Double HistoryMax and reduce aging

After history accounting rewrite in 1.6, a small
tweak of history parameters seems positive.

Note that these are not to be considered the optimal
values, just a wild guess that proved good.

Finding the optimal values would require a much longer
testing time.

After 967 games at 1+0

Mod vs Orig 240 529 198 +15 ELO

Functionality Signature: 21222553

Signed-off-by: Marco Costalba <mcostalba@gmail.com>

diff --git a/src/history.cpp b/src/history.cpp
index 949ba316420d8dc5351e412e32f1e23b414b261a..3cda37061d6d55a201413736323d927b24514158 100644 (file)
--- a/src/history.cpp
+++ b/src/history.cpp
@@ -61,7 +61,7 @@ void History::success(Piece p, Square to, Depth d) {
   if (history[p][to] >= HistoryMax)
       for (int i = 0; i < 16; i++)
           for (int j = 0; j < 64; j++)
-              history[i][j] /= 4;
+              history[i][j] /= 2;
 }
 
 

diff --git a/src/history.h b/src/history.h
index 5c394cc8f17e2ed75f043ba01b0251e607127222..ddab6cd09f442b57e139d3e2e146c35711c75cc8 100644 (file)
--- a/src/history.h
+++ b/src/history.h
@@ -61,13 +61,13 @@ private:
 
 /// HistoryMax controls how often the history counters will be scaled down:
 /// When the history score for a move gets bigger than HistoryMax, all
-/// entries in the table are divided by 4. It is difficult to guess what
+/// entries in the table are divided by 2. It is difficult to guess what
 /// the ideal value of this constant is. Scaling down the scores often has
 /// the effect that parts of the search tree which have been searched
 /// recently have a bigger importance for move ordering than the moves which
 /// have been searched a long time ago.
 
-const int HistoryMax = 25000 * OnePly;
+const int HistoryMax = 50000 * OnePly;
 
 
 #endif // !defined(HISTORY_H_INCLUDED)