Human versus Machine

Discussion of computer chess matches and engine tournaments.

Moderators: hgm, Harvey Williamson, bob

Post Reply
Lyudmil Tsvetkov
Posts: 6031
Joined: Tue Jun 12, 2012 10:41 am

Re: Human versus Machine

Post by Lyudmil Tsvetkov » Thu Nov 30, 2017 6:25 pm

Henk wrote:
Lyudmil Tsvetkov wrote:
Edsel Apostol wrote:Just continue with your work Lyudmil. Never mind the trolls!
Thanks Edsel!

The problem is they think I am the troll. :D


If your book is good then it is not original. For good ideas don't come out of nothing. Although Penicillin was discovered by accident.

And if it is original then it must be bad.


You excel at this. :)

Lyudmil Tsvetkov
Posts: 6031
Joined: Tue Jun 12, 2012 10:41 am

Re: Part III

Post by Lyudmil Tsvetkov » Thu Nov 30, 2017 6:29 pm

Seriously, I am just curious and can not remember any consistent measurements, so would be glad for some feedback.

If engine A draws a match of 20 games against engine B with both engines not pondering, what would be the score, how many elos will engine A that ponders gain in the same match against engine B that does not ponder?
How much dependent this will be on time control?

lkaufman
Posts: 3082
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Human versus Machine

Post by lkaufman » Thu Nov 30, 2017 11:11 pm

Henk wrote:
Lyudmil Tsvetkov wrote:
Edsel Apostol wrote:Just continue with your work Lyudmil. Never mind the trolls!
Thanks Edsel!

The problem is they think I am the troll. :D


If your book is good then it is not original. For good ideas don't come out of nothing. Although Penicillin was discovered by accident.

And if it is original then it must be bad.


It's probably somewhere around 20 elo at intermediate levels (more at blitz/bullet, less under TCEC conditions) which sounds like a significant amount but only implies a 10.5 to 9.5 score! That's in the ballpark of what the compiler bug is costing Komodo in TCEC in terms of speed lost.
Komodo rules!

Lyudmil Tsvetkov
Posts: 6031
Joined: Tue Jun 12, 2012 10:41 am

Re: Human versus Machine

Post by Lyudmil Tsvetkov » Fri Dec 01, 2017 12:55 pm

lkaufman wrote:
Henk wrote:
Lyudmil Tsvetkov wrote:
Edsel Apostol wrote:Just continue with your work Lyudmil. Never mind the trolls!
Thanks Edsel!

The problem is they think I am the troll. :D


If your book is good then it is not original. For good ideas don't come out of nothing. Although Penicillin was discovered by accident.

And if it is original then it must be bad.


It's probably somewhere around 20 elo at intermediate levels (more at blitz/bullet, less under TCEC conditions) which sounds like a significant amount but only implies a 10.5 to 9.5 score! That's in the ballpark of what the compiler bug is costing Komodo in TCEC in terms of speed lost.


That ballpark would make for 5 full points in 100 games, just the difference between Komodo or Houdini winning TCEC. :)

Milos
Posts: 2898
Joined: Wed Nov 25, 2009 12:47 am

Re: Human versus Machine

Post by Milos » Fri Dec 01, 2017 1:35 pm

lkaufman wrote:
Henk wrote:
Lyudmil Tsvetkov wrote:
Edsel Apostol wrote:Just continue with your work Lyudmil. Never mind the trolls!
Thanks Edsel!

The problem is they think I am the troll. :D


If your book is good then it is not original. For good ideas don't come out of nothing. Although Penicillin was discovered by accident.

And if it is original then it must be bad.


It's probably somewhere around 20 elo at intermediate levels (more at blitz/bullet, less under TCEC conditions) which sounds like a significant amount but only implies a 10.5 to 9.5 score! That's in the ballpark of what the compiler bug is costing Komodo in TCEC in terms of speed lost.

You seems to be answering in a wrong thread but your answer is shameless and wrong. First there is not even a remotely reliable figure regarding what is actual slow down of K 1960-1970 vs. 1959 on TCEC hardware. BS-ing about 23% when on 22 cores it is 8% is just ridiculous.
It is quite easy to compare Komodo in early opening phase of the games in stage 2 and now in stage 3. In stage 2 it was typically around 47Mnps and now is 40Mnps. So that is reduction of 15% at best.
Second, on TCEC conditions doubling single core nodes gives around 30-40Elo, doubling it in terms of additional cores hardly 20Elo. Question is how much of the so-called K bug is related to single core performance. The indication is quite strong that it is exclusively SMP speed loss. In that case 15% speed loss would translate into log(0.85)/log(0.5)*20 = 4.7 Elo.
That is like 0.5 points more for K in 100 games match against H.
And Robert Houdard with his 9Elo estimation was more than generous to you guys, but you obviously have no shame...

