The New Sudoku Players' Forum

[Yogi]

.................1..2..3.4........5......42...6..1...7...67......81.....5.9....8.
My Hodoku gives this puzzle a rating in its system as 926 - Hard, but

.342...5.....87....1.........24...3..5....2..6...........3....12.....7..8........
is rated as 972 - Medium difficulty.

Can anyone help elucidate why a larger number is not assigned at least an equal difficulty level?

[P.O.]

hi Yogi,
i guess you know that one puzzle is slightly harder than the other, SudokuCoach gives the first one SE 4.0 HODOKU 1046 and the second one SE 3.4 HODOKU 1016

[denis_berthier]

.................1..2..3.4........5......42...6..1...7...67......81.....5.9....8.
My Hodoku gives this puzzle a rating in its system as 926 - Hard, but
.342...5.....87....1.........24...3..5....2..6...........3....12.....7..8........
is rated as 972 - Medium difficulty.
Can anyone help elucidate why a larger number is not assigned at least an equal difficulty level?

The 2nd puzzle requires Naked and Hidden Pairs. The 1st also requires finned w-wings.

Such questions would better be in the Help section than in General.

[Maxito_Bahiense]

.................1..2..3.4........5......42...6..1...7...67......81.....5.9....8.
My Hodoku gives this puzzle a rating in its system as 926 - Hard, but

.342...5.....87....1.........24...3..5....2..6...........3....12.....7..8........
is rated as 972 - Medium difficulty.

Can anyone help elucidate why a larger number is not assigned at least an equal difficulty level?

The Hodoku rating is based on an cummulative sum of steps. The more difficult the step, the more the technique adds to the sum. So, you may have a smaller rating index for a puzzle that uses fewer but harder steps. Now, for the label Easy/Medium/Hard/Expert, there are two important points to be taken into account: Every technique is assigned a level: for instance, naked singles is usually labelled "easy", while death blossom is normally "unfair".

If you check the summary tab for a puzzle, you'll get how the rating index is built. The collection of techniques used in the solver path is summarised. Here you'll see the techniques used in the solution, tagged by colours according to their respective "hardness" level: white for easy, green for medium, etc. The last item in the summary is the cummulative rating index.

The tag for the puzzle is, in my understanding, based on two different conditions: firstly, there are rating total thresholds for each category: if a puzzle rating steps over the unfair level mark (in my configuration, 1800), it is marked as extreme. Now, the other condition is most difficult technique used: if a puzzle does not step over the threshold level between medium/hard, but the solution uses some hard technique, it is labelled hard. If all techniques used are medium or easy, then the puzzle is marked as medium.

All these aspects are configurable in Hodoku. Edit/Preferences opens the relevant dialog for altering default values. "Level/font" preferences hold thresholds, while "solver" tab holds record of every technique available, allowing for editing of level and solving value.

[Yogi]

Thanx All

The Google answer explained that the number used by Hodoku is a total of the values it assigns to all the moves required to solve a puzzle, while the difficulty Hard or Unfair or Extreme, etc, refers to the most difficult move used in the path.

[Wepwawet]

Whilst I am a great fan of Hodoku, its rating system leaves a lot to be desired, though it will give you a ballpark rating that can prove to be helpful. However, if you rotate the puzzle, despite it still using the same algorithms to solve the puzzle (prior to the rotation), you sometimes get a different rating, this is more evident in the more difficult puzzles.

[denis_berthier]

if you rotate the puzzle, despite it still using the same algorithms to solve the puzzle (prior to the rotation), you sometimes get a different rating, this is more evident in the more difficult puzzles.

There are always many different ways to reach a solution - different resolution paths. Suppose, in some of these paths, a uniqueness rule is applied (e.g. a UR1), but in another path another rule is applied before it, making it no longer available and leading to use more complex rules than in the first path. I haven't specific examples for Hodoku, but the same phenomenon is well documented for the Sudoku Explainer rating (SER).
The problem can be analysed in the general terms of non-confluence of families of resolution rules (see my book [PBCS]). It appears mainly for uniqueness rules and for non-reversible chain rules + in SER due to arbitrary thresholds in different types of chain rules.

As for ratings that don't suffer from such problems, the only ones I know of are those I introduced, which may combine rules such as Subsets, Subset-chains, braids, g-braids...

One more point about the rating of the total path: there has never been any constant definition of such a rating. Hodoku can't be specifically blamed about it.
.

[StrmCkr]

Rating programs gives you a "ball park" of skills needed to solve it, that's about it.

isomorphism affects rating: due to its search functions for a specific Technique it has a list of found by a Fixed finding sequence.

lets say xy wing has 7 found, but only "7" has direct impact: hodoku will apply all 7 if its the last in the list:
that inflates the rating by 6* xy wing rating.

we morph the puzzle and hodokus fixed search engine now instead finds the "7" xy wing First and has 6 others in que. the first is applied and not the others = way lower rating.

this happens for every single technique.

ontop of that messing with its internal rating system hierarchy : like mimic humans: Hidden befor naked and swap the score => lower grid rating on average.

hodokus unreliable as a "rating" when looking at others score of hodoku compared to your copy that both of u may have adjusted.



SE better rating system but still with its own flaws.
fix hierarchy: publicly displayed.
hardest step is the "rating score",
flaws => missing many modern solving methods: scores could be drastically lower: doesn't have A.I.C , uses forcing chains {which puts some A.I.C methods higher by nodal counts = higher rating {remote pair as an example} }

uniqueness assumptive techniques Skip many logic steps: these are state based checks, changing the sequence of moves may remove these as options.
= harder rating if they are no longer applicable.

[champagne]

And I think that for a player, all morphs don't have the same difficulty.

Some patterns are easy to spot in any morph, some are not.

One example is the symmetry of given. If the puzzle is morphed to show it, the player has no difficulty to spot it.
With other morphs, he has a good chance to miss the symmetry.

I am not a skill player, to work on a puzzle with the TH threat, I usually morph the puzzle to have the classical studied pattern.