The New Sudoku Players' Forum

[champagne]

But I persist that there are still too many degrees of freedom how to proceed.

Happily yes, without such degrees of freedom, the game would be dead for long

[coloin]

Another point .....

A while ago there were a shortage of players.
It seems we have more enthusiastic players now ....
The game would also completely change if we had say for example 10 players or more ! Or would it ?

I could easily publish my own idiots guide to making puzzles
with links to program download and specific command lines for each which work

do we really want that ?

[m_b_metcalf]

Here are some more statistics. The table below shows the maximum rating achieved for each of the possible clue counts. Thus, in the current game, the existing record is 11.2/11.2/10.1. A target to beat.

Regards,

Mike Metcalf

Code: Select all

  #    ER   EP   ED
clues
 19   9.3  9.3  3.4
 20  11.8 11.8  3.4
 21  11.3 11.3  3.4
 22  11.7  1.2  1.2
 23  11.4 11.4 10.8
 24  11.3 11.3 10.6
 25  11.3 11.3 11.1
 26  11.1 11.1 10.4
 27  10.7 10.7  8.9
 28  10.5 10.5  9.2
 29   9.5  1.2  1.2


[m_b_metcalf]

I have updated the table in the previous post to take account of the new record that papy999 has set. Interestingly, it beats the record for 24 clues too.

M

[champagne]

some comments on the game 315

The game is not yet closed, but the last steps will not change much the picture.

With 25 clues, the pattern has hundreds millions of valid puzzles. Although I ignored about 80% of the produced valid puzzles, I analyzed (pre rated) about 100 million grids.

Most of them did not pass the pre rating although the pattern produced many many "quasi diamonds and pearls". A rough estimate of the grids rated through skfr would be 10% at most of the analyzed grids, say 2% of the produced grids.

And less than 1% of these 2% grids have been rated using serate.

The first filter (80% of the produced grids) has a cost close to nil compared to serate.
The second filter (quick rating) is "several times" faster than skfr
and skfr is several 10th times faster than serate.

Meantime, the first filter has been significantly improved. The sources are not clean enough to be shared to-day, but I can easily share a .exe if anybody is interested.

IMO, such pattern can only be processed in one of the 2 ways

- classical tools with a huge power
- tolls similar to what I did to save ressources.

[papy999]

Some questions about "prerating"

In the game I think there are 3 fields
- low rating under 6.5 with rules that do not use chains
- intermediate rating between 6.5 and 9.5 : using all kinds of chains but no nested level
- high rating with nested levels

Usualy the intermediate rating comes very early in the game with some exception : 8.1 for instance

So there are 2 fields where tools and strategy will make the difference low rating and very high rating

For the low rating the best tool is a very fast limited rating tools that can run 100x fastest that serate... I don't see what a prerating tools can bring here ?!

I think that your prerating tools is very usefull for very high rating, particularly in a game like 315 where we are overflowded by very high rating puzzles.
But what are the basis of your prerating ?
- searching for near diamonds or near pearls based only on the length of chains ?
- how can you take in account the nested levels ?

I'm very interested on your algorithm !
Can we share ?

[champagne]

Hi papy999,

As a preliminary statement, what I am doing has sense mainly in games where you have many puzzles and many diamonds pearls.

BTW, other games are of small interest for what I am doing.

Agreed that for low ratings (say below 6.5) the priority is to have a very fast solving code. My work in progress is just there.
A point that you missed in your wording, in most cases, puzzles with a low ED will play no role. You can store them "just in case".

What I call pre rating is a code ignoring the length constraint.

All steps using the same chain level receive the same rating eg : 7.0 for a xchain, 7.6 for a nishio elimination ...
This pre rating is much faster we then get what I call "pseudo diamonds", "pseudo pearls"

At the beginning, the priority is given in my search to diamonds to search in pseudo diamonds.

If the pattern delivers all diamonds belonging to a group (for example 8.0 to 8.6) then the pearsl of this group can just be sent to the bin.
And if a group is covere in diamonds and pearls, other puzzles of the group are ignored.

So it's not true to write that in this area 6.2 to 9.5 the pre rating has no interest

And in the same area the ratings where diamonds are difficult to find are always the same 6.2 6.5 6.9 7.7 8.0 8.1 belonging to the same 2 groups in the pre rating.

But the most important is to postpone (usually for ever) as many puzzles as possible without any rating

And no problem to share more. My interest in the games remains the search of highest ratings.

[papy999]

Another point .....

A while ago there were a shortage of players.
It seems we have more enthusiastic players now ....
The game would also completely change if we had say for example 10 players or more ! Or would it ?

I could easily publish my own idiots guide to making puzzles
with links to program download and specific command lines for each which work

do we really want that ?

HI coloin

The number of regular players is quite low :
m_b_metcalf, Pat, champagne, dobrichev, papy999
coloin and Patrice (no news since his laptop broke)
200e200w came and has gone (patterns submission and player)

I think the story of 200e200w shows that
- the level of actual players is high - very difficult to win against champagne or dobrichev
- the best players uses their own always evoluting software (champagne dobrichev Patrice)
or have a lot of historical data that helps them (m_b_metcalf)
- to enter in the game : its difficult to find the appropriate software and mostly their manual
- new comers should be helped ... if not they will go away rapidly

So yes I think that publishing some reference software and the way to use them is mandatory to attract new players
A game with 10 players would be very different of the actual ones

[dobrichev]

Hi papy999,

After Glenn Fowler (aka gsf in this forum) I am the only one that keeps the whole used software in public space and adequately answers the questions (or at least I think so).
For example, after reading this single page introduction posted in this forum, how much time a motivated and literate newcomer should spend to join to the game? I think no more than a week, and after few played games he/she could be a real competitor.

