[Bug-gnubg] 0-Ply Vs 1-Ply Money Sessions Analysed By Snowie 4

[Michaeldepreli]

I was running some money sessions between 0-ply and 1-ply and then analysing at 2-ply and found that 0-ply and 1-ply checker error was very close. Cube errors were larger however by 1-ply. Wrong doubles & wrong passes due to its tendency to evaluate too high.
David Montgomery pointed out to me that 2-ply may be biased towards 0-ply so I ran through about 500 money games (in batches of 50) with Snowie 4 who shouldn't have any bias and hopefully is considered strong enough to use.

Snowie's opinion was that indeed GNU's 2-ply evals were consistently too biased in favour of 0-ply's checker play and showed 1-ply was actually stronger.
For cube however Snowie 4 confirmed that 0-ply is better.
Small sample but in each batch the same pattern emerged.

These tests assume skill neutral positions. I don't think this would be the case where 0-ply doesn't understand the position well
causing 1-ply to fluctuate wildy.



Michael