Patterns Game Strategies

Interactive on-site game threads go here

Re: Patterns Game Strategies

Postby champagne » Mon Sep 20, 2010 9:22 am

I have been asked thru PM to comment a little more on the “ –n+n” strategy.
I’ll do it on game 118, focusing on the last cycle.

In fact, I have the feeling that some of the other players are already doing something similar.

Let start with “n” value.

“n” value could be any number from 2 to 10 if not more.

“2” is very short to run, “3” relatively short as well, but they are not very efficient.
I use them at the start of a game mainly to open the search field.

“4” seems to be the best compromise. “5” is already very long.

I never experienced more than “5”. I think such numbers should be used with part of the clues not changed, whatever would be the reason.

In fact, to delay the moment where the generation does not produce new puzzles, I mix “4” and “5”.

Defining a batch.

For me, a batch includes about 30 to 100 puzzles (except for the two first runs).
This seems again a good compromise in the pattern game. In a free search, I, may be, would have bigger batches.

In game 118, the eighth batch was over 100 puzzles (I give it at the end of the post).
Choosing the puzzles is surely part of the strategy. In that game, I just took high “se rating” of the last runs.

The best strategy is not easy to define. One must cover if possible all the field. I have no idea on how to prevent the search to be locked in a sub field of the pattern potential.

The process is more or less over if all generated puzzles are morphs of known ones.

The last lot of game 118.

In game 118, one could expect some saturation effect at cycle 8. Reversely, I got an amazing result. It took me several days to extract and rate what I wanted to check,(but I used very limited resources after the game closure).

I have published as ::: more ::: the new low ratings. The game 119 started before I had closed the ratings in the high part.

The lot of puzzles selected as potential high rating included 1596 puzzles.
(note: I made somewhere a wrong manipulation in data handling. In fact 109 puzzles had already been rated, including the highest one).

I got 2 new interesting highest

Code: Select all
6..1...78......5.9.....861...9..48...3..2....4..7....5..4..7....2..3....7..6....1  ED=10.4/10.4/10.1
1..2...39......1.4.....426...8..36...7..5....3..4....1..9..7....5..8....2..3....6  ED=10.3/10.3/10.0


But the most surprising has been the density of puzzles rating more than 9. Here the table giving the number of puzzles per ED in that lot

Code: Select all
10.60  118
10.50  209
10.40  245
10.30   81
10.20   30
10.10   62
10.00  153


No surprise if the rating lasted so long.



I did not play in the ongoing game 119, but in such a game, the process saturate very soon. I doubt anybody will find more than the interesting highest rate found by gsf.

