Ultra decisive opening positions

[Dann Corbit]

See:
http://rybkaforum.net/cgi-bin/rybkaforu ... pid=574217
You can't make a big post or attach things here.

The objective of this test set is to provide a list of opening positions for which the deep computer evaluation is fairly small (in this case between 2 and 55 centipawns in absolute value) and for which there is a large ratio for wins by both sides compared to wins + draws.

In other words, lots of wins by white and black, and very few draws for these positions.

The notion is to find a test set where a big computer battle can take place between high-end programs and you do not end up with a huge collection of draws with one or two wins in it.

[Rodolfo Leoni]

See:
http://rybkaforum.net/cgi-bin/rybkaforu ... pid=574217
You can't make a big post or attach things here.

The objective of this test set is to provide a list of opening positions for which the deep computer evaluation is fairly small (in this case between 2 and 55 centipawns in absolute value) and for which there is a large ratio for wins by both sides compared to wins + draws.

In other words, lots of wins by white and black, and very few draws for these positions.

The notion is to find a test set where a big computer battle can take place between high-end programs and you do not end up with a huge collection of draws with one or two wins in it.

Ahh.. An "Age of Ultra"!

Interesting.

[zullil]

for which there is a large ratio for wins by both sides compared to wins + draws.

In other words, lots of wins by white and black, and very few draws for these positions.

Large ratio of decisive games determined how? I mean, who played the games and under what conditions?

[Dann Corbit]

for which there is a large ratio for wins by both sides compared to wins + draws.

In other words, lots of wins by white and black, and very few draws for these positions.

Large ratio of decisive games determined how? I mean, who played the games and under what conditions?

The collection includes:
Super GMs OTB (not blitz) {2700 Elo and up}
Correspondence masters {top 5%}
Top ten programs in CCRL and CEGT
TCEC top ten programs

A few others like those. My picks, I guess you would call it.

I don't put any counts into the database until the position has occurred at least 5 times.

[Lyudmil Tsvetkov]

See:
http://rybkaforum.net/cgi-bin/rybkaforu ... pid=574217
You can't make a big post or attach things here.

what do you mean by ultra-decisive positions?

no matter how hard I try, this is simply beyond my grasping power.

how many positions are included in the set?

the set evaluated as a whole, or just the single positions?

because, for me, both don't make much sense.

evaluating a single position and getting many draws both fro white and black would simply mean the engines and humans playing it are crazy, position is either better or worse for one side, you know, and that is usually reflected in the statistics.

evaluating a larger set of positions instead also does not make much sense to me, as in this case you don't know which positions favour white and which black.

if you were so kind to post here some examples for browsing, maybe you could get a lot more and more meaningful feedback.

[Gusev]

May we have a PGN version?

This is the correct link to the post:

http://rybkaforum.net/cgi-bin/rybkaforu ... ?tid=32232

Ah, maybe that will help

[Michel]

If you take a database of games then a certain fraction of the openings will appear to be "ultra-decisive" by your criterion, just by chance.

The opening positions you selected as "ultra-decisive" should be independently tested to validate the claim that they are indeed ultra-decisive.

The concept of an "ultra-decisive" opening position (if such a thing exists) is interesting but it should be investigated in a statistically correct manner.

[Dann Corbit]

If you take a database of games then a certain fraction of the openings will appear to be "ultra-decisive" by your criterion, just by chance.

The opening positions you selected as "ultra-decisive" should be independently tested to validate the claim that they are indeed ultra-decisive.

The concept of an "ultra-decisive" opening position (if such a thing exists) is interesting but it should be investigated in a statistically correct manner.

I agree totally. I guess that half of these or more will not be ultra decisive when top engines play them at decent time control.

Let's call them "Ultra Decisive Candidates"

[Dann Corbit]

See:
http://rybkaforum.net/cgi-bin/rybkaforu ... pid=574217
You can't make a big post or attach things here.

what do you mean by ultra-decisive positions?

In each case, we have at least 500 games played by high level opponents.

In actual games played, (white_wins + black_wins) / (white_wins + black_wins + draws) was greater than 0.85.

So in practice there was a very surprisingly tiny number of draws compared to wins for one side or the other.

no matter how hard I try, this is simply beyond my grasping power.

how many positions are included in the set?

170

the set evaluated as a whole, or just the single positions?

As M. Vandenberg pointed out, they are really just candidates for a set of positions designed to reduce draws in a high level contest. In practice, they may not work out so well. But I would say that they have much higher chances than positions chosen at random.

because, for me, both don't make much sense.

evaluating a single position and getting many draws both fro white and black would simply mean the engines and humans playing it are crazy, position is either better or worse for one side, you know, and that is usually reflected in the statistics.

Really strong and equally matched computer programs tend to get a lot of draws at long time control. The same is also true for super GM level matches when the opponents are extremely strong (e.g. the WC finals).

evaluating a larger set of positions instead also does not make much sense to me, as in this case you don't know which positions favour white and which black.

if you were so kind to post here some examples for browsing, maybe you could get a lot more and more meaningful feedback.

For this position:
[d]r1bqkb1r/5ppp/pn1ppn2/1p6/3NP1P1/2N1BP2/PPPQ3P/2KR1B1R b kq - acd 40; acs 2218; bm Nfd7; cce -1; ce -7; pm Bb7 {974} Nfd7 {611} Qc7 {16} b4 {4} Nc4 {1}; pv Nfd7 a3; white_wins 733; black_wins 729; draws 137;
there were 1,599 games completed. Of these games, there were 733 white wins, 729 black wins, and 137 draws.

[Dann Corbit]

May we have a PGN version?

This is the correct link to the post:

http://rybkaforum.net/cgi-bin/rybkaforu ... ?tid=32232

Ah, maybe that will help

There is now a PGN file attached here:
http://rybkaforum.net/cgi-bin/rybkaforu ... pid=574217