It may be an idea to add an extra option to Sukaku explainer to have the technique displayed for ED, EP and ER
Tarek
Revision of SE ratings and resolution rules
Re: Revision of SE ratings and resolution rules
by tarek » Tue Aug 27, 2019 2:37 pm
-
tarek - Posts: 3762
- Joined: 05 January 2006
Re: Revision of SE ratings and resolution rules
by 999_Springs » Tue Aug 27, 2019 4:34 pm
some thoughts
re: creating a new system of rating puzzles
so i thought the original project was to extend SE's rating at the top end to deal with pencilmark sudokus effectively as they can't be solved with nested dynamic fc (+ dynamic fc) which go up to existing se ~12.7 ish, but you're now looking to explore modifications to SE's system of ratings at the low end. can i ask what exactly is it that you're trying to achieve by doing this? i'm absolutely sure that nobody takes SE rating seriously at the low end <6.5 or so except patterns gamers, and that nobody realistically uses SE at the low levels for a difficulty guide on human solving except nick the guy who created it and MAYBE pat. SE's strength is its ability to rate puzzles at the high end effectively, unlike many solvers that fail to solve anything rated 10+. i suggest if you want to modify SE you do one of these two things:
either
1) keep the lower end rating as close to SE as possible so that you get cross compatibility with the existing version of SE, to avoid any confusion
or
2) completely abandon the existing rating system and revamp it to something that actually makes sense. so no "direct" anything, or fancy rarities for UR's with subsets and so on, maybe do it like how hodoku does it.
what you're doing looks like to me that you're doing a half-hearted mixture of both, and having pat and various people point out these random edge cases like the ultra-rare 5.3 that will basically never show up anywhere, unless the person who created it knows what a 5.3 is and how to create one, in which case that defeats the entire point of rating the puzzle in the first place lol. so what is it exactly you're trying to achieve with modifying SE ratings a little bit? and who is going to use this rating system? it can't be used in the patterns game and SE ratings don't seem to be used much outside it, especially at the low end
also can i suggest that if you do this it might be better to work with modifying skfr instead of se? it's 100x-1000x faster at the top end and the only advantage of using se over skfr is consistency over previous versions as there is a standard version for se that everyone uses, so if you change it you lose that advantage. so why not work with skfr instead?
re: overlapping ratings
so i think you took the description of the rating list in the first post from an old copy of what nick put up on his website, which has now been taken down. from my brief stint in the patterns game wiwth limited success the actual ratings for the chains 6.5+ are not the same as those in the first post; overly short chains at the bottom of each rating never exist because of degeneracy, and higher ones in the list can be rare or very rare, making it part of the challenge for patterns gamers to find. here is my revised list from 6.5 up:
x-chains/x-cycles: common 6.5-6.8; rare 6.9
y-cycles: common 6.6-6.8; rare 6.9-7.0
bidirectional cycles: common 7.0-7.2; rare 7.3 and up
forcing chains: common 7.1-7.3; rare 7.4-7.5
aligned triplet: always 7.5
nishio: common 7.6 and 7.8; semi-rare 7.7 (go ask champagne) and 7.9; rare 8.0-8.1
region/cell fc's: common 8.2-8.5 (8.2 only for region); rare 8.6-8.7
dynamic: common 8.8-9.4; rare 8.7 and 9.5(9.6?)
dynamic(+): comon 9.4-10.1; rare 9.3 and 10.2; i think 9.1-9.2 can exist
dynamic(+fc): common 10.2-10.7; rare 10.1 and 10.8(10.9?) ultra rare 9.9-10.0
dynamic(+multiple fc): 10.8-11.5?
dynamic(+dynamic fc): 11.4-11.9; (pencilmark only) 12.0-12.7
i'm sure someone who can actually do the patterns game properly can elaborate on this list.
the 12.7 comes from clicking on "get all hints" on a 11.9 and waiting 15 minutes for the gui to load properly when you click on the last hint that has like hundreds of chains in it. this kind of matches up with the 12.6-ish ratings that i see some of you were getting in the other pencilmark sudoku thread for some of the hardest puzzles
one of the pros of having overlapping ratings is that if you do not, i.e. if you space them out so that each level starts 1.0 above the rest, then you will get a LOT of "holes" in the rating system for ratings that cannot exist, or are exceptionally rare, and that will mess up the rating system by making puzzles at the bottom end of a chain category much harder than those at the top end of the category below, and so on, and the resulting rating won't be anything like SE if that was your intention.
