Chesstempo VS chess.emrald

[richard]

chipschap: Interesting to hear about your tournament gains. 

Most of the research done in this area tends to suggests tactics (at least a the higher ratings levels) is more about pattern recognition than calculation (although I expect both skills are important).  I'm surprised you didn't see much benefit from the 10K problems on CTS.  One of the concerns I have with CTS's extreme emphasis on time is that it leads some users to ignore time and concentrate on accuracy.  This means that their rating suffers and therefore they get easier problems.  I think there is an argument that if you are achieving a >95% accuracy level then you might be seeing a lot of positions that are already very easy for you, or in other words they are patterns you have already learned to recognize, and harder positions might be more beneficial in achieving OTB improvement at that point.

Just a theory and I imagine alot of the CTS "go for accuracy" crowd would disagree. I do agree that accuracy is very important - but on CTS where high accuracy usually translates to getting easier problems I think it might be problematic. Standard (and the more relaxed blitz timings here) should allow users to be concerned about accuracy but also get problems that increase in difficulty as their accuracy improves.  I'm sure the CTS style of doing 1000's of problems quickly can be useful, but maybe the "easy problem" issue is eroding the potential OTB gains.

Regards,
Richard.

[pollock]

If you actually pay attention to the time constraints on CTS it improves 3 0 blitz skill immensely . But if want general chess improvement I think working on accuracy with a reasonably high rating in tandem is the way. I accidently got good at fast blitz by using CTS  - peace

[tama]

Now that i have bin only solving problems in standard i fined myself having trouble making fast moves in a online game. I keep looking for some crazy line or i get to nerves to make a move with out calculating in advance.

I think i need to play at least 15 to 20 min games. But everyone that is high rated online only play 5 min games and allot of them play 1 min games witch is a complete waist of time.

[chipschap]

One of the concerns I have with CTS's extreme emphasis on time is that it leads some users to ignore time and concentrate on accuracy.  This means that their rating suffers and therefore they get easier problems.

This is exactly my experience.  I had thought to play for accuracy on CTS and was running over 87% correct.  My rating stayed low and I did not translate benefits to tournament games to any great degree.  That was for the first 9,000 problems or so.  Then I read somewhere about pattern recognition being the point of the excessively fast time control at CTS.  I started to play much faster.  My accuracy slipped to under 84% correct (a big fall after having done 9,000 problems at 87%) over the next 1,000 problems, but my rating went way up.

But it is the standard mode here at CT that is training my ability to select candidate move and to calculate them.  While I can't deny the importance of pattern recognition, the ability to think through a position is vital in tournament play --- and my recent results bear this out as my USCF rating is going up and I am winning a lot more than in the past.

[slacker00]

I just passed 3k problems attempted on CTS.  That's quite a lot!  I'm batting  around 82%, which I think is in the right range.  I try to solve with my kneejerk reaction to get the time bonuses when possible.  I've  noticed sometimes I  do some calculation, sometimes it's just pattern recognition. 

Sometimes I've seen the problem before!  That's my motivation for trying to improve my rating.  I don't really want to see the same problems again and again.  That starts to happen when you get stuck on a certain rating.  The problems start to repeat.  Hopefully, I actually learn from the problems so  the next time I see a specific problem I can solve it in 3 seconds and boost up  my rating to get up to the next level of problems.

[tama]

Does enyone notice that the higher your rating is the less problems people play.
Ex: 3 players on the best list are almost inactive and many others have solved 3 or 4 in the past month since the algorithim came out.

I was at 1600 standard when the algorithim came out n now im at 1900+ so all of use on top worming up my seat u better watch out lol.

[stormcrow]

I have 5000+ problems on CTS.  My rating is low, something like 1430, and at 75% my accuracy isn't anything to write home about.

CTS's time controls are just brutal for me.  I don't think fast.   It's common for me to stay within a 10 point rating range over 200 problems.  Sometimes, if I miss a relatively easy problem, it can take hours to get the points back.  I can't really say my skills are improving.

I am not 100% sure, but I believe that CTS's problems are computer generated.  The problems here are generated, but from real games.  I think that's one reason CTempo is better.

[richard]

Hi Stormcrow,

In CTS's defense I think their problems do come from real games. It is hard to tell, but I think that largely their problems are actually less generated than Chess Tempo's in that we both start with real positions but Chess Tempo chooses the best line suggested by the computer whereas CTS verifies that the line chosen by a human was sound. It seems that CTS *may* auto generate novel starting positions from the positions in real games but these variations may just come from games with very similar positions. The CTS approach (if I've guessed what they are doing correctly) has the advantage of the opponent's counter moves always being relatively obvious, sometimes the opponent moves on here can look a little strange as the computer is making decisions based on evaluations of positions 14+ moves into the future.

