TalkChess.com

White has given a lot for this attack, and this will succeed ! mate in #9

[d] r2q1rk1/ppp1bp2/3pbnp1/7p/4N3/1P1P1N2/PB1Q4/2K4R w - - 0 1

Mmmm just found this in SF case QCAPTURE: if (select<Best>([&](){ return depth > DEPTH_QS_RECAPTURES || to_sq(*cur) == recaptureSquare; })) return *(cur - 1); with DEPTH_QS_RECAPTURES = -5 This may very well be the trick for this kind of position !

Ok, I try recapture bonus even for negative SEE and it works quite well on this position, although still slower than qfutility here.
But in Minic, qfutility is a big elo drop, while this recapture bonus seems to be quite ok ... I'm doing some more testing about this ...

I indeed give a nice bonus to recapture with positive SEE in move ordering but that won't do the trick here.
Futility pruning in qsearch seems to help a lot in this case, but remains quite slow at depth 1 (4sec !), then speed up to reach depth 10 in 21 sec.
I'll try if delta can help also ...

Might be that futility/delta pruning is currently off inside Minic... Wait better when activated !

In this position [d] rnbqkbnr/rrrrrrrr/8/8/8/8/RRRRRRRR/RNBQKBNR w KQkq - 0 1 some engines are instantaneously going to depth 10 while other takes ages to go to depth 1 ... It looks like a qsearch explosion, but Stockfish or Ethereal or demolito have no limited depth qsearch and are very fast on thi...

I wonder if some of you know the speed of the evaluation function of your engine.

To know that, in Minic, I disable pawn hash table, and loop 10 times around the 2.500.000 position of lichess-quiet data.

This gives : 1.6Meval/s on an i7-9700K.

I wonder if this is too slow ...

Black will clear the board here !

[d] 5rk1/6pp/p1q1p3/1pPnP3/4Q1P1/7P/PP1B2K1/3R4 b - - 0 28

Can we ask Lichess crew to let some engines with pawn, knight, ... odds (depending on their rating) play against humans ?
Some hours of such a tournament will probably give some insight on the subject ?

For its worth, here is how I try to do it ... very basic PSO ... batch of 100 games at 1+0.01 tc I will converge ... but take your time ... #!/usr/bin/python # -*- coding: utf-8 -*- from subprocess import Popen, PIPE import sys import platform import random import math import multiprocessing from jo...