champagne

Batch used for cycle 8

Code: Select all
7..8...94......7.8.....465...8..15...3..2....5..9....6..1..9....2..3....4..6....7 10.6 10.6 10.40
6..5...24......5.8.....469...7..68...3..2....1..4....9..1..7....2..3....9..8....6 10.6 10.6 9.50
8..1...93......6.4.....751...9..48...3..2....5..9....6..4..9....2..3....1..6....7 10.6 10.6 9.40
1..6...39......1.4.....126...7..94...8..5....5..7....2..9..8....4..7....2..3....6 10.6 10.6 9.40
1..2...39......6.1.....325...4..71...8..9....7..6....2..3..5....9..4....8..7....6 10.6 10.6 9.40
4..3...97......5.4.....963...7..54...2..1....5..6....3..1..2....8..7....6..5....9 10.5 10.5 10.50
6..1...79......5.8.....561...9..48...3..2....4..8....5..4..9....2..3....7..6....1 10.5 10.5 10.00
9..2...34......7.2.....469...5..12...2..8....1..9....6..1..5....8..7....4..6....3 10.5 10.5 9.80
1..2...93......1.4.....426...7..36...8..5....3..4....9..9..7....5..8....2..3....6 10.5 10.5 9.40
6..9...31......6.4.....897...9..34...1..2....5..8....6..2..7....8..1....4..5....3 10.5 10.5 9.40
1..2...64......1.7.....423...7..54...8..9....3..7....2..9..7....5..8....4..6....1 10.5 10.5 9.40
1..5...43......2.1.....365...4..91...7..8....6..3....2..9..4....8..7....5..9....6 10.5 10.5 9.40
5..6...14......2.8.....463...7..31...8..9....6..7....2..4..7....9..8....2..1....3 10.5 10.5 9.40
6..2...74......5.8.....461...7..18...3..9....2..5....6..4..9....8..3....9..7....5 10.5 10.5 9.40
9..3...74......5.9.....763...7..54...1..8....5..2....6..4..9....8..1....6..5....3 10.5 10.5 9.40
4..7...38......6.4.....397...9..24...1..5....3..8....6..2..6....8..1....1..5....7 10.5 10.5 9.30
2..6...54......3.1.....326...4..75...8..9....7..1....2..1..4....9..8....3..7....6 10.5 10.5 9.30
2..1...98......2.1.....356...4..91...7..8....6..5....2..9..4....8..7....4..6....3 10.5 10.5 9.30
6..1...94......5.8.....961...9..48...3..2....5..9....6..4..3....2..9....1..6....7 10.5 10.5 9.30
6..2...94......7.8.....465...8..12...3..7....5..8....6..1..9....2..3....4..6....7 10.5 10.5 9.30
8..1...79......5.8.....516...9..48...3..2....5..9....6..4..9....2..3....1..6....7 10.5 10.5 9.30
1..2...39......6.1.....325...4..71...8..9....7..6....2..3..4....9..5....8..7....6 10.5 10.5 9.30
6..8...74......5.8.....461...9..18...3..2....5..9....6..4..5....2..3....9..6....7 10.5 10.5 9.30
6..7...41......5.8.....469...7..68...3..2....5..1....4..2..9....4..3....7..8....5 10.5 10.5 9.20
1..7...54......3.7.....516...9..14...7..2....5..8....6..2..9....8..3....4..6....1 10.5 10.0 9.40
6..9...41......5.8.....496...7..68...3..2....5..1....9..1..7....2..3....4..8....5 10.4 10.4 10.00
2..5...61......2.7.....253...4..91...1..8....8..3....2..9..4....8..7....5..1....6 10.4 10.4 10.00
1..8...57......3.8.....516...9..17...7..2....5..4....6..2..9....8..3....7..6....1 10.4 10.4 9.90
8..1...74......5.8.....761...9..68...3..2....5..4....1..4..9....2..3....7..6....5 10.4 10.4 9.90
8..1...72......5.8.....761...3..48...9..2....5..9....6..4..9....2..3....1..6....7 10.4 10.4 9.90
1..2...93......1.4.....426...7..36...8..5....3..6....2..9..7....5..8....4..3....6 10.4 10.4 9.90
8..9...74......9.8.....856...7..14...3..2....5..4....7..2..3....1..7....6..5....9 10.4 10.4 9.90
3..4...95......4.2.....963...7..15...5..8....4..5....6..1..2....8..7....6..3....9 10.4 10.4 9.90
5..7...31......6.4.....657...9..54...1..2....7..8....6..2..9....8..1....1..4....3 10.4 10.4 9.70
1..6...74......2.5.....316...7..45...8..9....3..7....2..4..2....9..8....5..1....6 10.4 10.4 9.50
6..5...94......7.8.....465...7..15...3..2....9..8....6..8..9....1..3....4..6....7 10.4 10.4 9.40
1..6...54......1.3.....126...7..34...8..9....4..7....2..3..7....9..8....5..1....6 10.4 10.4 9.40
9..5...38......5.1.....396...4..91...7..2....8..3....6..2..4....8..7....5..6....9 10.4 10.4 9.40
1..6...34......9.7.....361...9..54...8..2....5..1....6..8..4....2..7....7..5....3 10.4 10.4 9.40
1..2...45......3.8.....516...9..18...7..2....5..4....6..2..9....8..3....6..7....1 10.4 10.4 9.40
2..3...97......2.4.....963...7..14...2..8....5..4....6..1..5....8..7....3..6....9 10.4 10.4 9.40
1..3...62......4.5.....683...7..51...9..2....5..8....4..6..9....2..7....4..5....8 10.4 10.4 9.40
5..6...31......2.5.....376...4..71...8..9....7..3....2..5..4....9..8....1..7....6 10.4 10.4 9.40
6..7...48......5.2.....467...7..68...3..2....5..4....9..1..9....2..3....9..8....5 10.4 10.4 9.40
8..6...14......9.8.....156...7..28...3..1....5..4....1..2..3....8..7....4..5....9 10.4 10.4 9.40
6..5...74......1.3.....426...1..35...8..9....3..7....2..4..7....9..8....2..1....6 10.4 10.4 9.40
9..2...34......5.2.....369...7..12...4..8....5..9....6..1..4....8..7....4..6....3 10.4 10.4 9.40
6..1...74......2.8.....761...9..48...3..5....5..2....6..4..9....2..3....1..5....7 10.4 10.4 9.30
1..6...74......1.5.....526...7..34...8..2....5..7....1..3..7....9..8....2..4....6 10.4 10.4 9.30
8..1...69......5.8.....571...9..48...3..2....5..9....6..4..9....2..3....1..6....7 10.4 10.4 9.30
