Glaurung 2 - epsilon/2

[Marc MP]

Thank you very much for the engine,

Tactically epsilon2 is already looking very good. I ran epd tests Arasan7+ECMGCP (409 positions) at 30sec per positions, 64M hash on my old Athlon 1.4Ghz.

SlowChess WV 2.1 328
Glaurung 2-epsilon/2 326
Glaurung 1.2.1 324
Strelka v.1.0.beta 311
Rybka 1.0 Beta - very tactical 309
Toga 1.2.1a 308

Code: Select all

Results: Glaurung 2-epsilon/2

<=Sec Solved Total  PosNr
----- ------ ------ -----------------------------------------------------------
    1    135    135   1   7  15  21  32  35  45  48  50  51  52  54  57  60  61
                     66  69  72  74  76  77  79  89  92  93  94 100 102 108 117
                    121 123 127 133 136 138 145 152 168 181 186 188 189 203 204
                    205 211 218 224 228 231 235 237 239 240 242 243 245 246 247
                    248 249 250 253 255 256 262 264 265 266 269 271 272 274 275
                    276 277 283 284 285 286 288 289 291 292 295 298 299 300 301
                    302 304 306 307 309 312 313 314 315 316 317 318 322 325 326
                    327 328 333 336 339 341 342 343 344 349 351 355 356 357 358
                    360 361 369 373 376 385 389 391 393 395 397 398 401 404 407
    2     32    167   3  10  14  18  29  33  36  39  87  88 101 182 191 195 219
                    222 232 263 267 268 278 281 303 340 350 359 365 368 379 386
                    400 402
    3     28    195  13  19  20  27  31  41  71  85  99 107 110 150 153 156 172
                    192 194 233 241 273 293 294 334 337 348 366 381 399
    4     21    216  11  12  16  30  63 105 120 166 176 214 261 290 305 311 335
                    370 371 374 380 384 406
    5     10    226   4 124 161 193 252 254 260 279 396 403
    6     16    242   6  17  47  82  91  97 106 134 140 157 210 216 234 338 367
                    375
    7      6    248 104 170 198 212 329 331
    8      4    252  24  37  90 394
    9      4    256   8 141 184 280
   10      3    259  84 115 346
   11      6    265  75  96 287 319 364 387
   12      3    268 132 332 410
   13      6    274   9  42 162 169 345 390
   14      6    280  43  80  86 146 320 363
   15      3    283 171 200 229
   16      5    288 129 202 244 258 353
   17      5    293 201 208 372 382 383
   18      7    300  26  95 113 143 151 213 392
   19      5    305  23  70  83 118 128
   21      3    308  34 179 226
   22      3    311  40  55 119
   23      3    314  44 236 377
   24      4    318   2  56 116 175
   25      2    320 167 324
   26      1    321 177
   28      1    322 154
   29      2    324  98 112
   30      2    326  49 144
Failure               5  22  25  28  38  46  53  58  59  62  64  65  67  68  73
                     78  81 103 109 111 114 122 125 126 130 131 135 137 139 142
                    147 148 149 155 158 159 160 163 164 165 173 174 178 180 183
                    185 187 190 196 197 199 206 207 209 215 217 220 221 223 225
                    227 230 238 251 257 259 270 282 296 297 308 310 321 323 330
                    347 352 354 362 378 388 405 408 409

326 problems solved.
84 problems unsolved.

For those curious about Strelka (I don't have output like that for the other engines):

