Puzzles - request for comments
complete
A
Alfie Kirkpatrick [Developer]
I need to find some new puzzle positions soon and this gives an opportunity to improve the puzzles and how it works. Some ideas I already have:
- Exclude short matches. I will try with 7 points as minimum but could reduce to 5 if not enough positions are generated
- Exclude positions where there are two close options at the top so players less likely to lose a streak. Maybe 0.100 minimum gap? Alternative suggestion (by Aaron T) is to not break a player's streak if they are within 0.015 of the top play, which could work well.
- Introduce puzzle "series". Snapshot the leaderboard and start afresh. Mainly to get Ian off the top spot!
- Calendar to navigate past puzzles
- Show the overall match PR of the player who made the blunder, and the final score / who won the match
Other ideas welcome!
B
Backgammon Hub Admin
complete
A
Alfie Kirkpatrick [Developer]
New puzzles are live 🎉
S
Stephen Bailey 🇬🇧 (uptoapoint)
Alfie Kirkpatrick [Developer]Well done Alfie! A problem with the winning streaks is that not infrequently the answer given is wrong, sometimes by quite a bit. This is where Alex's suggestion for using rollouts would be good. But I realise that would probably be too much work for you and not realistic.
D
Denis-George Constantin (playhunter)
I am not much into puzzles, but yes it make sense to not break player's correct choice streak for small differences between top play and next good plays (sometimes a deeper analysis might put the top play in 2nd or 3rd best play). Anything within 0.020 should be considered a correct play.
And yes, I would like to see the average PR of the person who made the blunder or error. If possible, would be great if we can pick to see puzzles only from players with a certain PR level - I'd set it for 7 or lower (which I consider a competitive PR), but in the future I may set it to 4 or lower to study more positions where even GMs are making mistakes too.
I too think you should not exclude short matches because a blunder made at 1-0 in a 3 point match is equal to the same blunder made at 11-10 in a 13 point match. So I'd say do not exclude short matches BUT maybe try to limit the search only to players with a certain maximum average PR. So maybe players with average PR above 10 should not be included for generating puzzles ?
On the other hand, I think seeing a lot of 0.02x-0.03x errors in puzzles, while possibly interesting, might not be very useful for the players. Seeing blungers (0.080 or more difference between the best play) would be much more useful. IF I would be put to choose between the two options, I would choose seeing only blunders (so at least 0.08 difference between best play). [And preferably coming only from players with their average PRs under 7 (for now)].
A
Alfie Kirkpatrick [Developer]
Denis-George Constantin (playhunter), Alex Zamanian (z)
Denis-George, just saw your comments thanks.
Have been working on the next load of puzzles today. Hub has a lot of matches now, so I was able to be quite selective...
It only considers matches of at least 7 points in length and where the player PR (in the match overall) was 4 or better.
I take the point from both of you that shorter matches are also important, but it would be a lot of work (both code and processing time) to identify players with good average PR across lots of matches, especially as some players are playing short matches almost exclusively. It's just a bit too hard given my time available, and I don't think it detracts from the puzzles too much to focus on longer matches.
With the PR cutoff at 4, I hope we get some really interesting positions where truly strong players have made blunders.
I am adding the feature to leave a streak going when less than 0.020 error.
I have generated 465 new puzzles, roughly equal between checker and cube blunders.
A
Alex Zamanian (z)
Alfie Kirkpatrick
I don't understand why you would exclude short matches. In fact that's where a lot of the tough, interesting, instructive stuff is because so many concepts get skewed near the end of the match.
I agree with Aaron that excluding positions where the right option is close to the second best play is bad since it can give a hint and would limit the types of positions under discussion too much. I also like his threshold idea, though would set it to 0.02 (which is XG's definition of error).
Your other ideas seem fine to me, but not especially important.
The most important thing in my opinion would be to use a rollout to determine the best play. I think you do have to rollout lots of plays since the error size matters and comparing a rollout to an evaluation introduces more variance.
Also, the widget used for the discussion of the problem is a bit clunky, would prefer to see something closer to a thread on a discussion board where people can reply to others, navigate the comments easier, edit their comments, etc.
A
Alfie Kirkpatrick [Developer]
Alex Zamanian (z) thanks for input.
The reason for using longer matches is that I pick positions where the player's PR is low for the match, otherwise we end up picking really silly blunders from weak players. And for shorter matches the PR is generally lower. If there is some kind of factor that can be applied to scale the PR according to match length we could try that, but simpler just to use longer matches. What do you think?
A
Alex Zamanian (z)
Alfie Kirkpatrick [Developer] I see. Would it make sense to look for low/medium player PRs instead of looking at the match PRs?
Also, this might be too much work, but I thought it might be cool for a players to be able to submit one of their positions for the puzzle of the day. I know I often make a blunder and wonder if others would tend to make the same blunder.