Skip to content

Commit

Permalink
tune middle part of array. bench 5316057
Browse files Browse the repository at this point in the history
  • Loading branch information
xoto10 committed Jan 11, 2018
1 parent d3e6069 commit f24ec9e
Showing 1 changed file with 12 additions and 12 deletions.
24 changes: 12 additions & 12 deletions src/timeman.h
View file
Expand Up @@ -40,20 +40,20 @@ class TimeManagement {
{
{ // White
//Eval: <-30 -20 -10 <0 >=0 >10 >20 >30
{-21,-25,-28, -4, 0, 0, 0, 0}, // Diff < -20
{ -4, -4, 2, 3, 0, 0, 0, 0}, // -20 <= Diff < -10
{-20,-24,-28, -5, 1, 1, 0, -2}, // Diff < -20
{ -5, -3, 0, 0, 0, -2, 2, 1}, // -20 <= Diff < -10
{ 0, 0, 0, 0, 0, 0, 0, 0}, // -10 <= Diff < 0
{ 0, 0, 0, 0, 0, 0, 0, 0}, // 0 <= Diff < 10
{ 0, 0, 0, 0, -1, -1, 0, 5}, // 10 <= Diff < 20
{ 0, 0, 0, 0, 0, 0, 7, 11} // 20 <= Diff
{ 1, 3, 0, -2, 0, 0, 0, 5}, // 10 <= Diff < 20
{ 0, 2, 0, -2, 0, 0, 7, 12} // 20 <= Diff
},
{ // Black
{-27,-25, -6, 2, 0, 0, 0, 0}, // Diff < -20
{ -6, -1, 2, 0, 0, 0, 0, 0}, // -20 <= Diff < -10
{-27,-24, -6, 0, 1, 0, 3, -2}, // Diff < -20
{ -7, -1, 0, 0, -2, 0, 0, -4}, // -20 <= Diff < -10
{ 0, 0, 0, 0, 0, 0, 0, 0}, // -10 <= Diff < 0
{ 0, 0, 0, 0, 0, 0, 0, 0}, // 0 <= Diff < 10
{ 0, 0, 0, 0, 2, 3, -2, 5}, // 10 <= Diff < 20
{ 0, 0, 0, 4, 17, 15, 13, 11} // 20 <= Diff
{ 0, -2, -2, -1, 1, 3, 0, 5}, // 10 <= Diff < 20
{ -2, -3, 0, -1, 16, 14, 13, 11} // 20 <= Diff
}
}; // Us, (diff+30)/10-1, (sc+40)/10-1 to index into here

Expand All @@ -69,10 +69,10 @@ TUNE(SetRange(-72,72), whiteDef, whiteAtk, blackDef, blackAtk);
{
for (int j=0; j<4; j++)
{
dynCon[WHITE][i][j+4] = whiteDef[i][j];
dynCon[WHITE][i+4][j] = whiteAtk[i][j];
dynCon[BLACK][i][j+4] = blackDef[i][j];
dynCon[BLACK][i+4][j] = blackAtk[i][j];
dynCon[WHITE][i+2][j] = whiteDef[i][j];
dynCon[WHITE][i+2][j+4] = whiteAtk[i][j];
dynCon[BLACK][i+2][j] = blackDef[i][j];
dynCon[BLACK][i+2][j+4] = blackAtk[i][j];
}
}
}
Expand Down

2 comments on commit f24ec9e

@GuardianRM
Copy link

@GuardianRM GuardianRM commented on f24ec9e Jan 12, 2018 •
edited

Choose a reason for hiding this comment

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

Recently I held a tournament of engines on ultrashort management
(0m + 1s). On such a control, the situation with the discrepancy in the assessment
up to +3 for the first engine and -0.2 for the second at a level of 3000-3200 occurs quite often.
It is possible that the current version can be expanded to, for example,
0 4 10 18 28 40 54 70
or, perhaps, in more detail.

The tests seem close to success, and it looks like a more detailed division can improve the situation even more.

@xoto10
Copy link
Owner

@xoto10 xoto10 commented on f24ec9e Jan 12, 2018

Choose a reason for hiding this comment

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

Cheers!
My 50000 game test at home has finished now and gave +2.6 Elo (W-L +449), but somehow fishtest didn't agree (dynamic2 test only gave +69 in 45k games).
As you say, it does look like this can work, I will try a few more things ...

Please sign in to comment.