deepcrayon
/
stockfish
1
0
Fork 0
Code Releases Activity

Final touches 

Set the final stddev to 0.25 and add some comment

Current version starts much weaker than master (so to lower the 1350
minimum ELO) and improves consistently with level. Verified with 2 TC,
one short for lower levels and one longer for higher levels.

normal vs master: Level 8  (TC 30+0.1)  ->  -87 ELO
normal vs master: Level 6  (TC 30+0.1)  -> -105 ELO
normal vs master: Level 4  (TC 30+0.1)  -> -105 ELO

normal vs master: Level 4  (TC 10+0.1)  -> -120 ELO
normal vs master: Level 3  (TC 10+0.1)  -> -170 ELO
normal vs master: Level 2  (TC 10+0.1)  -> -295 ELO
pull/3777/head
xefoci7612 2021-11-07 20:33:43 +01:00
parent b41f6b42ae
commit 2d02ea46a1
1 changed files with 10 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

20
src/search.cpp
View File

@ -1768,29 +1768,29 @@ moves_loop: // When in check, search starts here
Move Skill::pick_move(const RootMoves& rootMoves, size_t multiPV) {
static PRNG rng(now()); // PRNG sequence should be non-deterministic
// RootMoves are already sorted by score in descending order
Move best = MOVE_NONE;
int maxScore = -VALUE_INFINITE;
Value topScore = rootMoves[0].score;
int delta = std::min(topScore - rootMoves[multiPV - 1].score, PawnValueMg);
int maxScore = -VALUE_INFINITE;
double weakness = 130 - 2 * level;
// Use a normal distribution to not spread too much the random values, so
// that moves quality remains consistent with the set skill level.
float mean = delta * weakness / 2;
float stddev = mean * 0.2;
std::default_random_engine generator;
std::normal_distribution<float> distribution(mean, stddev);
float stddev = mean * 0.25;
std::default_random_engine rng;
std::normal_distribution<float> normal(mean, stddev);
// Choose best move. For each move score we add two terms, both dependent on
// weakness. One is deterministic and bigger for weaker levels, and one is
// random. Then we choose the move with the resulting highest score.
// random with a normal probability distribution. Then we choose the move with
// the resulting highest score.
for (size_t i = 0; i < multiPV; ++i)
{
// This is our magic formula
int push = int(( weakness * int(topScore - rootMoves[i].score)
+ distribution(generator)) / 128);
int diff_to_top = topScore - rootMoves[i].score;
int push = int((weakness * diff_to_top + normal(rng)) / 128);
if (rootMoves[i].score + push >= maxScore)
{