13-03-18 | mco | 3698d9aa5573ca | diff |
ELO: 15.49 +-4.8 (95%) LOS: 100.0% |
20000
@ 60+0.05 th 1 |
Retest regression with new hash size of 128MB, previous result with 32MB hash size was +16.62 ELO |

13-03-18 | gli | no_pv_dangerous | diff |
ELO: 5.43 +-4.4 (95%) LOS: 100.0% |
24000
@ 15+0.05 th 1 |
Don't extend dangerous moves in PV |

13-03-18 | gli | no_pv_dangerous | diff |
ELO: 2.84 +-4.4 (95%) LOS: 97.9% |
24000
@ 15+0.05 th 1 |
Treat killers and ttMove as dangerous |

13-03-18 | gli | no_pv_dangerous | diff |
ELO: -8.65 +-6.2 (95%) LOS: 0.0% |
24000
@ 15+0.05 th 1 |
No restriction on LMR depth |

13-03-19 | mco | master | diff |
ELO: 26.11 +-15.3 (95%) LOS: 100.0% |
10000
@ 15+0.05 th 8 |
Quick test with SMP enabled (8 threads). This requirement can be satisfied only by Gary's 24 cores machine |

13-03-19 | gli | qsearch_pruning | diff |
ELO: -4.07 +-5.1 (95%) LOS: 0.6% |
24000
@ 15+0.05 th 1 |
Try again with new conditions |

13-03-19 | gli | all_cut_iid | diff |
ELO: 0.72 +-4.4 (95%) LOS: 69.6% |
24000
@ 15+0.05 th 1 |
Tweak IID based on all/cut |

13-03-19 | gli | all_cut_iid | diff |
ELO: 1.56 +-4.4 (95%) LOS: 86.7% |
24000
@ 15+0.05 th 1 |
Bring back different depth for pv/non-pv |

13-03-19 | gli | all_cut_iid | diff |
ELO: -3.71 +-4.4 (95%) LOS: 0.4% |
24000
@ 15+0.05 th 1 |
Use eval as another condition |

13-03-19 | gli | movepick | diff |
ELO: 1.51 +-4.4 (95%) LOS: 86.1% |
24000
@ 15+0.05 th 1 |
Test how valuable negative captures are when close to beta |

13-03-19 | gli | see_movepick | diff |
ELO: -4.56 +-7.6 (95%) LOS: 3.0% |
24000
@ 15+0.05 th 1 |
Use SEE always for ordering evasions |

13-03-20 | gli | f45b170 | diff |
ELO: -2.92 +-6.8 (95%) LOS: 7.6% |
16000
@ 60+0.05 th 1 |
Retest don't extend dangerous moves in PV at long TC |

13-03-20 | gli | all_cut_iid | diff |
ELO: -0.14 +-4.4 (95%) LOS: 45.9% |
24000
@ 15+0.05 th 1 |
Try higher depth for IID |

13-03-20 | gli | all_cut_iid | diff |
ELO: 0.64 +-6.2 (95%) LOS: 62.4% |
24000
@ 15+0.05 th 1 |
Remove another PV distinction |

13-03-20 | gli | lmr | diff |
ELO: 0.47 +-4.9 (95%) LOS: 61.6% |
24000
@ 15+0.05 th 1 |
Increase LMR depth |

13-03-20 | mco | passed_pawns | diff |
ELO: 8.49 +-6.6 (95%) LOS: 100.0% |
24000
@ 15+0.05 th 1 |
Eval tweak by Jundery, seems a +7 ELO on his tests |

13-03-21 | gli | iid | diff |
ELO: 0.43 +-5.4 (95%) LOS: 60.0% |
16000
@ 15+0.05 th 1 |
Always try IID |

13-03-21 | gli | 33f4172 | diff |
ELO: -0.13 +-3.1 (95%) LOS: 46.7% |
16000
@ 60+0.05 th 1 |
Re-test at long TC |

13-03-22 | mco | fix_mate_scores | diff |
ELO: 2.40 +-4.4 (95%) LOS: 95.7% |
24000
@ 15+0.05 th 1 |
Fix by Joona to prevent reporting bogus mate scores |

13-03-22 | mco | bestValue_pruning | diff |
ELO: 4.78 +-3.3 (95%) LOS: 99.8% |
16000
@ 15+0.05 th 1 |
Extend to search what we already do in qsearch |

13-03-22 | gli | all_cut_iid | diff |
ELO: 4.76 +-3.1 (95%) LOS: 99.9% |
24000
@ 15+0.05 th 1 |
Remove eval check for IID |

13-03-22 | gli | null_move | diff |
ELO: -9.45 +-6.3 (95%) LOS: 0.2% |
24000
@ 15+0.05 th 1 |
Bump R up to 4 base, no eval tweak |

13-03-22 | mco | increase_iid | diff |
ELO: 5.97 +-3.3 (95%) LOS: 100.0% |
16000
@ 20+0.05 th 1 |
My attemp at improving IID, this time search deeper Non-Pv in case |

13-03-22 | mco | passed_pawns | diff |
ELO: -0.80 +-5.4 (95%) LOS: 31.0% |
16000
@ 60+0.05 th 1 |
Jundary's patch bumped up at 60" per game |

13-03-23 | mco | passed_pawns | diff |
ELO: 1.69 +-3.1 (95%) LOS: 85.4% |
16000
@ 60+0.05 th 1 |
Another attempt by Jundery. This time we go directly at 60" becuase short TC is not sensible |

