[Bug-gnubg] 0ply doubles early

[Christian Anthon]

On gammonline "Stick" claimed that 0ply doubled early. That came as a
bit of a surprise to me, so I set gnubg up to play expert/expert.
After around 20000 games gnubg had used more than half of the
computers memory and I cancelled the match at that point. Then I did a
2ply analysis of the cube actions only. The statistics is striking,
Almost a third of 0ply's doubles are considered early and the ratio
(early doubles):(missed double takes) is 10:1. The cube action of 0ply
is still excellent though.

Perhaps all this is just a manifestation of gnubg 0ply overestimating
the equity of the player on roll, but I still found it interesting.

Christian.



Player                          gnubg0ply                   gnubg0ply

Luck statistics

Rolls marked very lucky         479                     466
Rolls marked lucky              1301                    1258
Rolls unmarked                  25483                   25436
Rolls marked unlucky            1054                    1152
Rolls marked very unlucky       167                     176
Luck rate (total)               +68.361 (+48.110)       +28.019 (+91.308)
Luck rate (per move)              +2.4 ( +0.002)          +1.0 ( +0.003)
Luck rating                     None                    None


Cube statistics

Total cube decisions            19141                   20119
Close or actual cube decisions  3132                    3002
Doubles                         966                     898
Takes                           624                     689
Passes                          274                     277
Missed doubles (below CP)        30 (-1.560 ( -2.828))   28 (-1.146 ( -1.543))
Missed doubles (above CP)         4 (-0.127 ( -0.255))    1 (-0.011 ( -0.022))
Wrong doubles (below DP)        292 (-9.209 (-10.496))  260 (-7.768 ( -9.034))
Wrong doubles (above TG)         11 (-0.110 ( -0.303))    6 (-0.025 ( -0.074))
Wrong takes                      24 (-0.895 ( -1.523))   16 (-0.721 ( -2.028))
Wrong passes                     15 (-0.723 ( -1.268))   16 (-0.720 ( -0.962))
Error rate (total)              -12.624 (-16.673)       -10.391 (-13.662)
Error rate (per cube decision)    -4.0 ( -0.005)          -3.5 ( -0.005)
Cube decision rating            World class             World class


Overall statistics

Error rate (total)              -12.624 (-16.673)       -10.391 (-13.662)
Error rate (per decision)         -4.0 ( -0.005)          -3.5 ( -0.005)
Equiv. Snowie error rate          -0.2                    -0.2
Overall rating                  World class             World class
Actual result                   -73.000                 +73.000
Luck adjusted result            -29.802                 +29.802
Advantage (actual) in ppg       -0.053                  +0.053
95% confidence interval (ppg)    0.163                   0.163
Advantage (luck adjusted) in ppg -0.022                  +0.022
95% confidence interval (ppg)    0.027                   0.027

stat
Description: Binary data