Some players have published code but their explanations are more or less encrypted. Nothing wrong there, all this is for free and for fun.

The differences in the used s/w are maybe the only thing that motivates the players at present. Else the game would become computer vs computer competition, bitcoins mining, or similar.

Hopefully the given reference is of use for you and Colin.

Cheers,
MD

[papy999]

The differences in the used s/w are maybe the only thing that motivates the players at present. Else the game would become computer vs computer competition, bitcoins mining, or similar.

May be you're right

For me a part of the game is to find new strategies and algorithm to explore the very large number of puzzle that can be compliant with a pattern.
But as in every game there is a part of chance also

BTW : your link "how to use gridchecker for gamepattern" is interesting ... I was trying to find how to use it ...
Thanks

papy999

[champagne]

I think the story of 200e200w shows that
- the level of actual players is high

I think that you can say that but they don't use all their potential to win.

- the best players uses their own always evoluting software (champagne dobrichev Patrice)

Sure, the game is a platform used to test some rating tools used in other tasks (at least for dobrichev and me).

- to enter in the game : its difficult to find the appropriate software and mostly their manual
- new comers should be helped ... if not they will go away rapidly

I delivered to enxio27 some .exe and the process used by me at that time, it has not been sufficient to attract him.
The problem is more complex. Newcomers are interested if they are bringing something in. Not easy in a mature business. This is a wider situation in the sudolu field. In the first years, anybody could come with fresh ideas. To-day, you have first to digest what has been done during years.

[champagne]

Here above an example where skfr rates 7.7 and serate 7.8

Code: Select all

1....2..3
.4..5..6.
..27..1..
..5.....8
.9.....7.
8.....9..
2....83..
.8..6..4.
..39....5;7.7;7.7;7.7;7.8;7.8;7.8

Code: Select all

... ..2 ...   
... ... 2.2 
..2 ... ...

.2. 22. 22.
... 22. 2.2
.2. 22. .22

2.. ... ...
... 2.. 2.2
... .2. 22.

Both will show that 2r4c4 can't be true, leading to a column 8 empty

serate writes (something equivalent to)

Code: Select all

2r4c4 -> -2r4c8
      -> -2(r456c5)  ->2r9c5 -> -2r9c8
      -> -2r4c2 ->2r6c2-> -2r6c8

for the second line, skfr chooses

Code: Select all

2r4c4 -> -2r8c4 ->2r9c5 -> -2r6c8


saving 2 in the total length, which makes the difference in the rating.

I am now revising my code in this area and I'll try to understand the serate sequence (likely the same used in the dynamic mode) to reduce the deviations in the rating

[champagne]

yes, this pattern {320} has plenty of (minimal) puzzles.

but all my puzzles go through SKFR
and that does use a significant amount of time.

the official rating is much more expensive --
so i only use it for a tiny portion of puzzles,
surely below 1%

some comments to pat's post in the pattern game.
My code revision goes very slow for many reasons, but with the process prepared

All the low ratings (1.0 to 4.4) are reduced at the generation time to one occurrence by ER/EP/ED higher than already seen in the batch
All low ED (say below 2.6) are put in "wait state"

To compare with other stats, this creates already two problems

. low ratings are not counted
. puzzles in wait state are not ED puzzles (not passed thru the canonicalization filter)

This process uses the brute force results and the first steps of the "serate solving clone".

Depending on the pattern, at that point, 50% to 90% of the created puzzles are just solved or put in wait state.

The rest is rated using my last tool, likely 10 times faster than SKFR.

I could have a fresh code available in test to process all puzzles rating below 8.3 for the next game, but this is not granted. If I am not ready, I'll skip that game as well.
As I said before, I'll release the code as soon as I reach the multi chains area (8.2 to 8.6_8.9)

[papy999]

Hi champagne,

With your promise the game will change a lot

I think I will have to drop down my own software

Will you release the source or just an exe ?
exe from an unknown source will not be accepted by Norton and there are some issue with C++ dll when you are not on the same level of Windows
and cannot run on Ubuntu as well

So it's will be nice to release the source

[champagne]

Will you release the source or just an exe ?

I'll release as much as possible:
sources, windows .exe; and comments on the overall process applied.
The code is C++ compiled with microsoft visual studio express 2013

exe from an unknown source will not be accepted by Norton and there are some issue with C++ dll when you are not on the same level of Windows
and cannot run on Ubuntu as well

SKFR had been designed to be open to DLL's and multi-thread operations.

As I don't use any of these tools, the current code could be hard to modify to produce a DLL or a multi-thread frame.
Regarding other compiling platforms, performance problems could appear if they don't accept "intrinsic" code. The design of this code uses intensively some intrinsic code as

popcount bitscan movs. instructions
128 bits sse2 set of instructions (but this already in SKFR)
...

and requires a 64 bits processor

BTW, I can email the current sources to anybody willing to see what kind of problem could appear in the platform in use for him
The current program frame will not change