User avatar
Guenther
Posts: 2220
Joined: Wed Oct 01, 2008 4:33 am
Location: Regensburg, Germany
Contact:

Re: Human versus Machine

Post by Guenther » Fri Dec 01, 2017 2:22 pm

Milos wrote:
lkaufman wrote:
Henk wrote:
Lyudmil Tsvetkov wrote:
Edsel Apostol wrote:Just continue with your work Lyudmil. Never mind the trolls!
Thanks Edsel!

The problem is they think I am the troll. :D


If your book is good then it is not original. For good ideas don't come out of nothing. Although Penicillin was discovered by accident.

And if it is original then it must be bad.


It's probably somewhere around 20 elo at intermediate levels (more at blitz/bullet, less under TCEC conditions) which sounds like a significant amount but only implies a 10.5 to 9.5 score! That's in the ballpark of what the compiler bug is costing Komodo in TCEC in terms of speed lost.

You seems to be answering in a wrong thread but your answer is shameless and wrong. First there is not even a remotely reliable figure regarding what is actual slow down of K 1960-1970 vs. 1959 on TCEC hardware. BS-ing about 23% when on 22 cores it is 8% is just ridiculous.
It is quite easy to compare Komodo in early opening phase of the games in stage 2 and now in stage 3. In stage 2 it was typically around 47Mnps and now is 40Mnps. So that is reduction of 15% at best.
Second, on TCEC conditions doubling single core nodes gives around 30-40Elo, doubling it in terms of additional cores hardly 20Elo. Question is how much of the so-called K bug is related to single core performance. The indication is quite strong that it is exclusively SMP speed loss. In that case 15% speed loss would translate into log(0.85)/log(0.5)*20 = 4.7 Elo.
That is like 0.5 points more for K in 100 games match against H.
And Robert Houdard with his 9Elo estimation was more than generous to you guys, but you obviously have no shame...


Well, you are in the wrong thread and LKs post has nothing to do with what you replied, obviously driven by some agenda.

He simply answered to a question of LT, but the orginal quote somehow slipped away:

If engine A draws a match of 20 games against engine B with both engines not pondering, what would be the score, how many elos will engine A that ponders gain in the same match against engine B that does not ponder?
How much dependent this will be on time control?
Guenther Simon
http://rwbc-chess.de/

Milos
Posts: 2898
Joined: Wed Nov 25, 2009 12:47 am

Re: Human versus Machine

Post by Milos » Fri Dec 01, 2017 2:44 pm

Guenther wrote:Well, you are in the wrong thread and LKs post has nothing to do with what you replied, obviously driven by some agenda.

He simply answered to a question of LT, but the orginal quote somehow slipped away:
If engine A draws a match of 20 games against engine B with both engines not pondering, what would be the score, how many elos will engine A that ponders gain in the same match against engine B that does not ponder?
How much dependent this will be on time control?
His answer has nothing to do with ponder at all. He doesn't mention ponder in it, and to estimate Elo in case of ponder engine against non-ponder one needs to know ponder-hit ratio which depends on the absolute strength of the engines.

On the other hand LT replies to LK's post with:
That ballpark would make for 5 full points in 100 games, just the difference between Komodo or Houdini winning TCEC.
which means it is you who obviously got it wrong.

Lyudmil Tsvetkov
Posts: 6031
Joined: Tue Jun 12, 2012 10:41 am

Re: Human versus Machine

Post by Lyudmil Tsvetkov » Fri Dec 01, 2017 8:33 pm

Milos wrote:
Guenther wrote:Well, you are in the wrong thread and LKs post has nothing to do with what you replied, obviously driven by some agenda.

He simply answered to a question of LT, but the orginal quote somehow slipped away:
If engine A draws a match of 20 games against engine B with both engines not pondering, what would be the score, how many elos will engine A that ponders gain in the same match against engine B that does not ponder?
How much dependent this will be on time control?
His answer has nothing to do with ponder at all. He doesn't mention ponder in it, and to estimate Elo in case of ponder engine against non-ponder one needs to know ponder-hit ratio which depends on the absolute strength of the engines.