Code: Select all

Results: Strelka v.1.0.beta

<=Sec Solved Total  PosNr
----- ------ ------ -----------------------------------------------------------
    1    135    135   1   4  15  18  21  30  37  39  44  47  48  50  51  54  56
                     57  61  63  68  69  72  89  93  94 100 102 103 107 110 118
                    121 123 124 138 152 154 155 159 163 168 181 183 186 205 210
                    211 221 225 228 230 231 232 233 234 236 237 238 240 245 246
                    247 248 249 250 256 258 260 263 264 265 266 269 274 275 276
                    277 278 280 283 284 285 286 288 291 292 294 295 298 300 301
                    302 304 306 307 309 314 315 316 317 322 324 325 327 328 333
                    336 339 341 342 348 349 351 353 355 356 357 360 361 365 366
                    369 373 376 378 385 387 389 393 395 397 398 400 401 402 407
    2     29    164   3   8  11  12  13  14  36  85  92 101 127 157 208 209 242
                    253 268 273 299 303 312 313 335 350 367 371 381 391 399
    3     21    185  19  29  32  35  41  45  60  64  95 108 133 206 262 267 272
                    281 289 320 380 394 406
    4     11    196   5  74  75 113 162 184 244 254 261 340 343
    5     17    213   2  24  38  59  73 185 195 198 214 239 279 326 331 363 368
                    375 404
    6     10    223  90  97 134 144 177 191 243 290 372 386
    7      7    230  82 143 153 171 229 235 374
    8      4    234  78 140 161 241
    9      7    241 131 189 202 216 252 271 396
   10      4    245  43  62  99 172
   11     11    256  31  53  55  91 115 147 200 226 227 318 358
   12      9    265  16  27  34  83 128 176 224 338 377
   13      6    271  52 112 146 213 218 346
   14      8    279  49  71 105 132 165 169 287 293
   15      4    283  88 337 359 364
   16      4    287  79  86 111 194
   17      4    291  10  77 329 410
   18      1    292 384
   20      2    294  70  87
   21      3    297  46 188 383
   23      3    300  20  33 175
   24      1    301 114
   25      2    303   6  28
   26      2    305 192 219
   28      1    306  40
   29      3    309  84 201 308
   30      2    311  26  98
Failure               7   9  17  22  23  25  42  58  65  66  67  76  80  81  96
                    104 106 109 116 117 119 120 122 125 126 129 130 135 136 137
                    139 141 142 145 148 149 150 151 156 158 160 164 166 167 170
                    173 174 178 179 180 182 187 190 193 196 197 199 203 204 207
                    212 215 217 220 222 223 251 255 257 259 270 282 296 297 305
                    310 311 319 321 323 330 332 334 344 345 347 352 354 362 370
                    379 382 388 390 392 403 405 408 409

311 problems solved.
99 problems unsolved.

[BBauer]

Thanks for your efford Tord!

I ran your Glaurungw64 on Vista with the position you posted before.

