-
Notifications
You must be signed in to change notification settings - Fork 1
/
Test results.txt
207 lines (104 loc) · 6.66 KB
/
Test results.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
For X - Y - Z the order is: Wins, Losses, Draws
I used to use 30+0.5 and 10+0.2, but lately I've been using 8+0.08
================================================================
Re-write v1 vs v3.0.8: 17 - 14 - 19
Elo +/- Games Score Draw
1 Re-write v2 63 58 100 59.0% 30.0%
2 Re-write v1 -7 58 100 49.0% 28.0%
3 Maxwell v3.0.8 -56 56 100 42.0% 34.0%
This is for when to return a TT evaluation
Re-write v3 (Not root) vs Re-write v3 (Not PV): 131 - 60 - 109
Re-write v3 (Not root) vs Re-write v3 (All): 33 - 33 - 36
Re-write v3 (Depth replace) vs Re-write v3 (Always replace): 74 - 60 - 66
TT replacement schemes
Elo +/- Games Points Draw
1 Re-write v3 (Higher depth, exact bound) 37 37 200 110.5 40.5%
2 Re-write v3 (Higher depth, equals key) -7 37 200 98.0 40.0%
3 Re-write v3 (Higher depth) -30 38 200 91.5 39.5%
v4 contains fixes for Null Move Pruning
Elo +/- Games Points Draw
1 Re-write v4 33 45 150 82.0 34.7%
2 Re-write v3 12 46 150 77.5 32.7%
3 Re-write v2 -44 43 150 65.5 40.7%
Promotion extension tests
Elo +/- Games Points Draw
Pawn one step from promotion 26 45 150 80.5 35.3%
Promotion flag -7 45 150 73.5 35.3%
Neither -19 45 150 71.0 34.7%
Re-write v4 vs Aspiration window drops after out of bounds: 275 - 263 - 262
Re-write v5 (Uses evaluation from the last iteration instead from the last search) vs Re-write v4: 106 - 93 - 101
Removed capture check before doing pruning techniques vs Re-write v5: 126 - 98 - 76
Capture check only before NMP vs Re-write v5 (Capture check before any pruning) 54 - 37 - 35
Elo +/- Games Points Draw
Re-write v6 (Capture check before NMP) 17 20 800 419.0 31.3%
Removed capture check 7 20 800 408.0 31.5%
Re-write v5 (Capture check before pruning) -23 20 800 373.0 29.8%
Re-write v7 (RFP before NMP) vs Re-write v6: 136 - 120 - 144
Re-write v8 (New PVS) vs Re-write v7: 161 - 117 - 122
Re-write v9 (LMR & PVS changes) vs Re-write v8 (New PVS): 418 - 241 - 341
Re-write v10 vs Re-write v9: 138 - 135 - 127
Re-write v10 vs Maxwell v3.0.8: 754 - 171 - 331 (1256 games!)
Re-write v10 vs Added extension to PVS searches: 99 - 85 - 116 (?)
Re-write v11 vs Re-write v10: 376 - 343 - 481
With alpha TT stores vs Without alpha TT stores (Re-write v11): 239 - 256 - 355
Don't allow more than one null move in the search tree at once vs Re-write v11: 175 - 167 - 268
Techincally a win, but the scores were fluctuating a good bit, and I'm not so sure about it
v12 vs v11 was equal, because v12 was just a bit of code cleanup
Strelka razoring vs Re-write v12: 234 - 304 - 368
NMP changes (> 1, and 3 + (depth - 2) / 5) vs Re-write v12: 255 - 254 - 391
Re-write v13 (NMP changes: > 1, and 3 + (depth - 2) / 3) vs Re-write v12: 171 - 122 - 207
History indexed by [from][to] vs Re-write v13: 76 - 77 - 113
Re-write v14 (History indexed by [side to move][from][to]) vs Re-write v13: 152 - 125 - 223
The move order penalty when the target square is attacked by the enemy
Elo +/- Games Points Wins Draws
Re-write v15 (Removed) 27 16 1000 539.0 320 438
Re-write v14 -13 16 1000 481.0 259 444
Re-write v15 (Quiets only) -14 16 1000 480.0 262 436
Countermoves vs Re-write v15: 122 - 177 - 201
v16 changed TT to use a Vec instead of a HashMap, so there should be no strength difference
Re-write v17 (Re-wrote killer moves) vs Re-write v16: 173 - 122 - 205
Re-write v17 vs Maxwell v3.0.8 (Patch 2): 968 - 136 - 476 (1580 games, and an estimated +190 Elo!!)