one of the cons of having overlapping ratings is that sometimes a puzzle's hardest move will be LOWER than its se rating. for instance top1465#2:
7.8...3.....2.1...5.........4.....263...8.......1...9..9.6....4....7.5...........
this is SE 9.5, but only because SE goes through all the dynamic chains up to 9.5 (which are useless) before moving on to the next category, finding the 9.3-rated dynamic(+) chain that is actually useful and solves a clue. you can check this by finding the 9.3 dynamic(+) chain earlier on by clicking "get next hint" until you get to it. this is almost certainly unwanted behaviour. another example is the 6.0 from my maximum clues per se rating thread. there are 2 bug3's, one in a block and the other in a row, but i think SE is designed to look for all block-themed bugs before row/column-themed ones so it finds the 6.0 bug3 in a block before the 5.8 bug3 in a row which you can see by clicking "get next hint" again
re: 5.2 uniqueness
SE's uniqueness ratings are currently as follows
UR type 1, 2, 4, or 3 with hidden pair: 4.5
UR3 with naked pair or hidden triplet: +0.1
UR3 with naked triplet or hidden quad: +0.2
UR3 with naked quad: +0.3
UL6: +0.1
UL8: +0.2
UL10: +0.5
the thing where it rates the UR/UL's with hidden subsets 0.1 lower than the naked subsets is supposed to be a bug. i read somewhere that nick commented in the SE code that they were supposed to be rated the same (i.e. hidden subsets are currently rated 0.1 too low) to be fixed later on, which he never did. note also that SE finds the UR/UL with the naked n-set before the hidden n-set if both exist.
UL10's by themselves aren't THAT rare so you do get examples of SE 5.0 uniqueness occasionally, but UL10 type 3's are exceptionally rare
so i posted a how-to guide when i found the SE 5.3 (UL10 with naked quad) here
you can do the same for SE 5.2 (UL10 with naked triplet) by modifying an existing UR/UL with naked triplet with a lot of clues, and i have one of those right up my sleeve in my favourite thread about SE ratings with lots of clues. you can start with my SE 4.8 with 61 clues (UL6 with naked triplet) and do the same approach by extending the UR by 4 cells into empty space. shouldn't take more than 5 minutes hold on i'll whip one up for you brb
edit: yep here you go
1..2..953...51.28..523..14.2.5.38761.617.2538...15642967..2381531.8.5672528671394
SE 5.2
re: direct anything
so firstly direct claiming cannot exist unless you clicked on get next hint/filter hints with same outcome. secondly if you're going to modify SE just get rid of these. nobody ever solves puzzles like this and it's just a by-product of nick being weird when he created this thing. this goes back to what i said at the start of the post, either keep it as close to SE as possible or change it completely
also if you're changing SE you might want to move quads to lower than ur/ul's since it makes zero sense to have quads rated HIGHER than ur's with quads in them like i mean what the hell is up with that
other stuff
here is an old thread about SE's bugs that i wrote at some point when (as you can see) i was obviously frustrated about life when i was 15. now i'm a decade older and still even more frustrated about life, but SE has stayed the same, so the bugs and stuff are still there. some discussion, including the first post which i have reposted, was lost in the big forum crash of 2009 so some posts may be missing, but a lot of points i mention are in that thread, and some more bugs you might want to take a look at. i don't know i haven't reread that thread in a decade
re: creating a new system of rating puzzles
so i thought the original project was to extend SE's rating at the top end to deal with pencilmark sudokus effectively as they can't be solved with nested dynamic fc (+ dynamic fc) which go up to existing se ~12.7 ish, but you're now looking to explore modifications to SE's system of ratings at the low end. can i ask what exactly is it that you're trying to achieve by doing this? i'm absolutely sure that nobody takes SE rating seriously at the low end <6.5 or so except patterns gamers, and that nobody realistically uses SE at the low levels for a difficulty guide on human solving except nick the guy who created it and MAYBE pat. SE's strength is its ability to rate puzzles at the high end effectively, unlike many solvers that fail to solve anything rated 10+. i suggest if you want to modify SE you do one of these two things:
either
1) keep the lower end rating as close to SE as possible so that you get cross compatibility with the existing version of SE, to avoid any confusion
or
2) completely abandon the existing rating system and revamp it to something that actually makes sense. so no "direct" anything, or fancy rarities for UR's with subsets and so on, maybe do it like how hodoku does it.
