Evaluation of moves from World Championship Games

[MikeGL]

By today's standards at move 12, we are still in the middle of the opening book. I suspect the results are biased. Try the same but start for example at move 20 to see if there is a difference between the results. In their matches, Kasparov & Karpov played the same opening over and over until beaten to death.

note there are 2 versions of kasparov posted in graph. 1987 and 2000. The 1987 Kasparov is way stronger than year 2000.

[jmartus]

Yeah restart the whole project and do from move 20 seams like opening is causing to much bias

[Ferdy]

By today's standards at move 12, we are still in the middle of the opening book. I suspect the results are biased. Try the same but start for example at move 20 to see if there is a difference between the results. In their matches, Kasparov & Karpov played the same opening over and over until beaten to death.

What do you mean exactly by "the results are biased"?

I added an option at what move number to start the evaluation.

Code: Select all

--evaluation-start-move <move number>

Code: Select all

python stats.py --pgn-folder ./docs/pgn --analysis-file ./docs/human_eval.csv --evaluation-start-move 20 >stats.txt

This is the plot.



I think I have to add an option for opening, middle and ending phases based on material remaining on the board. So that these players can be compared on those phases. One can be good at opening preparation but could be inferior in the middle for example. But of course the analysis quality can be improved to something like 60s/move if we are getting serious with this . There is evaluate.py module to analyze the games and save the positions then use stats.py to generate the plot and other stats.

[Karlo Bala]

By today's standards at move 12, we are still in the middle of the opening book. I suspect the results are biased. Try the same but start for example at move 20 to see if there is a difference between the results. In their matches, Kasparov & Karpov played the same opening over and over until beaten to death.

What do you mean exactly by "the results are biased"?

I added an option at what move number to start the evaluation.

Code: Select all

--evaluation-start-move <move number>

Code: Select all

python stats.py --pgn-folder ./docs/pgn --analysis-file ./docs/human_eval.csv --evaluation-start-move 20 >stats.txt

This is the plot.



I think I have to add an option for opening, middle and ending phases based on material remaining on the board. So that these players can be compared on those phases. One can be good at opening preparation but could be inferior in the middle for example. But of course the analysis quality can be improved to something like 60s/move if we are getting serious with this . There is evaluate.py module to analyze the games and save the positions then use stats.py to generate the plot and other stats.

Interesting, thanks.

Now I'm curious about the extreme case.
Can you restart the test from move 1, please?

[Ferdy]

Interesting, thanks.

Now I'm curious about the extreme case.
Can you restart the test from move 1, please?

Most analysis in human_eval.csv are from move 8, will fill it up later with move number from 1 to 7.