13-03-24 | jki | qs_check_depth_zero | diff |
ELO: -2.65 +-3.3 (95%) LOS: 5.7% |
16000
@ 15+0.05 th 1 |
QSearch checks only at depth = 0 |

13-03-24 | jki | qs_no_recaptures | diff |
ELO: 2.15 +-3.3 (95%) LOS: 89.9% |
16000
@ 15+0.05 th 1 |
Check if qsearch recapturing logic is harmful |

13-03-24 | jki | tuning_try_material | diff |
ELO: -3.26 +-3.3 (95%) LOS: 2.7% |
16000
@ 15+0.05 th 1 |
Retest one old tuning try for material coefficients |

13-03-24 | jki | qs_no_equal_see_pruning | diff |
ELO: -4.73 +-3.3 (95%) LOS: 0.3% |
16000
@ 15+0.05 th 1 |
Test usefulness of pruning moves with equal SEE in qsearch |

13-03-24 | jki | qsearch_TT_fixed_depth | diff |
ELO: -1.80 +-3.3 (95%) LOS: 14.5% |
16000
@ 15+0.05 th 1 |
Always use DEPTH_ZERO when accessing TT in qsearch |

13-03-24 | jki | always_check_repetition | diff |
ELO: 2.74 +-2.7 (95%) LOS: 97.6% |
24000
@ 15+0.05 th 1 |
Any gain in skipping repetition detection in qsearch? |

13-03-24 | jki | no_gain_tables | diff |
ELO: -4.19 +-3.3 (95%) LOS: 0.7% |
16000
@ 15+0.05 th 1 |
Measure value of Gain-tables in futility pruning |

13-03-24 | mco | master | diff |
ELO: 16.71 +-2.8 (95%) LOS: 100.0% |
20000
@ 60+0.05 th 1 |
Regression test with bestValue_pruning applied. It's the candidate for next TCEC stage |

13-03-25 | gli | Outpost_Weighting | diff |
ELO: 0.43 +-3.4 (95%) LOS: 60.0% |
16000
@ 15+0.05 th 1 |
RyanTaker idea, reduce outpost weight when major pieces gone |

13-03-25 | gli | Space_Weighting | diff |
ELO: 1.09 +-3.4 (95%) LOS: 73.4% |
16000
@ 15+0.05 th 1 |
RyanTaker idea, have space be valued in endgame as well |

13-03-25 | jki | razor_value_adjust | diff |
ELO: -1.02 +-3.3 (95%) LOS: 27.5% |
16000
@ 15+0.05 th 1 |
Return orthodox value when razoring |

13-03-25 | jki | pv_only_extend_checks | diff |
ELO: -1.15 +-3.4 (95%) LOS: 25.4% |
16000
@ 15+0.05 th 1 |
In PV nodes extend only checks (not all dangerous moves) |

13-03-25 | lbr | lucas | diff |
ELO: -1.63 +-3.4 (95%) LOS: 17.2% |
16000
@ 15+0.05 th 1 |
same reduction formula for PV and non PV nodes |

13-03-25 | lbr | lucas | diff |
ELO: -5.73 +-3.6 (95%) LOS: 0.1% |
16000
@ 15+0.05 th 1 |
LMR smooth transition between search and qsearch |

13-03-25 | gli | see_movepick | diff |
ELO: -3.49 +-4.3 (95%) LOS: 5.8% |
16000
@ 15+0.05 th 1 |
Take 3 for see_movepick |

13-03-25 | lbr | lucas | diff |
ELO: 4.89 +-3.3 (95%) LOS: 99.8% |
16000
@ 15+0.05 th 1 |
cleaup search extensions |

13-03-25 | gli | see_movepick | diff |
ELO: -3.77 +-4.1 (95%) LOS: 3.6% |
16000
@ 15+0.05 th 1 |
Try without sort evasions by SEE |

13-03-25 | jki | null_move2 | diff |
ELO: -1.87 +-3.3 (95%) LOS: 13.6% |
16000
@ 15+0.05 th 1 |
Null move: R = 3 + depth / 4 (no eval tweak) |

13-03-25 | mco | increase_iid | diff |
ELO: -0.07 +-3.2 (95%) LOS: 48.3% |
16000
@ 60+0.05 th 1 |
Retest at longer TC to see if it holds the +6 ELO |

13-03-26 | gli | see_search | diff |
ELO: -26.75 +-4.2 (95%) LOS: 0.0% |
16000
@ 15+0.05 th 1 |
Try sorting see < 0 non-captures to end of list |

13-03-26 | gli | see_search | diff |
ELO: 1.56 +-3.3 (95%) LOS: 82.3% |
16000
@ 15+0.05 th 1 |
Add one ply reduction if SEE is bad for non-capture above pruning depth |

13-03-26 | gli | fail_low_skip_ply | diff |
ELO: -3.63 +-4.6 (95%) LOS: 6.0% |
16000
@ 15+0.05 th 1 |
Old Garbochess idea, skip ply if we continue to fail low |

13-03-26 | jki | qs_check_depth_2 | diff |
ELO: -2.32 +-3.4 (95%) LOS: 9.0% |
16000
@ 15+0.05 th 1 |
Adjust QS Check depth to -2 |

13-03-26 | jki | prune_dangerous_neg_see | diff |
ELO: -2.13 +-3.4 (95%) LOS: 10.8% |
16000
@ 15+0.05 th 1 |
Prune also dangerous moves with negative SEE at low depths |

13-03-26 | gli | all_cut_iid | diff |
ELO: 0.28 +-3.1 (95%) LOS: 57.0% |
16000
@ 60+0.05 th 1 |
Long TC: Remove eval check for IID |