . O . .
. . . . . . . O :
. . . . O . . . O
. O . O . O . O .
O . . . O . O . .
: O . . . O . . .
. . O . O . . . O
. . : O . . . O .


[g.r.emlin]

::: game 0179 start at 2012-06-25+16:00:00-0000 :::

Code: Select all

dealer status

candidate patterns (via private message to g.r.emlin)

  23 tarek            87 g.r.emlin

past patterns

   1 claudiarabia      1 ravel             6 ronk             22 m_b_metcalf      35 tarek
   1 draco             1 wintder           9 gsf              23 Pat
   1 Eioru             3 g.r.emlin        19 TTHsieh          27 Mauricio
   1 pseudocoup        6 champagne        20 coloin           33 JPF


[tarek]

::: comment :::

But now there too few players ! How can we attract more players ?

That I don't know, but maybe some of our spectators and puzzle providers (coloin, tarek) could tell us why they don't participate.

Extra Time seems to be a luxury for me lately. This game as JPF mentioned some time ago will require dedicated time. Another thing is that this is a GAME. So you need to be enjoying it. I'm not enjoying it anymore. Targeting new players or programmers is probably a good idea.

tarek

[g.r.emlin]

::: game 0179 open 2012-06-25+16:00:08-0000 dealer tarek 25 clues diagonal (main) symmetry :::

Code: Select all

. 3 . . . 5 . . 4
8 . . . . . . 1 .
. . 1 9 8 . . . .
. . 9 . . . . 8 7
. . 2 . . . 6 . .
3 . . . . . 4 . .
. . . . 9 7 5 . .
. 1 . 2 . . . 4 3
2 . . 3 . . . 7 .  ED=2.3/1.2/1.2