[computer-go] thoughts on 100,000 cgos games

[Jerome Abela]

David G Doshay wrote:

> I can think of no solid way to benefit from an automated approach
> given only the information that I lost all of these games. I can think
> of ways to do this manually, going through the games one at a time
> and marking mistakes, but that is just too time consuming.

You don't need to mark them manually.

Given that the evaluation of the situation is 100% correct on the first 
and last moves of the game, just extract the moves where the evaluation 
dropped by more than a given threshold. This algorithm will not find all 
the bad moves, but it will find all the bad evaluations: there is never 
a good reason to evaluate that playing a move makes the situation being 
worth +3.00, and then find after the reply that it's only worth -10.00.

You may still don't know what to do with this set of bad evaluations, 
apart for from reviewing them manually. But at least, you don't have to 
review whole games, just individual moves, and you don't have to review 
all the bad moves, just the biggest blunders.


Jerome.