what you're doing looks like to me that you're doing a half-hearted mixture of both, and having pat and various people point out these random edge cases like the ultra-rare 5.3 that will basically never show up anywhere, unless the person who created it knows what a 5.3 is and how to create one, in which case that defeats the entire point of rating the puzzle in the first place lol. so what is it exactly you're trying to achieve with modifying SE ratings a little bit? and who is going to use this rating system? it can't be used in the patterns game and SE ratings don't seem to be used much outside it, especially at the low end
also can i suggest that if you do this it might be better to work with modifying skfr instead of se? it's 100x-1000x faster at the top end and the only advantage of using se over skfr is consistency over previous versions as there is a standard version for se that everyone uses, so if you change it you lose that advantage. so why not work with skfr instead?
re: overlapping ratings
so i think you took the description of the rating list in the first post from an old copy of what nick put up on his website, which has now been taken down. from my brief stint in the patterns game wiwth limited success the actual ratings for the chains 6.5+ are not the same as those in the first post; overly short chains at the bottom of each rating never exist because of degeneracy, and higher ones in the list can be rare or very rare, making it part of the challenge for patterns gamers to find. here is my revised list from 6.5 up:
x-chains/x-cycles: common 6.5-6.8; rare 6.9
y-cycles: common 6.6-6.8; rare 6.9-7.0
bidirectional cycles: common 7.0-7.2; rare 7.3 and up
forcing chains: common 7.1-7.3; rare 7.4-7.5
aligned triplet: always 7.5
nishio: common 7.6 and 7.8; semi-rare 7.7 (go ask champagne) and 7.9; rare 8.0-8.1
region/cell fc's: common 8.2-8.5 (8.2 only for region); rare 8.6-8.7
dynamic: common 8.8-9.4; rare 8.7 and 9.5(9.6?)
dynamic(+): comon 9.4-10.1; rare 9.3 and 10.2; i think 9.1-9.2 can exist
dynamic(+fc): common 10.2-10.7; rare 10.1 and 10.8(10.9?) ultra rare 9.9-10.0
dynamic(+multiple fc): 10.8-11.5?
dynamic(+dynamic fc): 11.4-11.9; (pencilmark only) 12.0-12.7
i'm sure someone who can actually do the patterns game properly can elaborate on this list.
the 12.7 comes from clicking on "get all hints" on a 11.9 and waiting 15 minutes for the gui to load properly when you click on the last hint that has like hundreds of chains in it. this kind of matches up with the 12.6-ish ratings that i see some of you were getting in the other pencilmark sudoku thread for some of the hardest puzzles
one of the pros of having overlapping ratings is that if you do not, i.e. if you space them out so that each level starts 1.0 above the rest, then you will get a LOT of "holes" in the rating system for ratings that cannot exist, or are exceptionally rare, and that will mess up the rating system by making puzzles at the bottom end of a chain category much harder than those at the top end of the category below, and so on, and the resulting rating won't be anything like SE if that was your intention.
one of the cons of having overlapping ratings is that sometimes a puzzle's hardest move will be LOWER than its se rating. for instance top1465#2:
7.8...3.....2.1...5.........4.....263...8.......1...9..9.6....4....7.5...........
this is SE 9.5, but only because SE goes through all the dynamic chains up to 9.5 (which are useless) before moving on to the next category, finding the 9.3-rated dynamic(+) chain that is actually useful and solves a clue. you can check this by finding the 9.3 dynamic(+) chain earlier on by clicking "get next hint" until you get to it. this is almost certainly unwanted behaviour. another example is the 6.0 from my maximum clues per se rating thread. there are 2 bug3's, one in a block and the other in a row, but i think SE is designed to look for all block-themed bugs before row/column-themed ones so it finds the 6.0 bug3 in a block before the 5.8 bug3 in a row which you can see by clicking "get next hint" again
re: 5.2 uniqueness
SE's uniqueness ratings are currently as follows
UR type 1, 2, 4, or 3 with hidden pair: 4.5
UR3 with naked pair or hidden triplet: +0.1
UR3 with naked triplet or hidden quad: +0.2
UR3 with naked quad: +0.3
UL6: +0.1
UL8: +0.2
UL10: +0.5
the thing where it rates the UR/UL's with hidden subsets 0.1 lower than the naked subsets is supposed to be a bug. i read somewhere that nick commented in the SE code that they were supposed to be rated the same (i.e. hidden subsets are currently rated 0.1 too low) to be fixed later on, which he never did. note also that SE finds the UR/UL with the naked n-set before the hidden n-set if both exist.