3..2...97......4.2.....965...7..13...2..8....8..5....6..1..4....8..7....6..3....9 10.4 10.4 9.30
1..5...47......2.9.....356...4..71...7..8....3..4....2..9..4....8..1....5..2....6 10.4 10.4 9.30
1..2...36......2.5.....341...8..91...4..2....6..7....2..3..5....5..8....7..6....9 10.4 10.4 9.30
1..5...37......2.1.....356...4..91...2..8....9..3....6..5..4....8..7....7..6....9 10.4 10.4 9.30
5..2...39......2.5.....318...4..95...7..6....9..1....2..3..7....6..8....1..9....4 10.4 10.4 9.30
4..6...39......1.4.....126...7..54...8..9....1..7....2..5..8....9..7....2..3....6 10.4 10.4 9.30
8..1...24......5.8.....769...9..84...3..2....5..9....6..4..9....2..3....1..6....5 10.4 9.50 9.40
6..4...79......5.4.....761...9..58...3..2....5..8....6..4..9....2..3....7..6....1 10.3 10.3 9.90
2..6...31......2.4.....356...4..91...7..8....9..5....2..3..4....8..7....4..9....6 10.3 10.3 9.70
6..2...89......3.2.....956...7..19...4..3....8..9....6..1..7....5..4....4..8....3 10.3 10.3 9.50
1..2...38......2.4.....356...4..91...7..8....9..5....6..2..4....8..7....3..6....9 10.3 10.3 9.50
9..7...41......5.8.....469...7..68...3..2....1..5....9..4..9....2..3....6..8....5 10.3 10.3 9.50
6..5...74......5.2.....461...9..18...3..2....7..8....6..4..9....2..3....3..7....5 10.3 10.3 9.40
8..1...75......3.4.....761...9..48...3..2....5..9....6..4..9....2..1....1..8....7 10.3 10.3 9.40
1..2...34......5.7.....321...8..91...4..2....6..7....2..5..8....6..4....7..6....9 10.3 10.3 9.40
2..3...97......6.4.....923...3..14...7..8....5..4....6..1..2....8..7....6..5....9 10.3 10.3 9.40
5..6...13......7.5.....326...4..75...8..9....7..1....2..2..4....9..8....3..7....6 10.3 10.3 9.40
9..2...54......3.2.....469...7..12...3..8....5..9....6..1..7....8..6....4..5....3 10.3 10.3 9.30
2..3...94......5.2.....563...7..14...2..8....5..4....6..1..2....8..7....3..6....9 10.3 10.3 9.30
4..5...63......5.1.....329...3..91...7..8....9..1....2..9..4....8..7....5..2....6 10.3 9.90 9.40
6..4...38......6.4.....651...9..24...1..7....5..8....6..2..8....7..9....1..5....3 10.3 9.80 9.30
6..8...79......5.8.....761...3..28...9..8....4..5....6..4..9....2..3....7..6....1 10.2 10.2 9.40
2..4...38......5.4.....362...9..74...1..2....5..6....3..7..9....3..1....1..5....6 10.2 10.2 9.40
6..7...84......5.6.....817...8..16...2..9....1..3....5..4..3....9..2....5..4....7 10.2 9.30 9.20
8..1...79......1.5.....768...9..48...3..2....5..9....6..4..9....2..3....1..6....7 10.1 10.1 9.90
4..3...97......5.4.....963...7..14...2..8....3..5....6..1..2....8..7....6..8....9 10.1 10.1 9.80
1..8...79......5.8.....761...3..28...9..4....4..5....6..4..9....2..3....7..6....1 10.1 10.1 9.80
4..5...37......2.1.....356...2..91...7..8....9..3....6..1..4....8..7....5..6....9 10.1 10.1 9.80
8..1...74......5.8.....761...9..58...3..2....5..4....6..1..9....2..3....6..5....7 10.1 10.1 9.80
4..6...39......1.4.....362...7..14...8..9....1..3....2..9..7....5..8....2..1....6 10.1 10.1 9.40
1..6...38......1.7.....326...7..93...8..5....4..7....2..9..1....5..8....2..5....6 10.1 10.1 9.40
5..6...74......1.5.....126...7..24...8..9....3..7....1..5..7....9..8....4..3....6 10.1 10.1 9.40
6..7...34......5.6.....617...5..14...2..9....3..8....5..6..8....9..2....5..3....7 10.1 10.1 9.30
1..6...54......3.8.....516...9..18...7..2....3..5....6..2..9....3..8....4..7....1 10.1 10.1 9.30
2..6...39......2.1.....356...4..81...7..6....9..5....3..1..4....8..7....4..9....5 10.1 10.1 9.30
1..5...49......5.1.....427...3..91...7..8....9..4....2..9..6....8..3....5..2....6 10.1 10.1 9.30
1..7...54......3.8.....516...9..18...7..2....5..4....6..2..9....4..3....6..8....1 10.1 10.1 9.20
1..5...34......5.1.....396...4..21...7..8....3..4....9..8..6....2..7....5..9....6 10.0 10.0 9.90
6..1...74......1.8.....865...9..48...3..2....5..9....6..4..9....2..3....1..5....7 10.0 10.0 9.90
5..2...73......6.1.....325...4..71...8..9....6..5....2..1..4....9..8....3..7....6 10.0 10.0 9.90
1..6...74......1.5.....526...7..34...8..9....3..7....2..4..7....9..8....5..2....6 10.0 10.0 9.90
1..9...34......1.9.....326...7..19...8..5....6..4....2..4..7....5..8....2..6....3 9.90 9.90 9.90
4..6...39......1.4.....326...8..54...7..9....1..7....2..5..7....9..8....2..3....6 9.90 9.90 9.90
6..5...43......2.1.....356...3..61...7..8....4..1....2..9..4....8..7....5..9....6 9.90 9.90 9.90
5..2...31......2.5.....376...4..75...8..9....7..1....2..6..4....9..8....1..7....6 9.90 9.90 9.90
8..1...74......5.8.....761...9..28...3..1....5..4....6..4..9....2..3....7..6....5 9.90 9.90 9.90
6..7...31......6.4.....357...9..54...1..2....5..8....6..2..9....8..1....4..5....3 9.90 9.90 9.90
6..4...38......5.7.....364...9..27...1..5....4..1....6..2..9....8..1....1..7....3 9.90 9.90 9.90
8..6...34......9.8.....365...7..14...8..2....5..4....9..1..8....2..7....9..5....3 9.90 9.90 9.90
9..3...24......5.9.....563...7..13...2..8....5..4....6..1..7....8..5....6..2....3 9.90 9.90 9.90
champagne
2017 Supporter
 
