Missing the old days :)

[Rebel]

Classic hand tuned engine programming started to fade away the moment autoplayers entered the scene, starting with AUTO232 somewhere in the early 90's. Nowadays it's all about the scientific approach playing 10-20K bullet games and never replay one single game.

While the progress is amazing the romance has gone and the scientific way does change the playing style of a program, F10 vs F11 being an excellent example. Mine as well BTW since I restarted about 2 years ago.

[Dr.Wael Deeb]

What I do miss the more is my own chess playing of 10 years ago. I was capable in those days of playing very long games and now I get tired and scarcely have the stamina to be playing more than two hours in a row.
My playing is not worst, but my enjoyment is lot less.


Fern

2 hours a day is a sign of good stamina actually Fern....

I play at best 2-3 hours when I have the chance regards,
Dr.D

[carldaman]

Maybe Fritz 14, done by the Pandix author, could bring back some of the lost spark -- we'll have to see.

Unlikely. According to the dendogram people, Fritz 14 (ie. Pandix) is very correlated to Robbolito. There is a discussion in the EO forum about it.

That being said, I like the playing style of Robbolito

Yeah, I had been aware of these suspicions for some time, but now we have some actual indications of this. It figures...
This is why I'd held back from buying DF14.

[fern]

Thank you doc.....

[PK]

There is one curious observation that might explain change of playing style of top engines. Some time ago I worked on implementing weaker levels in my engine. I managed to tune it to play at 2000 and 2200 on CCRL scale by crippling search function (details are here: https://github.com/nescitus/rodent_code ... /blunder.c). It seemed perfectly natural to combine this feature with already existing functionality of changing evaluation, to create entities such as "cautious beginner", "hyperagressive club player", "restraint-minded national master" etc. Then it transpired that changing eval function for lower levels caused great variance, amounting to as much as 100 Elo points. Stylish, aggressive eval function that performed suboptimally at the strongest level, substantially improved intended "national master" level! This observation - namely that imperfect search performs better with biased eval, whereas better search thrives on more balanced eval - would probably merit a scientific paper (were it not for the fact that I'm a philologist, not a computer scientist).

[fern]

Many years ago I noted that my Chess Challenger champion comp, played sometimes better if random function non, say, it could choose between two or thre moves more or less equally good instead of going for the very best according his program.
I understood that the very best was such only if we assumed the programming was the very best, which it was not. So in a sense that random option gave the comp a chance to run away of the limits of his programming.
Maybe some similar in your case?


Fern