- Timestamp:
- 03/06/13 17:33:38 (12 years ago)
- Location:
- branches/CMAES/HeuristicLab.Algorithms.CMAEvolutionStrategy/3.3
- Files:
-
- 5 edited
Legend:
- Unmodified
- Added
- Removed
-
branches/CMAES/HeuristicLab.Algorithms.CMAEvolutionStrategy/3.3/CMAOperators/CMAMutator.cs
r9245 r9291 20 20 #endregion 21 21 22 using System;23 22 using HeuristicLab.Common; 24 23 using HeuristicLab.Core; … … 29 28 using HeuristicLab.Parameters; 30 29 using HeuristicLab.Persistence.Default.CompositeSerializers.Storable; 30 using HeuristicLab.Random; 31 using System; 31 32 32 33 namespace HeuristicLab.Algorithms.CMAEvolutionStrategy { … … 113 114 RealVectorParameter.ActualValue = arx; 114 115 } 115 alglib.hqrndstate state; 116 alglib.hqrndseed(random.Next(), random.Next(), out state); 116 var nd = new NormalDistributedRandom(random, 0, 1); 117 //alglib.hqrndstate state; 118 //alglib.hqrndseed(random.Next(), random.Next(), out state); 117 119 118 120 for (int i = 0; i < lambda; i++) { … … 122 124 for (int k = 0; k < arx[i].Length; k++) { 123 125 do { 124 arx[i][k] = xmean[k] + sp.Sigma.Value * sp.D[k] * alglib.hqrndnormal(state);126 arx[i][k] = xmean[k] + sp.Sigma.Value * sp.D[k] * nd.NextDouble(); //alglib.hqrndnormal(state); 125 127 inRange = bounds[k % bounds.Rows, 0] <= arx[i][k] && arx[i][k] <= bounds[k % bounds.Rows, 1]; 126 128 if (!inRange) tries++; … … 138 140 var artmp = new double[arx[0].Length]; 139 141 for (int k = 0; k < arx[0].Length; ++k) { 140 artmp[k] = sp.D[k] * alglib.hqrndnormal(state);142 artmp[k] = sp.D[k] * nd.NextDouble(); //alglib.hqrndnormal(state); 141 143 } 142 144 -
branches/CMAES/HeuristicLab.Algorithms.CMAEvolutionStrategy/3.3/CMAOperators/CMAUpdater.cs
r9148 r9291 237 237 DegenerateStateParameter.ActualValue = new BoolValue(false); 238 238 } else { 239 var c = new double[sp.C.Rows, sp.C.Columns];239 /*var c = new double[sp.C.Rows, sp.C.Columns]; 240 240 for (int i = 0; i < sp.C.Rows; i++) 241 241 for (int j = 0; j < sp.C.Columns; j++) … … 243 243 244 244 double[] eVal; 245 double[,] eVec; 246 var success = alglib.smatrixevd(c, N, 1, false, out eVal, out eVec); 245 double[,] eVec;*/ 246 247 // set B <- C 248 for (int i = 0; i < N; ++i) 249 for (int j = 0; j <= i; ++j) 250 sp.B[i, j] = sp.B[j, i] = sp.C[i, j]; 251 252 var success = Eigendecomposition(N, sp.B, sp.D); // alglib.smatrixevd(c, N, 1, false, out eVal, out eVec); 247 253 DegenerateStateParameter.ActualValue = new BoolValue(!success); 248 sp.B = new DoubleMatrix(eVec);249 250 for (int i = 0; i < sp.D.Length; i++) {254 //sp.B = new DoubleMatrix(eVec); 255 256 /*for (int i = 0; i < sp.D.Length; i++) { 251 257 if (eVal[i] < 0) { 252 258 DegenerateStateParameter.ActualValue.Value = true; 253 259 sp.D[i] = 0; 254 260 } else sp.D[i] = Math.Sqrt(eVal[i]); 255 } 261 }*/ 256 262 257 263 if (sp.D.Min() == 0.0) sp.AxisRatio.Value = double.PositiveInfinity; … … 260 266 return base.Apply(); 261 267 } 268 269 private bool Eigendecomposition(int N, DoubleMatrix B, DoubleArray diagD) { 270 bool result = true; 271 // eigendecomposition 272 var offdiag = new double[N]; 273 tred2(N, B, diagD, offdiag); 274 tql2(N, diagD, offdiag, B); 275 276 // assign diagD to eigenvalue square roots 277 for (int i = 0; i < N; ++i) { 278 if (diagD[i] < 0) // numerical problem? 279 result = false; 280 else diagD[i] = Math.Sqrt(diagD[i]); 281 } 282 283 return result; 284 } // eigendecomposition 285 286 287 // Symmetric Householder reduction to tridiagonal form, taken from JAMA package. 