Problem Rating

(Note: Please read the discussion of the proposed problem type (PT) and problem difficulty (PD) tags on the SGF Wishlist page before reading this page.)

In discussing the content type of the SGF problem difficulty tag proposed in the SGF Wishlist, Benjamin Teuber had the idea of using ratings for the difficulty, just like EGF ratings are used for players' grades.

Therefore a problem intended for a 1d would have a rating of 2100; a problem for a 15k would have a rating of 600 and so on. Programs offering a problem solving mode should allow the user to filter by the rating. For example, the user might search for "all tsumego with a rating from 800 to 1100".

How do you obtain a problem's rating? Maybe a good starting point would be to compare a given problem to a set of other problems whose rating we already know. The Graded Go Problems for Beginners series has had some discussion about the difficulty of their problems (initially because the ranks printed in the books were so wildly off). Judging from this book review page and my own experience (which for now, admittedly, only stretches to the end of volume two), the following correspondences are proposed:

• Volume 1 = 30k-20k = rating 100
• Volume 2 = 19k-11k = rating from 200 (19k) to 1000 (11k)
• Volume 3 = 10k-5k = rating from 1100 (10k) to 1600 (5k)
• Volume 4 = 4k-1d = rating from 1700 (4k) to 2100 (1d)

Conveniently, each book is divided into four "levels" of increasing difficulty, which make determining the problem rating even easier. For example, level 1 of volume 2 might correspond to the 19k-17k range, meaning a rating from 200 to 400.

All problems from Volume 1 would have a rating of 100 and although there certainly is a difference between the first problems and the last problems of that book, the EGF rating doesn't distinguish between ranks from 30k-20k.

For reference, the Igo Hatsuyoron problems might be given a rating of 2700, corresponding to 1p.

Benjamin: Yes, I think your proposed ratings are a good guess to start with - later the problem's rating should just depend on who's able to solve it and who not. This means that people using such a database should register using their rating/rank (or the database should be used from a context where people's strengh is known anyways - of course I'm thinking of UGS^^ )

Calvin: I got these books when the old estimates where in place, and I always felt the old rank targets were more accurate, unless the intent is to solve them in 2 seconds. I work on Volume 4 these days in my quest to consistently break 10k because that's what my book says. My time limit is about one minute and I do them in my head while on my exercise bike to save time and simulate the stress of a game. I never understood why they rescaled them. Maybe the new ones are more accurate, but personally I'll never feel like a real 10k until I've mastered Volume 4 and have almost no chance of missing them in games, so I prefer to ignore these new "inflated" breakdowns.

LithiumTwo: http://www.goproblems.com is a place where this problem has been tackled, and anyone with an interest should check it out... I think the main thing we need to do is get away from the idea of ranking problems in kyus/dans/pings, because it's simply unrealistic... if the idea is to make them easy to find for your rank, I'd say that there are loads and loads and loads of different completely inconsistent rank systems out there so it's all a lost cause anyway if you choose to pick that kind of rank system. If problems are simply given a number value for their difficulty rating, then the person using the source (whether it be book, go server, etc) need only try around 5 problems at random to get an idea of where their level fits in to this.

One point to consider is that SGF is not the most efficient format for diagrams. ISHI is a better choice I believe.