tarek wrote:I’m guessing that these 5.0-5.3 unique loop will be rare!
UL10's by themselves aren't THAT rare so you do get examples of SE 5.0 uniqueness occasionally, but UL10 type 3's are exceptionally rare
Pat wrote:
5.2 non-Jellyfish
would be even rarer than 5.1
we may need help from 999_Springs
to construct one
so i posted a how-to guide when i found the SE 5.3 (UL10 with naked quad) here
you can do the same for SE 5.2 (UL10 with naked triplet) by modifying an existing UR/UL with naked triplet with a lot of clues, and i have one of those right up my sleeve in my favourite thread about SE ratings with lots of clues. you can start with my SE 4.8 with 61 clues (UL6 with naked triplet) and do the same approach by extending the UR by 4 cells into empty space. shouldn't take more than 5 minutes hold on i'll whip one up for you brb
edit: yep here you go
1..2..953...51.28..523..14.2.5.38761.617.2538...15642967..2381531.8.5672528671394
SE 5.2
re: direct anything
so firstly direct claiming cannot exist unless you clicked on get next hint/filter hints with same outcome. secondly if you're going to modify SE just get rid of these. nobody ever solves puzzles like this and it's just a by-product of nick being weird when he created this thing. this goes back to what i said at the start of the post, either keep it as close to SE as possible or change it completely
also if you're changing SE you might want to move quads to lower than ur/ul's since it makes zero sense to have quads rated HIGHER than ur's with quads in them like i mean what the hell is up with that
other stuff
here is an old thread about SE's bugs that i wrote at some point when (as you can see) i was obviously frustrated about life when i was 15. now i'm a decade older and still even more frustrated about life, but SE has stayed the same, so the bugs and stuff are still there. some discussion, including the first post which i have reposted, was lost in the big forum crash of 2009 so some posts may be missing, but a lot of points i mention are in that thread, and some more bugs you might want to take a look at. i don't know i haven't reread that thread in a decade
- 999_Springs
- Posts: 591
- Joined: 27 January 2007
- Location: In the toilet, flushing down springs, one by one.
Re: Revision of SE ratings and resolution rules
by coloin » Tue Aug 27, 2019 5:35 pm
indeed food for thought .... and i have deliberately not tried to understand what SE does ....maybe "it aint broke" 
- coloin
- Posts: 2391
- Joined: 05 May 2005
- Location: Devon
Re: Revision of SE ratings and resolution rules
by tarek » Wed Aug 28, 2019 11:48 am
Now that is a REPLY 999_Springs,
Rejigging the ratings to a more manual solver friendly at the lower end of the ratings is not a big task … Most of the hard work has been and is being done at that top end of the ratings with improving efficiency by other collaborators.
Unfortunately it is broke Colin so these minor changes can hopefully make a difference to some
I agree that the job of understanding how SE works is close to madness and therefore I partly agree that it is best to tweak and improve rather than make radical changes in the rating system.
The number of collaborators is small and therefore the task of revamping the whole rating system is already under-resourced (and obviously with no budget).
I'll bump up the Direct stuff so that they are immediately preceding their non direct counterparts which sounds logical. I'll also add to GUI in the analysis window the most difficult technique encountered next to the rating which currently the overall difficulty only (Analyse F9)
Thanks for the 5.2 UL 10 type 3 puzzle. Thanks for the clarification of the current SE ratings. I'll update the current wiki page attached to the repository with that.
