Simplification: use Arctan for the optimism S-curve

This was an idea of Gontran Lemaire (gonlem), but the graphs
he published did not seem accurate to me. I did my own graphs,
got my own constants and here is the result:

STC:
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 42179 W: 8704 L: 8622 D: 24853
http://tests.stockfishchess.org/tests/view/5a9db6270ebc590297cb611b

LTC:
LLR: 2.96 (-2.94,2.94) [-3.00,1.00]
Total: 17737 W: 2702 L: 2577 D: 12458
http://tests.stockfishchess.org/tests/view/5a9eb5b20ebc590297cb61b2

Note: we are now back to 70% draw rate in selfplay mode!

Bench: 5544908

diff --git a/src/search.cpp b/src/search.cpp
index 548ea9143d37c48f3aeec316744803a797c13aa1..832501f2f70d94437625b35c836692b63f86c2fa 100644 (file)
--- a/src/search.cpp
+++ b/src/search.cpp
@@ -351,10 +351,7 @@ void Thread::search() {
               ct =  Options["Contempt"] * PawnValueEg / 100; // From centipawns
 
               // Adjust contempt based on current bestValue (dynamic contempt)
-              int sign = (bestValue > 0) - (bestValue < 0);
-              ct +=  bestValue >  500 ?  70 :
-                     bestValue < -500 ? -70 :
-                     bestValue / 10 + sign * int(std::round(3.22 * log(1 + abs(bestValue))));
+              ct += int(std::round(48 * atan(float(bestValue) / 128)));
 
               Eval::Contempt = (us == WHITE ?  make_score(ct, ct / 2)
                                             : -make_score(ct, ct / 2));