288 private void tred2(int n, DoubleMatrix V, DoubleArray d, double[] e) { 289 290 // This is derived from the Algol procedures tred2 by 291 // Bowdler, Martin, Reinsch, and Wilkinson, Handbook for 292 // Auto. Comp., Vol.ii-Linear Algebra, and the corresponding 293 // Fortran subroutine in EISPACK. 294 295 for (int j = 0; j < n; j++) { 296 d[j] = V[n - 1, j]; 297 } 298 299 // Householder reduction to tridiagonal form. 300 301 for (int i = n - 1; i > 0; i--) { 302 303 // Scale to avoid under/overflow. 304 305 double scale = 0.0; 306 double h = 0.0; 307 for (int k = 0; k < i; k++) { 308 scale = scale + Math.Abs(d[k]); 309 } 310 if (scale == 0.0) { 311 e[i] = d[i - 1]; 312 for (int j = 0; j < i; j++) { 313 d[j] = V[i - 1, j]; 314 V[i, j] = 0.0; 315 V[j, i] = 0.0; 316 } 317 } else { 318 319 // Generate Householder vector. 320 321 for (int k = 0; k < i; k++) { 322 d[k] /= scale; 323 h += d[k] * d[k]; 324 } 325 double f = d[i - 1]; 326 double g = Math.Sqrt(h); 327 if (f > 0) { 328 g = -g; 329 } 330 e[i] = scale * g; 331 h = h - f * g; 332 d[i - 1] = f - g; 333 for (int j = 0; j < i; j++) { 334 e[j] = 0.0; 335 } 336 337 // Apply similarity transformation to remaining columns. 338 339 for (int j = 0; j < i; j++) { 340 f = d[j]; 341 V[j, i] = f; 342 g = e[j] + V[j, j] * f; 343 for (int k = j + 1; k <= i - 1; k++) { 344 g += V[k, j] * d[k]; 345 e[k] += V[k, j] * f; 346 } 347 e[j] = g; 348 } 349 f = 0.0; 350 for (int j = 0; j < i; j++) { 351 e[j] /= h; 352 f += e[j] * d[j]; 353 } 354 double hh = f / (h + h); 355 for (int j = 0; j < i; j++) { 356 e[j] -= hh * d[j]; 357 } 358 for (int j = 0; j < i; j++) { 359 f = d[j]; 360 g = e[j]; 361 for (int k = j; k <= i - 1; k++) { 362 V[k, j] -= (f * e[k] + g * d[k]); 363 } 364 d[j] = V[i - 1, j]; 365 V[i, j] = 0.0; 366 } 367 } 368 d[i] = h; 369 } 370 371 // Accumulate transformations. 372 373 for (int i = 0; i < n - 1; i++) { 374 V[n - 1, i] = V[i, i]; 375 V[i, i] = 1.0; 376 double h = d[i + 1]; 377 if (h != 0.0) { 378 for (int k = 0; k <= i; k++) { 379 d[k] = V[k, i + 1] / h; 380 } 381 for (int j = 0; j <= i; j++) { 382 double g = 0.0; 383 for (int k = 0; k <= i; k++) { 384 g += V[k, i + 1] * V[k, j]; 385 } 386 for (int k = 0; k <= i; k++) { 387 V[k, j] -= g * d[k]; 388 } 389 } 390 } 391 for (int k = 0; k <= i; k++) { 392 V[k, i + 1] = 0.0; 393 } 394 } 395 for (int j = 0; j < n; j++) { 396 d[j] = V[n - 1, j]; 397 V[n - 1, j] = 0.0; 398 } 399 V[n - 1, n - 1] = 1.0; 400 e[0] = 0.0; 401 } 402 403 // Symmetric tridiagonal QL algorithm, taken from JAMA package. 404 private void tql2(int n, DoubleArray d, double[] e, DoubleMatrix V) { 405 406 // This is derived from the Algol procedures tql2, by 407 // Bowdler, Martin, Reinsch, and Wilkinson, Handbook for 408 // Auto. Comp., Vol.ii-Linear Algebra, and the corresponding 409 // Fortran subroutine in EISPACK. 410 411 for (int i = 1; i < n; i++) { 412 e[i - 1] = e[i]; 413 } 414 e[n - 1] = 0.0; 415 416 double f = 0.0; 417 double tst1 = 0.0; 418 double eps = Math.Pow(2.0, -52.0); 419 for (int l = 0; l < n; l++) { 420 421 // Find small subdiagonal element 422 423 tst1 = Math.Max(tst1, Math.Abs(d[l]) + Math.Abs(e[l])); 424 int m = l; 425 while (m < n) { 426 if (Math.Abs(e[m]) <= eps * tst1) { 427 break; 428 } 429 m++; 430 } 431 432 // If m == l, d[l] is an eigenvalue, 433 // otherwise, iterate. 