I may suggest to my collaborators who are working on serate to add technique name as an option similar to the ER/EP/ED:
so
so an ER/EP/ED ERtN/EPtN/EDtN could be another added dimension to the Patterns game
The changes in the experimental re-rating are now:
tarek
Rejigging the ratings to a more manual solver friendly at the lower end of the ratings is not a big task … Most of the hard work has been and is being done at that top end of the ratings with improving efficiency by other collaborators.
Unfortunately it is broke Colin so these minor changes can hopefully make a difference to some
I agree that the job of understanding how SE works is close to madness and therefore I partly agree that it is best to tweak and improve rather than make radical changes in the rating system.
The number of collaborators is small and therefore the task of revamping the whole rating system is already under-resourced (and obviously with no budget).
I'll bump up the Direct stuff so that they are immediately preceding their non direct counterparts which sounds logical. I'll also add to GUI in the analysis window the most difficult technique encountered next to the rating which currently the overall difficulty only (Analyse F9)
Thanks for the 5.2 UL 10 type 3 puzzle. Thanks for the clarification of the current SE ratings. I'll update the current wiki page attached to the repository with that.
I may suggest to my collaborators who are working on serate to add technique name as an option similar to the ER/EP/ED:
so
- Code: Select all
ERtN = Most difficult technique name
EPtN = Pearl technique name
EDtN = Diamond technique name
so an ER/EP/ED ERtN/EPtN/EDtN could be another added dimension to the Patterns game
The changes in the experimental re-rating are now:
- Code: Select all
* <li>3.0: Naked pair//3.0 ---> 3.1 ---> 3.0
* <li>3.1: Direct Hidden Triplet//2.5 ---> 3.0 ---> 3.1
tarek
-
tarek - Posts: 3762
- Joined: 05 January 2006
Re: Revision of SE ratings and resolution rules
by tarek » Thu Aug 29, 2019 8:55 pm
Patterns Game 354 puzzles rated with the experimental rating above and with added R/P/D options to show the technique name
Hidden Text: Show
-
tarek - Posts: 3762
- Joined: 05 January 2006
Re: Revision of SE ratings and resolution rules
by 1to9only » Thu Aug 29, 2019 9:48 pm
As I really hate scrolling right to see what's there, an idea (improvement) might be:
- shorten names: NS, HS, NP, HP, NT, HT, NQ, HQ, XW, SF, JF, XY, XYZ, etc.
- avoid repetition:
1.2/1.2/1.2 HS
1.7/1.5/1.5 DP/HS
1.7/1.7/1.5 DP/HS <- rating does not exist, but you get the idea!
5.7/5.0/2.6 BUG2/NQ/DP
- shorten names: NS, HS, NP, HP, NT, HT, NQ, HQ, XW, SF, JF, XY, XYZ, etc.
- avoid repetition:
1.2/1.2/1.2 HS
1.7/1.5/1.5 DP/HS
1.7/1.7/1.5 DP/HS <- rating does not exist, but you get the idea!
5.7/5.0/2.6 BUG2/NQ/DP
-
1to9only - Posts: 4176
- Joined: 04 April 2018
Re: Revision of SE ratings and resolution rules
by tarek » Fri Aug 30, 2019 6:24 am
Proposed shorter version then
Patterns Game 354: Show
-
tarek - Posts: 3762
- Joined: 05 January 2006
Re: Revision of SE ratings and resolution rules
by tarek » Fri Aug 30, 2019 4:47 pm
The next mini-Release from Sukaku explainer (1.3.0.1 so still backward compatible with current serate) will have these added options to use with the batch solver
If there are several techniques that can qualify as %R %S and/or %P %T (due to SE overlapping ratings) in the solution path then the technique name encountered 1st will be mentioned.
I'll also participate in the Patterns Game for a game to test it
- Code: Select all
%R Explainer rating (most difficult) Technique name
%P Pearl rating Technique name
%D Diamond rating Technique name
%S Explainer rating Technique name (short)
%T Pearl rating Technique name (short)
%U Diamond rating Technique name (short)
If there are several techniques that can qualify as %R %S and/or %P %T (due to SE overlapping ratings) in the solution path then the technique name encountered 1st will be mentioned.
I'll also participate in the Patterns Game for a game to test it
-
tarek - Posts: 3762
- Joined: 05 January 2006
Re: Revision of SE ratings and resolution rules
by tarek » Mon Sep 02, 2019 7:17 am
Uniqueness techniques are bunched in a group
BUG techniques are bunched in another
Chaining techniques are into several groups too.