Code: Select all

FEN: rq1r2k1/1ppn1pp1/p1n1p2p/P1PpP2P/1P1P2Nb/2NQ4/R2B1PP1/1R4K1 w - - 0 1 

Glaurung-w64:
   2	00:00	         197	439	-0,13	Ng4xh6+ g7xh6 Bd2xh6
   3	00:00	         507	1.129	-1,13	Ng4xh6+ g7xh6 Bd2xh6 f7f5
   3	00:00	         720	1.603	-0,31	Bd2f4 Bh4e7 Ra2e2
   4	00:00	       3.867	8.555	-0,17	Bd2f4 Rd8e8 Ng4e3 Qb8a7
   4	00:00	       4.649	10.262	-0,07	Ng4xh6+ g7xh6 Bd2xh6 Rd8e8 Ra2e2
   4	00:00	       5.778	12.726	 0,00	Bd2xh6 g7xh6 Ng4xh6+ Kg8f8 Qd3h7
   5	00:00	       8.122	17.772	-0,80	Bd2xh6 g7xh6 Ng4xh6+ Kg8g7 b4b5 Kg7xh6 b5xc6
   5	00:00	       9.315	20.294	-0,21	Ng4xh6+ g7xh6 Bd2xh6 Rd8e8 g2g3 Bh4e7
   6	00:00	      22.983	48.385	-0,05	Ng4xh6+ g7xh6 Bd2xh6 Ra8a7 Ra2c2 Rd8e8 Rc2e2
   7	00:00	      38.742	79.065	-0,21	Ng4xh6+ g7xh6 Bd2xh6 Rd8e8 Ra2e2 Qb8a7 g2g3 Bh4e7
   8	00:00	     125.651	221.998	-0,29	Ng4xh6+ g7xh6 Bd2xh6 Rd8e8 Nc3e2 Qb8d8 b4b5 a6xb5 Rb1xb5 Qd8c8
   9	00:00	     250.193	370.107	 0,00	Ng4xh6+ g7xh6 Bd2xh6 Nc6e7 g2g3 Ne7f5 Bh6f4 Bh4e7 Nc3xd5 e6xd5 Qd3xf5
  10	00:01	     671.675	634.254	 0,00	Ng4xh6+ g7xh6 Bd2xh6 Nc6e7 g2g3 Ne7f5 Bh6f4 Bh4e7 Nc3xd5 e6xd5 Qd3xf5
  10	00:02	   2.113.426	903.946	+0,35	Nc3e2 Bh4e7 Qd3g3 Kg8h8 Bd2xh6 g7xh6 Ng4xh6 Rd8f8 Qg3g4 Qb8a7
  11	00:03	   2.978.003	962.820	+0,19	Nc3e2 Qb8a7 Qd3h3 Bh4e7 Qh3g3 Kg8h8 Bd2xh6 g7xh6 Ng4xh6 Rd8f8 Ra2d2
  12	00:04	   4.523.132	1.018.035	+0,23	Nc3e2 Kg8h8 b4b5 a6xb5 Rb1xb5 Qb8a7 Ra2b2 Ra8b8 Bd2xh6 g7xh6 Ng4xh6 Rd8f8 Ne2f4 Nc6xa5
  13	00:07	   7.737.196	1.068.230	+0,29	Nc3e2 Kg8h8 b4b5 a6xb5 Rb1xb5 Qb8a7 Ra2b2 Ra8b8 Qd3b1 Qa7a6 Bd2xh6 f7f5 e5xf6/ep g7xh6 Ng4xh6 Bh4xf6
  14	00:16	  19.085.752	1.124.742	+0,47	Nc3e2 Qb8a7 Qd3f3 Kg8f8 Ng4xh6 g7xh6 Bd2xh6+ Kf8e8 Qf3g4 Bh4e7 Qg4g8+ Nd7f8 Ra2d2 Ra8c8 Ne2f4 Qa7a8
  14	00:20	  23.300.174	1.133.221	+0,62	Bd2xh6 g7xh6 Ng4xh6+ Kg8g7 Nh6xf7 Kg7xf7 Qd3g6+ Kf7e7 Qg6g4 Ke7f7 Qg4xh4 Rd8g8 h5h6 Qb8d8 Qh4f4+ Kf7e7 h6h7
  15	00:27	  32.003.320	1.146.086	+0,66	Bd2xh6 g7xh6 Ng4xh6+ Kg8g7 Nh6xf7 Kg7xf7 Qd3g6+ Kf7e7 Qg6g4 Ke7f7 Qg4xh4 Rd8h8 Qh4f4+ Kf7e7 h5h6 Qb8f8 Qf4h4+ Ke7f7 h6h7
  16	00:44	  51.670.829	1.162.605	+0,62	Bd2xh6 g7xh6 Ng4xh6+ Kg8g7 Nh6xf7 Kg7xf7 Qd3g6+ Kf7e7 Qg6g4 Ke7f7 Qg4xh4 Rd8h8 Qh4f4+ Kf7e7 h5h6 Ke7e8 Ra2e2 Qb8a7 Re2e3
  16	01:21	  95.303.309	1.171.882	+1,29	g2g3 Bh4g5 Bd2xg5 h6xg5 f2f3 Kg8h8 h5h6 f7f5 e5xf6/ep g7xf6 Qd3e3 f6f5 Qe3xe6 Nc6xd4 Qe6f7 Nd4xf3+ Kg1g2 Rd8g8 Qf7xd7 f5xg4 Nc3xd5
  17	01:33	 110.061.140	1.175.866	+1,33	g2g3 Bh4g5 Bd2xg5 h6xg5 f2f3 Kg8h8 h5h6 f7f5 e5xf6/ep g7xf6 Qd3e3 f6f5 Qe3xe6 f5xg4 Qe6f7 Rd8g8 Qf7xd7 g4xf3 Qd7xd5
  18	02:30	 178.715.395	1.188.409	+1,31	g2g3 Bh4g5 Bd2xg5 h6xg5 f2f3 Rd8f8 h5h6 g7xh6 Nc3xd5 e6xd5 e5e6 Kg8g7 e6xd7 f7f5 Ng4e5 Nc6xe5 d4xe5 c7c6 Ra2e2 Qb8c7

So Glaurung_eps_2 finds g3 too.
However, I am wondering why the task manager reports only 30% CPU usage. Rybka gives 70%.

Kind regards og en fantastic 17. mai

Bernhard

[Tord Romstad]

Thanks very much for the engine Tord! Everything seems to work in the Shredder interface. I think this Glaurung is indeed playing differently so far than old Glaurung.

Yes, they are two entirely different programs, which are not sharing much apart from the name and the author. I hope to eventually make the style of the new engine resemble the old, though. The current version probably has the most ugly and boring style ever seen in the history of chess. It's disgusting to watch.

But if you implement things like Space again as an eval parameter it can change things dramatically of course. It is possible I think to give it a fairly large value, in the latest Glaurung 1.2.1 I had Space at 130 and I was not really tuning for spectacular play.