Posts: 5650
Joined: 02 August 2007
Location: France Brittany

PG 122

Postby surbier » Fri Oct 15, 2010 7:31 am

Hi,

This is a slightly off-topic question, and only loosely related to
Patterns Game Strategies.
Better I post it here than in the grid submission thread for game 122.

Could somebody of the PG players confirm:

My following findings for pattern game 122

Code: Select all
000003008007060020090700300008000003010090040700000900006004010030010200200600000 6.6/1.2/1.2
000009005006070010030100600003000009020050080700000300009002050080060700500400000 8.9/1.2/1.2
000004009005010070030600400006000003070080010100060200019006080020070100800100000 7.3/1.2/1.2
000007009007010080030400200008000001050080070400000500004005090020090600100300000 7.3/1.2/1.2
000006001009070050030100800007000005050030090400000700005001020090080400100300000 1.5/1.2/1.2
000003002005040010030800900006000001040060050300000700001005020080030400900400000 7.3/1.2/1.2
000007008001080070050900200007000005080020030900000400004005060010030500600800000 7.4/1.2/1.2
000003002004050080020800600006000008040060030300000900008005070010020400900100000 7.1/1.2/1.2
000009001001070090050800600003000002080010050600000900008007040010040300900500000 8.3/1.2/1.2
000004005008090070060300400009000008040080020300000100005009010090050300100400000 8.3/1.2/1.2
000003002007020060010800900009000003050060020700000100006008010020010800400900000 7.2/1.2/1.2
000006001001090030040100700007000008060020090200000400002004050090050200800700000 7.8/1.2/1.2
000003002007020060010800900009000003050060020700000100006008010020010800400900000 7.2/1.2/1.2
000008004007060030010200700009000002070020090800000400001004050050030800200100000 2.6/1.2/1.2
000003004003010060050200700008000006090050070600000500001007090070020800200600000 6.6/1.2/1.2
000005002007020080030400100005000004060080050100000300003001040010070900600200000 7.2/1.2/1.2
000006005002090080040500100004000003030080010500000200003004090020010700600300000 2.8/1.2/1.2
000001003005080070080900400009000004010070080200000900007004050060030100300700000 6.6/1.2/1.2
000005002008030040020800900003000009080040010400000700004002070010060300900500000 9.2/1.2/1.2
000001006002050030010200800009000004030070060400000200007005080040060100200800000 4.2/1.2/1.2
000007005005040080030900700001000007020060010600000900007003040090070600100200000 7.6/1.2/1.2


are calculated insane (in the spirit of the competitive character of the pattern game).
They all are already trumped, and it does not make sense to show them later after
PG 122 will have been finished under the
Code: Select all
::: more :::

tag (section).

If I (my code) stays on the level of producing only secont rate ER/EP/ED = X.X/1.2/1.2 grids,
I will continuously be trumped by grids with EP>1.2, ED>1.2 and I will not get any score > 0 ?

Given in addition the amount of compute power consumption for my addmitedly less optimized code,
this is a real challanging game.
surbier
2012 Supporter
 
Posts: 54
Joined: 06 June 2008

Re: PG 122

Postby champagne » Fri Oct 15, 2010 8:15 am

surbier wrote:Hi,

This is a slightly off-topic question, and only loosely related to
Patterns Game Strategies.
Better I post it here than in the grid submission thread for game 122.