SE has to exhaust all techniques in a group before moving on to the next set of resolution rules even if the next set of rules may have a lower rating technique.
From examples and testing: a UL10 3 5.3 puzzle may have an 5.2 Jellyfish that will never be used. That Jellyfish if used prior to the UL10 3 technique it may solve the puzzle and should have had a lower rating of 5.2.
Similarly, I tried to see what happens if I specifically change the rating of turbot fish to let's say 4.3 …. That didn't result in a significant change in resolution rule solving path change because SE has to go through all the techniques that are available before chaining. I can't move turbot fish alone down the technique order without dragging some chaining techniques with it.
In that case unless we change how SE handles techniques in general, it may be worth creating gaps between these groups to remove inter-group overlapping …
This should be straight-forward and would allow 2 things:
1. To add techniques in the future in these gaps
2. To correct technique's inconsistency (as did Iksudoku for unique loops)
BUG techniques are bunched in another
Chaining techniques are into several groups too.
SE has to exhaust all techniques in a group before moving on to the next set of resolution rules even if the next set of rules may have a lower rating technique.
From examples and testing: a UL10 3 5.3 puzzle may have an 5.2 Jellyfish that will never be used. That Jellyfish if used prior to the UL10 3 technique it may solve the puzzle and should have had a lower rating of 5.2.
Similarly, I tried to see what happens if I specifically change the rating of turbot fish to let's say 4.3 …. That didn't result in a significant change in resolution rule solving path change because SE has to go through all the techniques that are available before chaining. I can't move turbot fish alone down the technique order without dragging some chaining techniques with it.
In that case unless we change how SE handles techniques in general, it may be worth creating gaps between these groups to remove inter-group overlapping …
This should be straight-forward and would allow 2 things:
1. To add techniques in the future in these gaps
2. To correct technique's inconsistency (as did Iksudoku for unique loops)
-
tarek - Posts: 3762
- Joined: 05 January 2006
Re: Revision of SE ratings and resolution rules
by dobrichev » Mon Sep 02, 2019 10:03 am
Actually this is one of the several dimensions where techniques are/can/should be grouped. We must take care not to make additional terminology soup.
The techniques are grouped by rating similarity and I presume that the initial goal was the techniques from one group not to intersect the ratings with another group. In general this works at the lower end, but for the complicated techniques/ratings there could be inconsistencies as mentioned in the example. For the high end, where chain lengths are involved in the final rating, the techniques have only the lower rating limit defined and nobody was/can/will investigate the precise higher limit.
Technically fragmenting the existing groups to smaller ones isn't difficult, as well as consolidating the existing groups to more general ones.
Another dimensions for grouping the techniques include the technical implementation similarities, ability (perhaps synthetically determined) of the technique to be nested within another, the precision of the logic that is handled and reported, the number of houses involved, etc., etc.
The techniques are grouped by rating similarity and I presume that the initial goal was the techniques from one group not to intersect the ratings with another group. In general this works at the lower end, but for the complicated techniques/ratings there could be inconsistencies as mentioned in the example. For the high end, where chain lengths are involved in the final rating, the techniques have only the lower rating limit defined and nobody was/can/will investigate the precise higher limit.
Technically fragmenting the existing groups to smaller ones isn't difficult, as well as consolidating the existing groups to more general ones.
Another dimensions for grouping the techniques include the technical implementation similarities, ability (perhaps synthetically determined) of the technique to be nested within another, the precision of the logic that is handled and reported, the number of houses involved, etc., etc.
- dobrichev
- 2016 Supporter
- Posts: 1854
- Joined: 24 May 2010
re: revision of Sudoku Explainer ratings
by Pat » Tue Sep 03, 2019 8:21 am
999_Springs wrote:nobody realistically uses SE at the low levels for a difficulty guide on human solving
except nick the guy who created it
and MAYBE pat
not me. i use gsf's solver to filter e.g. "superior"
it might be better to work with modifying skfr instead of se?
it's 100x-1000x faster at the top end
but no GUI
5.2 non-Jellyfish1..2..953...51.28..523..14.2.5.38761.617.2538...15642967..2381531.8.5672528671394
lovely, thanks!