Glaurung 2-ε has no problem with 2 times eight Knights on the board, I was offerered a draw playing Glaurung here and should have accepted of course, but got into trouble a bit later. At the end Glaurung ran out of iterations in the ponder search seeing mate in two and Shredder GUI was grumbling a bit, it says here:

[D]6k1/8/3NNN2/n2n4/1n4N1/2n5/3K4/8 w - -

PD zet: 28...Pxf6

PB: 96.00 0:00 +M2 29.Pxf6+ Kh8 30.Pf7+ (5.663) 72
PB: 97.00 0:00 +M2 29.Pxf6+ Kh8 30.Pf7+ (5.785) 74
PB: 98.00 0:00 +M2 29.Pxf6+ Kh8 30.Pf7+ (5.908) 53
PB: 99.00 0:00 +M2 29.Pxf6+ Kh8 30.Pf7+ (6.031) 54
PB: 100.00 0:00 +M2 29.Pxf6+ Kh8 30.Pf7+ (6.174) 56
Engine 'Glaurung 2-epsilon/2': UCI implementation fault: illegal end of ponder search!
beste zet: Pg4xf6 tijd: 0:00.172 min n/s: 56.127 nodes: 6.174

I always forget this when I implement UCI: For some reason I have never understood, when pondering, it is not allowed for the engine to give its best move until the GUI explicitly tells it that it is OK to do so. This bug may cause problems when running Glaurung in ponder-on games, at least in some GUIs. I'll try to remember to solve it before the next version.

Otherwise no bugs seen yet!

Good!

Tord

[Tord Romstad]

Hello Marc!

Thank you very much for the engine,

Tactically epsilon2 is already looking very good. I ran epd tests Arasan7+ECMGCP (409 positions) at 30sec per positions, 64M hash on my old Athlon 1.4Ghz.

That's surprising. The last time I tried running ECMGCP, the results were very poor. Admittedly this was a few weeks ago, but I can't imagine that any of my recent changes has a big impact on the tactical performance.

Thanks for the test results!

Pedantic remark: It is not "epsilon 2", but "epsilon/2", i.e. "epsilon divided by 2". It is usual in mathematics to use the greek letter epsilon to denote some small positive number. Hence "2-epsilon" means a number slightly below 2, while "2-epsilon/2" is a number even closer to 2. I will call my development versions "2-epsilon", "2-epsilon/2", "2-epsilon/3", ... all the way until it is time to release the final Glaurung 2.

Tord

[Tord Romstad]

Thanks for your efford Tord!

Thanks to you for your interest, too!

I ran your Glaurungw64 on Vista with the position you posted before.

Code: Select all

  18	02:30	 178.715.395	1.188.409	+1,31	g2g3 Bh4g5 Bd2xg5 h6xg5 f2f3 Rd8f8 h5h6 g7xh6 Nc3xd5 e6xd5 e5e6 Kg8g7 e6xd7 f7f5 Ng4e5 Nc6xe5 d4xe5 c7c6 Ra2e2 Qb8c7

So Glaurung_eps_2 finds g3 too.

Yes, but not Bxh6, which is the move I would have liked to see it play. I agree with what people observed in the other thread: Both of the moves g3 and Bxh6 seem to win, and the choice of move is probably more a matter of style than a matter of strength or tactical speed. However, I think most of us agree that Bxh6 is a much more attractive move.

However, I am wondering why the task manager reports only 30% CPU usage. Rybka gives 70%.

Yes, that's strange - but I think everything is OK. If I understand correctly, Glaurung searches about 1.2 MN/s from the position above, which is very good.

Kind regards og en fantastic 17. mai

Thanks, but actually I traditionally do my best to stay far away on May 17: This year I am seeking refuge in Amalfi, Italy.

For those who don't know, May 17 is the national day of Norway, celebrated with pompous and patriotic speeches, big parades of children shouting and waving with flags, noisy brass music performed by school bands from the local schools, and huge crowds of "russ", the graduating class in high school, who spend the whole month being dead drunk, making an effort to break as many social taboos as possible, and driving around in buses equipped with monstrous stereos with which they terrorize the neigborhood with music played at maximum volume 24 hours for day, thus making sure that hardly anyone in the major cities of Norway gets a good night of sleep in May. The rationality of doing all these celebrations a short time before their final exam is perhaps questionable, but Norwegians are not among the most intelligent people of the world.

Most Norwegians - and a surprising number of foreigners - find it all very charming. I do no share their view.

Tord