Count plies since null moves in is_repetition vs Re-write v17: 140 - 149 - 211
Re-write v18 (Add null moves to repetition lookback) vs Re-write v17: 350 - 327 - 523
Re-write v19 (FINALLY FIXED REPETITION DETECTION) vs Re-write v18: 509 - 429 - 662
Incremental move sorter like Rustic vs Re-write v19: 153 - 171 - 176
Extensions outside the move loop vs Re-write v19: 576 - 581 - 593
Only check extension, outside the move loop vs Re-write v19: 272 - 276 - 252
Re-write v20 (Rewrote extension code) vs Re-write v19: 171 - 163 - 166
(Re-write v20's RFP Threshold is 60 centipawns per ply)
RFP = 80 vs Re-write v20: 165 - 172 - 173
RFP = 90 vs Re-write v20: 168 - 181 - 151
RFP = 100 vs Re-write v20: 423 - 519 - 508
At this point I changed RFP to return beta (because of fail hard)
RFP = 16 * (depth - 1).pow(2) + 50 vs Re-write v20: 196 - 217 - 191
RFP = 30 * (depth - 1).pow(2) + 60 vs Re-write v20: 133 - 150 - 176
RFP = 20 * (depth - 1).pow(2) + 65 vs Re-write v20: 140 - 163 - 147
RFP = 40 * (depth - 1).pow(2) + 70 vs Re-write v20: 486 - 498 - 516
FINALLY
Re-write v21 (RFP = 50 * (depth - 1).pow(2) + 55) vs Re-write v20: 276 - 252 - 272
Hmm
Store Alpha bound TT entries vs Re-write v21: 283 - 308 - 311
No TT Cutoff on PV-nodes vs Re-write v21: 514 - 521 - 573
Internal Iterative Reductions
Only on PV, Depth > 2, Depth -= 1 vs Re-write v21: 254 - 255 - 253
Re-write v22 (All nodes, Depth > 1, Depth -= 1) vs Re-write v21: 285 - 234 - 247
All nodes, Depth > 2, Depth -= 2 vs Re-write v22: It lost horribly :P
Only return TT eval if not PV, or depth == 0 vs Re-write v22: 498 - 554 - 618
RFP = 55 * (depth - 1).pow(2) + 60 vs Re-write v22: 142 - 145 - 153
Razoring change (depth -= 2) vs Re-write v22: 533 - 556 - 687
Razoring change (threshold <= alpha) vs Re-write v22: 122 - 132 - 146
Razoring change (285 + 200 * (depth - 1)) vs Re-write v22: 205 - 198 - 207
Razoring change (300 + 150 * (depth - 1)) vs Re-write v22: 265 - 225 - 266
Only NMP if no hash move vs Re-write v23: 485 - 520 - 597
Aspiration Windows
Elo +/- Games Points Wins Draws Draw
AW=30 12 24 500 258.5 165 187 37.4%
AW=60 -2 25 500 248.5 167 163 32.6%
Re-write v23 (40) -5 25 500 246.5 164 165 33.0%
AW=50 -5 24 500 246.5 148 197 39.4%
Hmm
AW=30 vs Re-write v32: 147 - 148 - 145
Maxwell v3.1 (Re-write v23) vs Maxwell v3.0.8 (Patch 2): 1533 - 155 - 312
Elo difference: 293.9 +/- 17.6, LOS: 100.0 %
Tests TODO:
TT enabled vs TT disabled