On the other hand LT replies to LK's post with:
That ballpark would make for 5 full points in 100 games, just the difference between Komodo or Houdini winning TCEC.
which means it is you who obviously got it wrong.
I guess MS and GS did not understand LK and LT, or vice-versa.

My question was actually very simple: in case I have 5 minutes for the game, and the machine 2 minutes, just as an example, and the machine ponders, how much strength it would gain in comparison to the case, when it will not ponder?
Could this be translated into time increment?

lkaufman
Posts: 3082
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: Human versus Machine

Post by lkaufman » Sat Dec 02, 2017 4:11 am

Lyudmil Tsvetkov wrote:
Milos wrote:
Guenther wrote:Well, you are in the wrong thread and LKs post has nothing to do with what you replied, obviously driven by some agenda.

He simply answered to a question of LT, but the orginal quote somehow slipped away:
If engine A draws a match of 20 games against engine B with both engines not pondering, what would be the score, how many elos will engine A that ponders gain in the same match against engine B that does not ponder?
How much dependent this will be on time control?
His answer has nothing to do with ponder at all. He doesn't mention ponder in it, and to estimate Elo in case of ponder engine against non-ponder one needs to know ponder-hit ratio which depends on the absolute strength of the engines.

On the other hand LT replies to LK's post with:
That ballpark would make for 5 full points in 100 games, just the difference between Komodo or Houdini winning TCEC.
which means it is you who obviously got it wrong.
I guess MS and GS did not understand LK and LT, or vice-versa.

My question was actually very simple: in case I have 5 minutes for the game, and the machine 2 minutes, just as an example, and the machine ponders, how much strength it would gain in comparison to the case, when it will not ponder?
Could this be translated into time increment?
I suppose ponder hits are much lower against a human than against another computer. Maybe in your example 2 min with ponder might be about as strong as 3.5 or so minutes without ponder, which would in turn be about like 2 min plus 3 sec increment without ponder. Note that the greater the time odds, the more ponder adds to elo. With equal time it might be 20 elo or so in fast games, but at say one hour to one minute odds it might be more than a hundred elo. That's why when we played GM Erenburg with 90 to 3 min time odds we also turned off ponder, as with ponder the odds would have been much less.
Komodo rules!

Lyudmil Tsvetkov
Posts: 6031
Joined: Tue Jun 12, 2012 10:41 am

Re: Human versus Machine

Post by Lyudmil Tsvetkov » Sat Dec 02, 2017 3:40 pm

lkaufman wrote:
Lyudmil Tsvetkov wrote:
Milos wrote:
Guenther wrote:Well, you are in the wrong thread and LKs post has nothing to do with what you replied, obviously driven by some agenda.

He simply answered to a question of LT, but the orginal quote somehow slipped away:
If engine A draws a match of 20 games against engine B with both engines not pondering, what would be the score, how many elos will engine A that ponders gain in the same match against engine B that does not ponder?
How much dependent this will be on time control?
His answer has nothing to do with ponder at all. He doesn't mention ponder in it, and to estimate Elo in case of ponder engine against non-ponder one needs to know ponder-hit ratio which depends on the absolute strength of the engines.

On the other hand LT replies to LK's post with:
That ballpark would make for 5 full points in 100 games, just the difference between Komodo or Houdini winning TCEC.
which means it is you who obviously got it wrong.
I guess MS and GS did not understand LK and LT, or vice-versa.

My question was actually very simple: in case I have 5 minutes for the game, and the machine 2 minutes, just as an example, and the machine ponders, how much strength it would gain in comparison to the case, when it will not ponder?
Could this be translated into time increment?
I suppose ponder hits are much lower against a human than against another computer. Maybe in your example 2 min with ponder might be about as strong as 3.5 or so minutes without ponder, which would in turn be about like 2 min plus 3 sec increment without ponder. Note that the greater the time odds, the more ponder adds to elo. With equal time it might be 20 elo or so in fast games, but at say one hour to one minute odds it might be more than a hundred elo. That's why when we played GM Erenburg with 90 to 3 min time odds we also turned off ponder, as with ponder the odds would have been much less.
Without having done any tests, I have pretty much the same intuitive understanding.

Post Reply