This patch reverts the recent commit called "Tweak reductions formula, etc."
The decisions for the revert decision were as follows:
1) The original commit called "Tweak reductions formula: 0.88 * depth + 0.12"
showed bad scaling at in a Very Long Time Control (VLTC) test:
VLTC (180+1.8):
LLR: -1.59 (-2.94,2.94) [0.00,5.00]
Total: 14968 W: 2247 L: 2257 D: 10464
http://tests.stockfishchess.org/tests/view/
5b559ffa0ebc5902bdb84f36
2) So there was a suspicion that the original fast passing LTC test which lead
us to accept the patch may have been a statistical accident, so we organized
a match against the previous master at LTC to get an Elo estimate for the
patch:
LTC match:
ELO: -1.83 +-2.1 (95%) LOS: 4.3%
Total: 36018 W: 6018 L: 6208 D: 23792
http://tests.stockfishchess.org/tests/view/
5b55f8110ebc5902bdb8526f
3) Based on these results, we ran a simplification test with [-3..1] bounds
for the revert at LTC:
LTC:
LLR: 2.95 (-2.94,2.94) [-3.00,1.00]
Total: 41501 W: 7107 L: 7020 D: 27374
http://tests.stockfishchess.org/tests/view/
5b5738670ebc5902bdb86932
4) So we revert.
Bench:
4491691
for (int d = 1; d < 64; ++d)
for (int mc = 1; mc < 64; ++mc)
{
- double slope = d > 2 ? 0.88 * d + 0.36 : d;
- double r = log(slope) * log(mc) / 1.95;
+ double r = log(d) * log(mc) / 1.95;
Reductions[NonPV][imp][d][mc] = int(std::round(r));
Reductions[PV][imp][d][mc] = std::max(Reductions[NonPV][imp][d][mc] - 1, 0);