Oct 13, 2011

[Info] Not so futile attempts

Title refers to an earlier post where I tried to add futility pruning in the past and never really got it to work.

This time though:

Rank Name              Elo    +    - games score draws
1 Mediocre 1.0 -2 31 29 28 100 60% 31%
2 Mediocre 1.0 -1 -31 28 29 100 41% 31%

So in worst case 2 elo better, in best case 60 (the -2 refers to futility pruning added to that version). But somewhere around 30 elo is probably accurate.

Combining this with the test last night would give this:
Rank Name              Elo    +    - games score draws
1 Mediocre 1.0 -2 53 39 38 100 60% 31%
2 Mediocre 1.0 -1 -11 12 12 1100 53% 24%
3 Mediocre v0.34 -42 13 13 1000 46% 24%

Probably flawed to bits with no statistical significance. But I wonder if I can't say that Mediocre has gained somewhere upwards of almost 100 elo.

Probably not, but considerably better it is.

On to internal iterative deepening.

1 comment:

Thomas Petzke said...

Hi Jonatan, good to see you progress. Don't worry much about the absolute ELO gain. Running matches of A vs A' is usually good to prove whether one version is stronger than the other. But in my experience the ELO difference from that is over estimated. You need matches against a set of other opponents for that. It seems you will have a strong engine when Chess War starts division E. Keep it up !

Thomas...