How i s Komodo getting on? (now, end of December 2016)

[S.Taylor]

10.3 still not been able to find Keres attack. It looks like grandmaster evaluation doesn't pay off

They sacrificed knowing the Keres Attack for knowing thousands of other opening positions very well. You can't judge an engine on one positioon. There are thousands of of games and hundreds of thousands of positions that Komodo has proven to know very well, so what if it didn't see one opening move. Komodo is a genius of a chess program, and the smartest of all engines in my opinion from a positional and strategic point of view, almost to the point of appearing to have plans. Hardly ever do I see Kpmodo shuffling pieces unless there is an absolute objective draw, and then you can't blame it. So Larry Kaufman has added tons of smarts to Komodo and I'm sure he always monitors how Komodo is playing for ideas of changes. And I know Mark adds great search imiprovements, move ordering, speedups, and eval weights tuning. It's a great team to create such a great engine.

Sincerely,

Tim.

So is komodo better than sf and houdini positionally?

[tpoppins]

Because we can't see how an engine is improving when it is 2 moving targets at the same time.

I'm sure Komodo developers don't need to test their latest engine version against a Stockfish version that's several months behind the latest commit to know how much progress they made; they have statistical tools and can extrapolate.

What I'm interested to know is what a test like that would give you, an end user, besides a warm fuzzy feeling along the lines of "yeah, the latest SF might beat K10.3 ten-to-one (just an arbitrary example) but had K10.3 been released a few months earlier it might have had a good shot at winning last TCEC". ;)

[S.Taylor]

Because we can't see how an engine is improving when it is 2 moving targets at the same time.

I'm sure Komodo developers don't need to test their latest engine version against a Stockfish version that's several months behind the latest commit to know how much progress they made; they have statistical tools and can extrapolate.

What I'm interested to know is what a test like that would give you, an end user, besides a warm fuzzy feeling along the lines of "yeah, the latest SF might beat K10.3 ten-to-one (just an arbitrary example) but had K10.3 been released a few months earlier it might have had a good shot at winning last TCEC".

I want to see what it takes to outwit SF from TCEC 9, i want to see the beauty over the board, not just to hear an extrapolation findings. And if i feel it handles things so well, then i want to use it myself as my best computer setup, for analyzing and for training personally and testing ideas.. I also don't want to be dependant on internet for ever but to have my favorite engine and know it's the best for the next few years. Rybka 4.1 was the best for some years and even after then, it was almost unbeatable or at least uncrushable.

[Guenther]

Because we can't see how an engine is improving when it is 2 moving targets at the same time.

I'm sure Komodo developers don't need to test their latest engine version against a Stockfish version that's several months behind the latest commit to know how much progress they made; they have statistical tools and can extrapolate.

What I'm interested to know is what a test like that would give you, an end user, besides a warm fuzzy feeling along the lines of "yeah, the latest SF might beat K10.3 ten-to-one (just an arbitrary example) but had K10.3 been released a few months earlier it might have had a good shot at winning last TCEC". ;)

I want to see what it takes to outwit SF from TCEC 9, i want to see the beauty over the board, not just to hear an extrapolation findings. And if i feel it handles things so well, then i want to use it myself as my best computer setup, for analyzing and for training personally and testing ideas.. I also don't want to be dependant on internet for ever but to have my favorite engine and know it's the best for the next few years. Rybka 4.1 was the best for some years and even after then, it was almost unbeatable or at least uncrushable.

That makes no sense as there is always progress and you'll never have any 'setup' which does what you want _for years_.
Also Rybka 4.1 wasn't the best for years. just in your memory it appears like that.

[S.Taylor]

Because we can't see how an engine is improving when it is 2 moving targets at the same time.

I'm sure Komodo developers don't need to test their latest engine version against a Stockfish version that's several months behind the latest commit to know how much progress they made; they have statistical tools and can extrapolate.

What I'm interested to know is what a test like that would give you, an end user, besides a warm fuzzy feeling along the lines of "yeah, the latest SF might beat K10.3 ten-to-one (just an arbitrary example) but had K10.3 been released a few months earlier it might have had a good shot at winning last TCEC".

I want to see what it takes to outwit SF from TCEC 9, i want to see the beauty over the board, not just to hear an extrapolation findings. And if i feel it handles things so well, then i want to use it myself as my best computer setup, for analyzing and for training personally and testing ideas.. I also don't want to be dependant on internet for ever but to have my favorite engine and know it's the best for the next few years. Rybka 4.1 was the best for some years and even after then, it was almost unbeatable or at least uncrushable.

That makes no sense as there is always progress and you'll never have any 'setup' which does what you want _for years_.
Also Rybka 4.1 wasn't the best for years. just in your memory it appears like that.

It was for a very long time. And only now it's beginning to crumble.
True, the infamous clones had some success at the top!

But how can i ever know if Komodo 3 is any better than e.g. the Stockfish which played in TCEC 9, if it is not tested vs that version?

Extrapolation does not work ever! I remember that there have been engines which were good for a long time, until a new one came out and everyone said that it was obviously fantastic, because it was obviously a whole lot stronger than the previous version which was at the top for so long (shredder 7.32?) but when it got on to rating lists, it showed a 1 elo point rise, or maybe it was a 1 elo drop? Something like that.

We need to test older versions in order to know if we have improved since then or not. We can't just say "how dare you say that not? we are keeping up with the joneses!"