Could somebody of the PG players confirm:

My following findings for pattern game 122
are calculated insane (in the spirit of the competitive character of the pattern game).


If I (my code) stays on the level of producing only secont rate ER/EP/ED = X.X/1.2/1.2 grids,
I will continuously be trumped by grids with EP>1.2, ED>1.2 and I will not get any score > 0 ?

Given in addition the amount of compute power consumption for my addmitedly less optimized code,
this is a real challanging game.



some of your puzzles are not minimal.

these ones would have been valid

Code: Select all
000003008007060020090700300008000003010090040700000900006004010030010200200600000ED=6.6/1.2/1.2
000009005006070010030100600003000009020050080700000300009002050080060700500400000ED=8.9/1.2/1.2
000007009007010080030400200008000001050080070400000500004005090020090600100300000ED=7.3/1.2/1.2
000006001009070050030100800007000005050030090400000700005001020090080400100300000ED=1.5/1.2/1.2
000003002005040010030800900006000001040060050300000700001005020080030400900400000ED=7.3/1.2/1.2
000007008001080070050900200007000005080020030900000400004005060010030500600800000ED=7.4/1.2/1.2
000003002004050080020800600006000008040060030300000900008005070010020400900100000ED=7.1/1.2/1.2
000009001001070090050800600003000002080010050600000900008007040010040300900500000ED=8.3/1.2/1.2
000004005008090070060300400009000008040080020300000100005009010090050300100400000ED=8.3/1.2/1.2
000006001001090030040100700007000008060020090200000400002004050090050200800700000ED=7.8/1.2/1.2
000008004007060030010200700009000002070020090800000400001004050050030800200100000ED=2.6/1.2/1.2
000006005002090080040500100004000003030080010500000200003004090020010700600300000ED=2.8/1.2/1.2
000001003005080070080900400009000004010070080200000900007004050060030100300700000ED=6.6/1.2/1.2
000005002008030040020800900003000009080040010400000700004002070010060300900500000ED=9.2/1.2/1.2
000001006002050030010200800009000004030070060400000200007005080040060100200800000ED=4.2/1.2/1.2
000007005005040080030900700001000007020060010600000900007003040090070600100200000ED=7.6/1.2/1.2


regarding the generation process some remarks:

1) it is anyway a cpu consuming work.

2) to give an order of magnitude, I found up to now about 6.7 millions diferent puzzles in game 122 (can be doubled or tripled to have raw puzzles)
this means that as soon as your process is improving, you are facing a volume issue.
I let you compute the average per hour.

3) I described in that thread some efficient processes, you should rally to the generation -+n if this is not what you are doing.

4) we are thinking of ways to help newcomers. One, not yet discussed, but that I experienced in game 122, is for skilled players not to play in the lowest part of the game during a first phase (say one or two days).

but newcomers can have ideas as well;

champagne
champagne
2017 Supporter
 
Posts: 5650
Joined: 02 August 2007
Location: France Brittany

Re: Patterns Game Strategies

Postby surbier » Fri Oct 15, 2010 1:51 pm

I agree with you, that some of the puzzles I posted above are still not minimal.
These are indeed older solutions from my generator before I had integrated the minimal test.

3) I described in that thread some efficient processes, you should rally to the generation -+n if this is not what you are doing.


I was reading this thread more carefully and I think I'm beginning to understand some of the tricks.
In your post you describe how to find a reasonable choice of -n +n.
But it is not said what the '-n +n process' stands for.
Could you give a short hint ?

thanks in advance

PS: does it have something to do with Nearest-Neighbor search when starting from a solved
seeds grid ?
surbier
2012 Supporter
 
Posts: 54
Joined: 06 June 2008

Re: Patterns Game Strategies

Postby champagne » Fri Oct 15, 2010 4:13 pm

surbier wrote:I was reading this thread more carefully and I think I'm beginning to understand some of the tricks.
In your post you describe how to find a reasonable choice of -n +n.
But it is not said what the '-n +n process' stands for.
Could you give a short hint ?

thanks in advance

PS: does it have something to do with Nearest-Neighbor search when starting from a solved
seeds grid ?



I think you got it.

The underlying idea is that in the vicinity of an existing solution, you have others.
And it works.

I can add that if you have a solution with a value of ED, you have fullfilled some specific conditions and you have good chances to find other puzzles with the same ED in the vicinity.

The process itself (in my case) works in blind mode.

eg:-4+4 take any quad in the puzzle and try all other digits.

So the number of possibilities to test grows rapidly with 'n'.

Looking for a non blind mode is not so easy, so in fact, the best central choice is -4+4.

This is not enough if one wants to eliminate as soon as possible solutions not in the targee of what he is looking for.

As I am generally looking for high ED, I stop generation as soon as a partial solution is already below the target.

