Some problems could end sooner than they do. Here’s a few to investigate:
66
361
514
567
619
627
683
1427
2024
2025
Maybe too long at the beginning?
82
Thanks for your help, I’ll add these to my test cases as soon as I get a chance.
My plan is to use the next version of the problem generator rebuild these problems to end sooner, but keep the ratings from the current problems.
But that version isn’t even started yet so it will be a while. Do you think any of these are frustrating enough to warrant being deactivated in the mean time?
Problem #: 1297
Not sure why this one ends where it does.
That one is pretty sloppy, there are a lot of alternates. The engine likes the path it does because with the trades it enters the endgame database and sees the win.
It ends where it does because the heuristic tries to shorten problems when it can. In this case it would have ended after the two forced captures as red’s king is stuck in the corner. But since those two captures are the only moves available in ends early.
I’d like this pruning to be both more aggressive to generate fewer problems that continue too long after the winning combination but also smarter to avoid confusing endings like this one. But it’s not yet.
here’s another I find baffling. I was with it until the last move.
2041
Oh wait, I see it now, but still, it ends too early.
I think this problem works pretty well, you wouldn’t play 14-18 if you didn’t see that you get the material back. The pruning works by checking if you’re up already up material (including kings) and moves are forced. The goal of that pruning was to avoid forcing the player to play out every capture when the defending side runs out of moves. Otherwise it might force the player to play out five captures in a row when they obviously see the problem.
So now in that scenario you run him out of safe squares, capture the first piece, and then the problem ends. I hadn’t come up with a good way of telling the difference. Maybe it should check the intermediate moves and realize you’re sacrificing material (briefly) and let you play it out. It’s an easy thing to try anyway. I worry that will force players to trade repeatedly when up and again extend the problem needlessly.
Thanks for pointing these out, it’s an easy thing to try for version two of the problem generator. I’m not sure it will be an improvement, but it’s worth trying.
1327
I went 3-7, it wanted 3-8, but there is no functional difference where the problem ends.
1327. It looks to me like there is a difference. If you played 1. 3-7, the final move 11-15 would be impossible, giving white a double jump. And because the checker on 7 can’t move forward it gets ganged up on and taken returning to material parity.
There are problems where it really makes no difference and one is an alternate and one is the winning line which is silly and I’ll disable those. And it is possible the engine missed things, so they’re worth pointing out. But here I think it matters, even if only very far into the future. Any ideas on how I could have made it more instructive?
1306
Too many “That was a good move, but the computer thinks there’s better”
Yep, that one needs to be a couple moves shorter. I’m working on a way to do that.
1306 is now two moves shorter. It ends at a more natural place and avoids the frustrating alternates. Thanks for pointing it out.
I’m closing this topic. If and when you find other problems that need pruning, please create a new topic.