Skip to content

Commit

Permalink
different stat bonus equation.
Browse files Browse the repository at this point in the history
  • Loading branch information
protonspring committed Mar 4, 2019
1 parent 82ff04b commit 199ecc6
Showing 1 changed file with 1 addition and 1 deletion.
2 changes: 1 addition & 1 deletion src/search.cpp
View file
Expand Up @@ -82,7 +82,7 @@ namespace {
// History and stats update bonus, based on depth
int stat_bonus(Depth depth) {
int d = depth / ONE_PLY;
return d > 17 ? 0 : 29 * d * d + 138 * d - 134;
return d > 17 ? 0 : 11550 - 12 * (d - 32) * (d - 32);
}

// Add a small random component to draw evaluations to keep search dynamic
Expand Down

4 comments on commit 199ecc6

@pb00068
Copy link

@pb00068 pb00068 commented on 199ecc6 Mar 5, 2019

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

FYI: from my experience increasing the linear coefficient does very well on STC but gains nothing/or even hurts at longer TC's, so I fear this is passing STC but failing LTC finally.

@protonspring
Copy link
Owner

@protonspring protonspring commented on 199ecc6 Mar 5, 2019 via email

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

@pb00068
Copy link

@pb00068 pb00068 commented on 199ecc6 Mar 5, 2019 •
edited

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

VoyagerOne explained this once with saturation & clumping.
Big linear coefficient does very well on STC because then values in histories grow very fast and we can distinct good moves from bad moves early on. On higher TC's in turn this has the effect that our statistics get saturated: this is when several good moves touch the ceiling. Once several moves return maxvalue, we cannot distinct anymore which one is actually the best one, they are all clumped to the same value.

P.S.: Anyhow I like to see you testing also for elo-gaining patches now. GL with your tests.

@protonspring
Copy link
Owner

@protonspring protonspring commented on 199ecc6 Mar 5, 2019 via email

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Please sign in to comment.