Pre checking minimality and no single is very easy. Above, you need an integrated process using a solver.

champagne
champagne
2017 Supporter
 
Posts: 5650
Joined: 02 August 2007
Location: France Brittany

Re: Patterns Game Strategies

Postby surbier » Fri Oct 15, 2010 8:06 pm

Thanks for the clarification.
This makes all sense to me.

I entered the PG with methods
a) generate a random solved grid, take off all clues according to the pattern, and perfrom the unique solution test.
This simple monte carlo method works for patterns with more than ~ 30 clues.
b) set iteratively/recursively clues in the cells defined by pattern until a grid with a unique solution is found and stop.

I could improve b) by continuing at the point a solution is found.
The only optimization I once have introduced is on the order the pattern cells are filled.
For each cell of the pattern I calculate the number of pattern cells in their peers, as a measure for the number
of constraints per pattern cell. I start the iteration (from scratch, none of the clues set) with the most constraint cells.

Since the number of constraints per pattern cell is constant with the pattern
(it is just geometry) and independent on the specific unique solution grid,
the information is available at run-time to decidw which of the n=2-4 clues to delete to iterate for further
in a non-blind manner, searching for more unique solution grids.
surbier
2012 Supporter
 
Posts: 54
Joined: 06 June 2008

Re: Patterns Game Strategies

Postby champagne » Mon Apr 04, 2011 12:43 pm

papy999 wrote:::: comment :::
I've received congratulations three times during this game... 2 times from joel64, once from champagne

Thanks a lot, it was a difficult game for me, I Started very late and I had to pile post on post to come back

I will not participate to the next ones. My computer stayed at home and since Saturday I had just some puzzles in cache to submit with an old laptop
I will come back in May.

champagne : I'm interested about your trick to select "promising puzzles"... meet you in "Patterns Game Strategies" stream.

Best regards
papy


Hi papy999,

If I consider the game 137, I am not sure at all I have anything to bring you to improve your process.

I have the feeling that, unless I use my best process, which means high time consumption, you are at least as quick as I have been to find the highest rating.

Game 137 was somehow special in that point that, as far as I could see, all high ratings have been found with a very limited number of patterns in the middle band, all of them with one of the lines 4 or 6 having one of the digits in box 5 line 5.

I made the last cycle based on that property, but it seems that all high ratings of that kind were already there.
I tested again a systematic search, but the results were highly disappointing. It looks like the overall best strategy is to search in the vicinity of existing puzzles.

Usually, you have many different pattern leading to high ratings, but no clear rule to locate them.
So I could split my strategy in 2 phases (plus a zero phase if I am prepared to spend more time):

phase0: Morph the puzzle in a way the canonical form is not too far from the working form.
This works pretty well, but one have to be very cautious to morph back the puzzle.


Phase 1 not very selective: expand +-3 +-4 and cycle with some relatively high rating. Try to cover as much space as possible.

Phase 2: in that phase, i try to focus on the highest rating authorized by the pattern.

I filter the generation to keep only puzzles reaching an high ER after 2 resolution steps.
I now use for that my SEclone in preparation.
I add highest ED
If the number of puzzles is big, I diminish the size of the lot trying to keep some puzzles in each area of the game

I extend the +-n process to +-5 and over on the best puzzles of the lot, keeping unchanged part of the seed.

Note: if a morphed form has been used, it is relatively easy to sort out the next cycle out of the canonical form.
This makes much easier the coverage of the entire field;

I think this is a good description of how I am working (or thinking of working) to-day.

I am sure that questions will come if necessary

champagne
champagne
2017 Supporter
 
Posts: 5650
Joined: 02 August 2007
Location: France Brittany

Re: Patterns Game Strategies

Postby papy999 » Mon Apr 04, 2011 2:47 pm

Hello champagne,

I'm not sure to understand your process
phase0: Morph the puzzle in a way the canonical form is not too far from the working form.
This works pretty well, but one have to be very cautious to morph back the puzzle.

What is the use of this phase ?

I now use for that my SEclone in preparation

Do you have a software that mimics Serate but is quicker ?
Does your software give exactly same rating as Serate ?
If the answer is yes to the 2 questions, you have a terrific advantage !

Your description seems to focus on highest ratings ?
What do you do to find for instance puzzles in the range 4.0 to 7.0 ?

For my side,
I use a generation based on {-n+n} with an integrated "one solution and minimal verification"
I discard all puzzles with a single as first move (with my own test)
I keep only the first found puzzles between the different isomorph forms (even in some cases different morph form can have different rating)

I filter with a modified version of Serate that enable to stop Serate if Ed doesn't meet a given condition, or if it is not a diamond or not a pearl, this means I've in my database puzzles with partial or full rating. I use this for the seed choice !