-
Pat - Posts: 4056
- Joined: 18 July 2005
Re: Revision of SE ratings and resolution rules
by StrmCkr » Tue Sep 03, 2019 11:13 pm
Any plans to add the missing techniques to this?
L23, w, m, s, iW, h wings
Wxyz wing & (barn size 5-9)
Als xz, xy
2rc als
Nrc als
Als chains
Uniqueness ur1.1, hidden ur 1,2
Or would that just fup too much and cause everything to be re rated.
L23, w, m, s, iW, h wings
Wxyz wing & (barn size 5-9)
Als xz, xy
2rc als
Nrc als
Als chains
Uniqueness ur1.1, hidden ur 1,2
Or would that just fup too much and cause everything to be re rated.
Some do, some teach, the rest look it up.
stormdoku
stormdoku
-
StrmCkr - Posts: 1427
- Joined: 05 September 2006
Re: Revision of SE ratings and resolution rules
by tarek » Tue Sep 03, 2019 11:47 pm
After improving the current explainer (Backward compatible) I think the next step would be in the direction of Re-rating & Resolution rule revision/addition. The task is not easy as it would require knowledge of how SE works which is a task by itself.StrmCkr wrote:Any plans to add the missing techniques to this?
.
.
.
Or would that just fup too much and cause everything to be re rated.
-
tarek - Posts: 3762
- Joined: 05 January 2006
Re: Unique loops
by tarek » Tue Sep 03, 2019 11:53 pm
Here is a summary of unique loop techniques ratings from what everybody mentioned and after reviewing the code
This hint rated at 4.7
This is the old code controlling difficulty rating according to loop length
I can see from the comment that there was an intention to spread the rating over 0.3 not over 0.5. It also neglects the UL 12
a proposed change which adds difficulty in a linear fashion based on loop length:
The old Type 3 Hidden code:
with the ratings table changing to this:
And in case you missed it. I posted one of SE's own original puzzles that has this UL14
This is overlapping now with the Bug fixes thread as well. I probably should have continued that thread instead of starting a new one
[EDIT: No mistake. The table was wrong … Corrected]
[EDIT2: added proposed change to UL loop difficulty rating]
[EDIT3: Typo]
[EDIT4: Added Hidden ratings and proposed table]
[EDIT5: Added UL14 example and updated proposed ratings table]
- Code: Select all
UL Type1 Type2 Type3+HP Type3+NP/HT Type3+NT/HQ Type3+NQ Type4
4 4.5 4.5 4.5 4.6 4.7 4.8 4.5
6 4.6 4.6 4.6 4.7 4.8 4.9 4.5
8 4.7 4.7 4.7 4.8 4.9 5.0 4.5
10 5.0 5.0 5.0 5.1 5.2 5.3 5.0
12 5.0 5.0 5.0 5.1 5.2 5.3 5.0
This hint rated at 4.7
- Code: Select all
Patterns game puzzle
1............5............9......7...2..............8....4..........6.....3.........4.......3..6.....3..6.......5............91..............7...2..............8........8..2.............7....34.6.....3..6.....34.6...........91............5......3.56...........91......8...34.6.......5.7...2...........56.........78.1..4.......3.56......4..7...2.....8...34.6...1..........34.6..9.2..56.........78....4....9....56......4..7..12..............8.....5.7.....4.6..9.2..56.....3......1..4....9........91...........4......2..............8.......7....3..........5.........6.........7.........8...3..6...........9..3..6.......5....1...........4......2........2.........3..6.......5....1...........4.......3..6..........8.........9......7..