CT does have a few areas where it's generator has the edge (IMO).  For example CT removes (or makes allowed alternatives) for positions where the second best is deemed winning, where winning in the current generator is > 2 pawn advantage, although the current set has cutoffs of 2.75 (the new generator with a +2 cutoff will be used to update the current set in the next few weeks).  CTS appears to have no cutoff threshold. I've seen positions where moves that win up to a Rook are marked wrong where the best move may win a queen for example.  The overall perception seems to be that these are not that common in the CTS set but I recently looked at the top 20 rated problems and 15% of the CTS problems have winning alternatives >2 pawns.  It is likely that these problems are over represented in the high rated problems (they are on here, and looking at the 20 problems from the 10,000th ranked problem on CTS only showed 1 problem (5%) with an alternative line that would have been marked wrong [note choosing samples of 20 problems is way to low to make conclusions so don't take these results too seriously]).

CT also tends (again IMO) to do a better job of forcing users to prove they understand the point of the tactic.  CT still has some problems that end at positions that make it hard to see the tactical point , however it appears to have a lot less of these than CTS, this means that CTS can more often reward users for having a good guess at the first move without understanding the position.

Anyway CTS and CT both have different things to offer, those that prefer ultra fast tactics can choose CTS and those that like to take longer (either untimed or slightly slower than CTS) can choose CT (lots of users seem to use both).


Regards,
Richard.

[tomohawk]

I've noticed that the CTS has a number of positions based on variations not played in famous games. I saw a variation on Kasparov v Lautier, Linares 1994 just recently. There are also quite a few problems devoted to unplayed variations in the amazing Kasparov v Topalov, Wijk aan Zee 1999.

[richard]

tomohawk: Thanks for that observation. Sounds like the CT and CTS approaches might not be as different as I thought. I guess it is possible that they are pulling out the alternative lines from annotations in the game's PGN file rather than having purely computer generated lines, this would account for why I don't hear anyone saying CTS lines can look computerish, what is your opinion here?

Regards,
Richard.

[tomohawk]

That's a tough question to answer. Most of the problems that I have done at CTS looked like real positions, at least to me. I would say maybe 15%-25% appear to be composed to illustrate some point (to demonstrate a smothered mate, for example). Generally the winning moves look normal and the range of defenses are not as surprising as they are here.

For more advanced solvers, it is probably a good idea to have many computerish defensive resources. After all, more people are using computers to train with, so their play is likely to be more computerish (in the best sense of the term) in that they will be more likely to look for hidden resources in positions for both themselves and their opponents.

[richard]

I hadn't thought of the line that computerish defenses are good due to the amount of computer training players are doing these days. I'll keep that one up my sleave next time someone complains :-)

I think the worst thing about the computerish responses is that often it can mean the user doesn't have to prove the tactical point as the computer may be happy to drop instant material on the assumption that the human player has already seen what it sees therefore drop a little material now to avoid dropping more later. Obviously humans might try a different defense forcing the player to prove they really can make the tactic work (but maybe leading to a more lost position).

Regards,
Richard.

[tomohawk]

Agreed with that absolutely. Players should have to prove the point of the combination. The "best defense" is the one that a person is least likely to have considered and the one they will have the most difficulty facing in a real game situation. Sometimes that is some bizarre computerish defense that hopefully expands the players' idea of what is possible at the board (I am constantly amazed at the defensive resources in what look to me to be hopeless positions). Sometimes that is an objectively sub-optimal defense but one that poses tough problems to prove its incorrectness. I am not a computer person, but that seems like it would require a lot of work to figure out for each problem.

[richard]

Hi Tomohawk,

Unfortunately you are right, automatically detecting when  sub-optimal (from the computers point of view) responses would make the problem a better one is a fairly difficult problem and one for which I don't currently have any viable ideas (little own solutions).  Probably the best that can be done here is to allow users to enter suggested alternatives (in a much more structured way than just comments so that I can automatically process them) and use some kind of voting process to amend solutions.  Unfortunately this requires a number of UI improvements which I will not be able to get to for a while.

Regards,
Richard.

[boybawang]

Hi richard!

Is it a problem such as the computer sacrifices all materials just to delay mate?
I have a simple solution to make the computer reply like human.

SOLUTION:
   Consider this ITERATIVE DEEPENING process inside a chess engine.
   Internally, the engine first analyzes the position using a single depth(PLY).. Then analyzes the position again at two plies deep.. Then three plies and so on. 
    below is an example where it's black(computer) to move:

    PLY   SCORE                MOVE
     1      0.50                   ..Kh8             // no mate seen yet
     2      0.51                   ..Kh8             // no mate seen yet
     3      0.48                   ..Kh8             // no mate seen yet
     4      0.21                   ..Kh7             // no mate seen yet
     5      +999.99 Mate5     ..Qg5            // Mated! offer queen to delay mate

     The above shows that the computer discovers at PLY 5 that it's about to get mated. therefore, it sacrifices it's queen by ..Qg5 in-order to delay it.
      The simple solution is to use the result from the previous PLY before it discovered the mate. PLY 4 in this case.
      That way the computer will move the more human (..Kh7) instead of computerish Queen suicide (..Qg5).
      This also guarantees that ..Kh7 is not a dumb move because it's just one PLY shallower than the mate discovery.

SUMMARY:
     Use the move from previous depth whenever the evaluation shows sudden loosing score.


I hope it helps..