I have a generation based on "one" seed process, so I use n=4, n=5 or in the night n=6, (i can't process a batch at the moment) I cancel this generation if the rate of generation is too low (number of new puzzles / s) especially at the beginning of the game.
I have to relaunch that process too many times because I process seeds one by one and I have no queued processes. This enables also to analyze the results immediately.

For the seed, I choose a high ED or a high ER/EP/ED in my puzzles database or in the other players posts to find the highest rating
and some condition on puzzles that have their first steps rated like
er<7.0 and ed>4.0 to find puzzles in the field 4.0 - 7.0 !
In the last game I tried to use rare rating as seed, it gave some good results

Well, I think I will include a batch process or a queuing process to facilitate the generation.

I'm also interested in SE clone, the Serate slowness is the cornerstone of that game !

I am far from home with an old laptop for the next month, no time and no tools to modify my software. I'll will contact you again in May.
By the way : I use exclusively Java and I'm too old to try an other language !

Best regards and good games !
papy is not yet decrepit :lol:
papy999
 
Posts: 92
Joined: 15 September 2010

Re: Patterns Game Strategies

Postby papy999 » Mon Apr 04, 2011 3:03 pm

champagne,

I have a question

I think it would be interesting to have a program that could calculate the "isomorph distance" between 2 puzzles that have the same pattern :

A definition ? The minimum number of givens that you must change in the first puzzle to have an isomorph form of the second puzzle.

To explore new zone in puzzles generation I am interested to take seeds that are as far as possible (with my definition) from all already found puzzles. This means that the generation will produce a lot of new non isomorph puzzles.
At the moment for some generation I have more than 95% of generated puzzles that are isomorph of a previous one. This means that the seed was not far enough from previous found puzzles !

What do you think of this idea ?
I've tried to find a efficient algorithm for this calculation - not yet found. If you can help me !!!
papy999
 
Posts: 92
Joined: 15 September 2010

Re: Patterns Game Strategies

Postby ronk » Mon Apr 04, 2011 3:53 pm

papy999 wrote:I think it would be interesting to have a program that could calculate the "isomorph distance" between 2 puzzles that have the same pattern :

A definition ? The minimum number of givens that you must change in the first puzzle to have an isomorph form of the second puzzle.

To explore new zone in puzzles generation I am interested to take seeds that are as far as possible (with my definition) from all already found puzzles. This means that the generation will produce a lot of new non isomorph puzzles.

I have no idea what you mean by "isomorph distance." An isomorph would have zero distance, by definition I think. I suspect you're referring to Hamming distance, which can be obtained with gsf's software.
ronk
2012 Supporter
 
Posts: 4764
Joined: 02 November 2005
Location: Southeastern USA

Re: Patterns Game Strategies

Postby champagne » Mon Apr 04, 2011 3:57 pm

Hello papy999,

Have a good time during the next month.

I answer to some of your points





papy999 wrote:
I now use for that my SEclone in preparation

Do you have a software that mimics Serate but is quicker ?
Does your software give exactly same rating as Serate ?


Cloning “serate” is an old dream. You’ll find long discussions in that thread;

http://forum.enjoysudoku.com/team-project-c-or-c-explainer-like-rating-program-t30083.html

If you read the thread, you’ll see also that

1) I am working on producing such a clone based on my solver design
2) cloning 100% “serate” seems not realistic;

I cover for the time being the ratings 1.0 to 7.5 (slightly more) in a reasonable way, with a speed ratio somewhere in the area 1:30 and with additional functions very efficient in the game.

I tested that program for the first time in that game.

I am not alone in that task, but the group is not formal. The door is open to anybody willing to contribute. Already, lksudoku and Patrice brought some key feed to go ahead.

I use directly the rating of the clone (but I check the results) in that range except that in the area UR UL BUG I have small differences in the design, so I have to re rate the best puzzles.

Later, I have other deviations in Y cycles … so rerating is unavoidable. The clone just sort out the best puzzles;

papy999 wrote:
Your description seems to focus on highest ratings ?
What do you do to find for instance puzzles in the range 4.0 to 7.0 ?



Let me, thru that question cover several other points.

I don’t play specifically to win the game, it’s to much time consuming
So I focus on the highest part of the game as you understand. This is also in line with my wider research on Sudoku.

I get puzzles in the lowest range

- at the start of the game
- later on in the generation just due to the fact that I de activate the option to eliminate puzzles having low ED for the last cells.

Each cycle includes a significant number of puzzles (usually several tens to some hundreds), processed in batch mode.

papy999 wrote:I'm not sure to understand your process
phase0: Morph the puzzle in a way the canonical form is not too far from the working form.
This works pretty well, but one have to be very cautious to morph back the puzzle.

What is the use of this phase ?


Processing batches of a significant size, I find easier to work on a form as close as possible of the canonical form.