+----------------+----------------+----------------+
| 1 5 9 | 7 2 8 | 4 6 3 |
| 4 36 36 | 5 9 1 | 7 2 8 |
| 8 2 7 | 346 36 346 | 9 1 5 |
+----------------+----------------+----------------+
| 356 9 18 | 346 57 2 | 56 78 14 |
| 356 47 28 | 346 1 3469 | 256 78 49 |
| 56 47 12 | 8 57 469 | 256 3 149 |
+----------------+----------------+----------------+
| 9 1 4 | 2 8 7 | 3 5 6 |
| 7 8 36 | 9 36 5 | 1 4 2 |
| 2 36 5 | 1 4 36 | 8 9 7 |
+----------------+----------------+----------------+
Unique loop 8 Type 1
This is the old code controlling difficulty rating according to loop length
- Code: Select all
public double getDifficulty() {
double result = 4.5;
if (loop.size() >= 10)
result += 0.3;
if (loop.size() >= 8)
result += 0.2;
else if (loop.size() >= 6)
result += 0.1;
return result; // 5.5 - 5.8
}
I can see from the comment that there was an intention to spread the rating over 0.3 not over 0.5. It also neglects the UL 12
a proposed change which adds difficulty in a linear fashion based on loop length:
- Code: Select all
public double getDifficulty() {
double result = 4.5 + (loop.size() / 2 - 2) * 0.1
return result; // 4.5 for UR(UL4) - 4.9 for UL12
}
The old Type 3 Hidden code:
- Code: Select all
public double getDifficulty() {
double toAdd = (hiddenIndexes.length - 1) * 0.1; // Pair=0.1, Quad=0.3
return super.getDifficulty() + toAdd;
}
- Code: Select all
public double getDifficulty() {
double toAdd = (hiddenIndexes.length) * 0.1; // Pair=0.1, Quad=0.3
return super.getDifficulty() + toAdd;
}
with the ratings table changing to this:
- Code: Select all
UL Type1 Type2 Type3+NP/HP Type3+NT/HT Type3+NQ/HQ Type4
4 4.5 4.5 4.6 4.7 4.8 4.5
6 4.6 4.6 4.7 4.8 4.9 4.6
8 4.7 4.7 4.8 4.9 5.0 4.7
10 4.8 4.8 4.9 5.0 5.1 4.8
12 4.9 4.9 5.0 5.1 5.2 4.9
14 5.0 5.0 5.1 5.2 5.3 5.0
And in case you missed it. I posted one of SE's own original puzzles that has this UL14
- Code: Select all
+----------------+----------------+----------------+
| 28 78 27 | 9 13 13 | 4 5 6 |
| 6 49 49 | 5 2 7 | 13 8 13 |
| 3 1 5 | 4 8 6 | 2 9 7 |
+----------------+----------------+----------------+
| 29 5 29 | 7 6 13 | 8 13 4 |
| 14 3 14 | 8 9 2 | 7 6 5 |
| 7 6 8 | 13 4 5 | 9 2 13 |
+----------------+----------------+----------------+
| 5 2 6 | 13 7 8 | 13 4 9 |
| 148 48 134 | 6 13 9 | 5 7 2 |
| 19 79 1379 | 2 5 4 | 6 13 8 |
+----------------+----------------+----------------+
Unique Loop 14 type 4: Cells E1,F1,F4,D6,D7,G7,G2,I2,I6,H4,H9,C9,C8,E8 on 3, 1
==> C9 nor C8 contains the value 1.
This is overlapping now with the Bug fixes thread as well. I probably should have continued that thread instead of starting a new one
[EDIT: No mistake. The table was wrong … Corrected]
[EDIT2: added proposed change to UL loop difficulty rating]
[EDIT3: Typo]
[EDIT4: Added Hidden ratings and proposed table]
[EDIT5: Added UL14 example and updated proposed ratings table]
-
tarek - Posts: 3762
- Joined: 05 January 2006
Re: Revision of SE ratings and resolution rules
by tarek » Fri Sep 06, 2019 4:33 pm
Has anybody got txt file with all patterns game entries (happy to download from a link as well)
I'm going to treat the Unique Rectangle/Loop type 3 (with Hidden set) current rating as a bug with an easy fix as detailed above. I will still state that sukaku explainer would be backward compatible with the patterns game serate 1.2.1 but would like to see the impact it may have on puzzles submitted to patterns game to see how much of an effect it has on puzzles in the range (4.5-5.0)
The other puzzles would be helpful in the experimental rater as well
Thanks
tarek
I'm going to treat the Unique Rectangle/Loop type 3 (with Hidden set) current rating as a bug with an easy fix as detailed above. I will still state that sukaku explainer would be backward compatible with the patterns game serate 1.2.1 but would like to see the impact it may have on puzzles submitted to patterns game to see how much of an effect it has on puzzles in the range (4.5-5.0)
The other puzzles would be helpful in the experimental rater as well
Thanks
tarek
-
tarek - Posts: 3762
- Joined: 05 January 2006
