CTS Opponent's Clock

[roq]

Recently I've changed my approach to solving the problems (I only do Blitz because the ticking clock simulates a real game and creates tension). Instead of entering the first likely move that i see on the board, I now make myself calculate and see a continuation leading to advantage before playing it, however long it takes. Initially my Glicko rating went down a lot, but accuracy increased.

In this respect I much prefer Chess Tempo's system to the CTS one (where you get really heavy penalties for time). However, one thing I really like on CTS is the opponent's move feature whereby they allow you to see the position for 5 seconds or so with the opponent's clock ticking before the move is made. On chess Tempo I often fail problems when there are two or more winning lines of play, such as say a mate in two and winning the queen. This isn't such a problem on CTS since looking at the position on your opponent's time often indicates if there is more than one tactical theme in the position and is closer to what happens in a real game situation.


             

[richard]

Hi Roq,

I think this is a good idea, although I think the rating system also helps take care of this type situation.  Positions with say a mate in 2 versus a material win tend to get relatively higher ratings, so whilst they are harder, they also represent a greater reward if you can see the "correct" move.  There is also an argument that problems should be eliminated where there is a mate competing with a move that produces a winning material advantage but no immediate mate. I'm somewhat sympathetic to this argument, especially where the mate is long and complicated, a human player is probably better off going for the immediate material advantage over the long mate.  One of the first things I plan to do when I have time is to try and remove some of the more obvious examples of long mates versus material wins.

Regards,
Richard.

[roq]

Hi again Richard

Personally, I don't think that the material gain versus mate problem is particularly serious and also to fix it would probably require you eyeballing lots of individual borderline problems and i suspect you have better things to do. Also the ambiguity does encourage me to analyze a position knowing that there may be multiple ways to an advantage, which is not a bad thing. I think a more common and insidious (from the point of view of improving real world chess performance) problem is where the solution moves are trivial and the complexity is hidden in later moves that you are not required to find - e.g. the computer will sacrifice a lot of material to prevent a mate you may not even have seen. These problems increases the efficacy of guessing to get a quick Glicko hit. I tend to mark these with one star and a comment. Personally I want to be challenged and fooled by the problem set (even though I explode with frustration when I miss something!), but I notice that many of the problems I consider good (often ones I've failed) are sometimes given low marks by other users, so it may be hard to filter the problem quality based on user ratings.

BTW This is all somewhat relative. I think the problem set as a whole is good already (many of the problems are excellent *****) and the few suspect problems detract very little from my enjoyment of the site.   




       

[richard]

Hi Roq,

Thanks, its nice to hear you are mostly satisfied with the current problem set quality.  I think its more natural for people to speak up when they see issues, so its a bit tricky for me to assess what the overall level of satisfaction is like. 

The general consensus seems to be that CTS still has the overall edge on problem quality. I can see strengths and weaknesses in both problem sets.  CTS definitely has less of the "only a computer would play that move" type problems. The long mates versus material wins are often of this type.  CTS don't make it clear but I believe all of their tactics sequences were played over the board and they use the computer only to verify ambiguity/soundness issues.  I've taken a different approach where I take positions from real games and then let the computer tell me what the non-ambiguous tactic is (if any). Sometimes that was the tactic played in the actual game, often it is not (sometimes game tactics are flawed [at least from the computers point of view] or lead to longer mates than was necessary).  Whilst CTS's approach leads to almost no "computerish" tactics, my approach can extract interesting tactics from games where the two players miss them (and this does seem to happen fairly often, even at the GM level [although it can be hard to know if the GM 'missed' the tactic or just found a safer, less immediately winning sequence]).

I think one place where my position generator appears to have an edge is in how long I allow the tactic to go on.  Whilst I still have some problems where the tactic finishes before the point is obvious, my (non quantified) feeling is that CTS has a lot more problems where you can guess a first move without having fully grasped the position.  I'd guess that this is partly why CTS tends to have higher % correct than here.

I'd still like to continue to improve the generator.  Ideally with the material gain versus mate I could come up with something automated to filter out some of those.  The main challenge here is that using a chess engine for evaluations which are fed into my own code is a little problematic for these cases.  Chess engines value a mate above any material winning move and often once a clear forced mate opens up there are other longer sequences that will also lead to mate. So the "best N" moves don't often include the "winning but non-mate" material grab.  I'd have to hack directly at the chess engine to modify this behaviour and I'd like to avoid that as it ends up shackling me to a particular engine.  At the moment I can get input from any UCI aware chess engine.  This gives me a large choice of free and commercial engines (I'm currently using Toga 1.3 after playing around with a couple of other free engines).

After I've finished improving the generator I'm going to look at using the problem rating and problem quality data to cull bad moves from the set.  There is a degree of noise in the star ratings, with some comments suggesting alternative moves that have straightforward refutations.  Obviously the higher ranked the comment maker, the more likely their analysis will be accurate (which isn't to say that lower rated player don't come up with valuable comments!). I think combining the user rating with the star rating and also looking at very highly rated problems manually should allow a number of the more annoying problems left in the set to be removed.

 
Anyways, thanks again for the feedback, its very useful in helping me decide how I should prioritize the time I spend on the site.

Regards,
Richard.

[roq]

Hi

I think one place where my position generator appears to have an edge is in how long I allow the tactic to go on.  Whilst I still have some problems where the tactic finishes before the point is obvious, my (non quantified) feeling is that CTS has a lot more problems where you can guess a first move without having fully grasped the position.  I'd guess that this is partly why CTS tends to have higher % correct than here.

The usual answer to tactics problems is to sac a piece somewhere close to your opponent's king. IMHOP your system does generate a wider range of possibilities. Some of the problems are interesting and unlike anything I’ve seen elsewhere – But many give me a feeling of déjà vu as positions that I’ve seen before in games. I hope that any filtering won’t reduce this mix - Some of the problems aren’t pretty and some are a bit messy – but that’s good I think.