In that example, to be more efficient, I should at least have morphed the given puzzle in the following way: (the best form is slightly different)

Rows 546123789
Columns 4656123789

Anyway, the box 5 is box 1 in my canonical form. (maxtext), and columns have to be arranged to get the 2 digits of the row 5 in position 1 and 2.




papy999 wrote:
I use a generation based on {-n+n} with an integrated "one solution and minimal verification"
I discard all puzzles with a single as first move (with my own test)
I keep only the first found puzzles between the different isomorph forms (even in some cases different morph form can have different rating)

I filter with a modified version of Serate that enable to stop Serate if Ed doesn't meet a given condition, or if it is not a diamond or not a pearl, this means I've in my database puzzles with partial or full rating. I use this for the seed choice !




Nothing surprising in that, and it worked pretty well in the last game;


Best regards

champagne
champagne
2017 Supporter
 
Posts: 5650
Joined: 02 August 2007
Location: France Brittany

Re: Patterns Game Strategies

Postby champagne » Mon Apr 04, 2011 4:10 pm

papy999 wrote:champagne,

At the moment for some generation I have more than 95% of generated puzzles that are isomorph of a previous one. This means that the seed was not far enough from previous found puzzles !



No theoretical answer to your point, but this surprises me.

In a game with 21 or 22 given, we come quickly to a situation where most of the puzzles have been found and new generations don't bring many new ones.

With 24 given, I have never seen that. My worst ratio is likely in the range of 60% isomorphic puzzles. Usually, more than 50% are new ones.

champagne
champagne
2017 Supporter
 
Posts: 5650
Joined: 02 August 2007
Location: France Brittany

Re: Patterns Game Strategies

Postby champagne » Wed Nov 23, 2011 10:42 am

Hi,

The game 157 is not yet closed, but on my side, I stopped any work on it 2 days ago.

I worked in that game in 2 directions

1) I tested a new set of programs nearly public and produced all my work out of three programs

Code: Select all
   Gridchecker as puzzles generator
   skfr as puzzles filter
   ACCESS as data base

any new player could easily do the same

2) I gave the priority to the search of high ratings

just to give an idea I got the following results


count per ER

Code: Select all
346   10.3
1036   10.4
823   10.5
341   10.6
14   10.7
1   10.8


count of diamonds per ER

Code: Select all
34   10.3   10.3
113   10.4   10.4
59   10.5   10.5
22   10.6   10.6
2   10.7   10.7



I have no doubt that all diamonds would have finally been found by other players (not 100% sure for the 10.7).
what is amazing is that it took less than one day to get all of them plus the only rating 10.8 I have seen, found first by Mike.
This seems to confirm eleven theory that high ratings come quickly in a search focusing on such ratings.


My infrastructure is very simple. I did not invest much to perform well in the game.

Out of the first lot of seeds, my search has exclusively been a +/- n vicinity exploration.

I did not post during the first day diamonds below 9.6, but I could have done it for a significant number of ER.

Reversely, ronk and mike were faster than me for several diamonds in high ratings and, congratulations to him, Mike was very fast to sort out the highest rating.


I don't know what will be the future of the game, but as soon as the main players will use such tools, the game will go very very fast.

And players not using them will have to find some equivalent tools if they want to keep a chance to win.

champagne
champagne
2017 Supporter
 
Posts: 5650
Joined: 02 August 2007
Location: France Brittany

Re: Patterns Game Strategies

Postby ronk » Wed Nov 23, 2011 11:29 am

champagne wrote:I don't know what will be the future of the game, but as soon as the main players will use such tools, the game will go very very fast.

And players not using them will have to find some equivalent tools if they want to keep a chance to win.

Then, without significant changes in the Patterns Game rules, the long-term future of the game looks grim IMO.
ronk
2012 Supporter
 
Posts: 4764
Joined: 02 November 2005
Location: Southeastern USA

Re: Patterns Game Strategies

Postby m_b_metcalf » Thu Nov 24, 2011 3:18 am

ronk wrote:
champagne wrote:I don't know what will be the future of the game, but as soon as the main players will use such tools, the game will go very very fast.

And players not using them will have to find some equivalent tools if they want to keep a chance to win.

Then, without significant changes in the Patterns Game rules, the long-term future of the game looks grim IMO.

Hear, hear!

Basically, for the me, the game is an interesting way of keeping my hand in at active programming*. Thus, using software from any other source is quite pointless. I've even devised two new filters to keep my use of SE to just the rating of puzzles before submission, avoiding serate as much as possible. If we all use the same programs, the game reduces to determining who happens to be awake when it starts.

Mike Metcalf

* A happy outcome of this was the acceptance of one of my programs for consideration as part of a benchmark. It has passed the initial evaluation and tests and some $$$ have changed hands.
User avatar
m_b_metcalf
2017 Supporter
 
Posts: 8290
Joined: 15 May 2006
Location: Berlin

PreviousNext

Return to Interactive games