434 435 if (m > l) { 436 int iter = 0; 437 do { 438 iter = iter + 1; // (Could check iteration count here.) 439 440 // Compute implicit shift 441 442 double g = d[l]; 443 double p = (d[l + 1] - g) / (2.0 * e[l]); 444 double r = hypot(p, 1.0); 445 if (p < 0) { 446 r = -r; 447 } 448 d[l] = e[l] / (p + r); 449 d[l + 1] = e[l] * (p + r); 450 double dl1 = d[l + 1]; 451 double h = g - d[l]; 452 for (int i = l + 2; i < n; i++) { 453 d[i] -= h; 454 } 455 f = f + h; 456 457 // Implicit QL transformation. 458 459 p = d[m]; 460 double c = 1.0; 461 double c2 = c; 462 double c3 = c; 463 double el1 = e[l + 1]; 464 double s = 0.0; 465 double s2 = 0.0; 466 for (int i = m - 1; i >= l; i--) { 467 c3 = c2; 468 c2 = c; 469 s2 = s; 470 g = c * e[i]; 471 h = c * p; 472 r = hypot(p, e[i]); 473 e[i + 1] = s * r; 474 s = e[i] / r; 475 c = p / r; 476 p = c * d[i] - s * g; 477 d[i + 1] = h + s * (c * g + s * d[i]); 478 479 // Accumulate transformation. 480 481 for (int k = 0; k < n; k++) { 482 h = V[k, i + 1]; 483 V[k, i + 1] = s * V[k, i] + c * h; 484 V[k, i] = c * V[k, i] - s * h; 485 } 486 } 487 p = -s * s2 * c3 * el1 * e[l] / dl1; 488 e[l] = s * p; 489 d[l] = c * p; 490 491 // Check for convergence. 492 493 } while (Math.Abs(e[l]) > eps * tst1); 494 } 495 d[l] = d[l] + f; 496 e[l] = 0.0; 497 } 498 499 // Sort eigenvalues and corresponding vectors. 500 501 for (int i = 0; i < n - 1; i++) { 502 int k = i; 503 double p = d[i]; 504 for (int j = i + 1; j < n; j++) { 505 if (d[j] < p) { // NH find smallest k>i 506 k = j; 507 p = d[j]; 508 } 509 } 510 if (k != i) { 511 d[k] = d[i]; // swap k and i 512 d[i] = p; 513 for (int j = 0; j < n; j++) { 514 p = V[j, i]; 515 V[j, i] = V[j, k]; 516 V[j, k] = p; 517 } 518 } 519 } 520 } 521 522 /** sqrt(a^2 + b^2) without under/overflow. **/ 523 private double hypot(double a, double b) { 524 double r = 0; 525 if (Math.Abs(a) > Math.Abs(b)) { 526 r = b / a; 527 r = Math.Abs(a) * Math.Sqrt(1 + r * r); 528 } else if (b != 0) { 529 r = a / b; 530 r = Math.Abs(b) * Math.Sqrt(1 + r * r); 531 } 532 return r; 533 } 262 534 } 263 535 } -
branches/CMAES/HeuristicLab.Algorithms.CMAEvolutionStrategy/3.3/HeuristicLab.Algorithms.CMAEvolutionStrategy-3.3.csproj
r9260 r9291 93 93 </PropertyGroup> 94 94 <ItemGroup> 95 <Reference Include="ALGLIB-3.6.0, Version=3.6.0.0, Culture=neutral, PublicKeyToken=ba48961d6f65dcec, processorArchitecture=MSIL">96 <SpecificVersion>False</SpecificVersion>97 <HintPath>..\..\..\..\trunk\sources\bin\ALGLIB-3.6.0.dll</HintPath>98 <Private>False</Private>99 </Reference>100 95 <Reference Include="HeuristicLab.Analysis-3.3, Version=3.3.0.0, Culture=neutral, PublicKeyToken=ba48961d6f65dcec, processorArchitecture=MSIL"> 101 96 <SpecificVersion>False</SpecificVersion> -
branches/CMAES/HeuristicLab.Algorithms.CMAEvolutionStrategy/3.3/Plugin.cs.frame
r9115 r9291 33 33 [PluginDependency("HeuristicLab.Core", "3.3")] 34 34 [PluginDependency("HeuristicLab.Data", "3.3")] 35 [PluginDependency("HeuristicLab.Encodings.RealVectorEncoding", "3.3")] 35 36 [PluginDependency("HeuristicLab.Operators", "3.3")] 36 37 [PluginDependency("HeuristicLab.Optimization", "3.3")] -
branches/CMAES/HeuristicLab.Algorithms.CMAEvolutionStrategy/3.3/Properties/AssemblyInfo.cs.frame
r9129 r9291 1 1 #region License Information 2 2 /* HeuristicLab 3 * Copyright (C) 2002-201 2Heuristic and Evolutionary Algorithms Laboratory (HEAL)3 * Copyright (C) 2002-2013 Heuristic and Evolutionary Algorithms Laboratory (HEAL) 4 4 * 5 5 * This file is part of HeuristicLab. … … 31 31 [assembly: AssemblyCompany("")] 32 32 [assembly: AssemblyProduct("HeuristicLab")] 33 [assembly: AssemblyCopyright("(c) 2002-201 2HEAL")]33 [assembly: AssemblyCopyright("(c) 2002-2013 HEAL")] 34 34 [assembly: AssemblyTrademark("")] 35 35 [assembly: AssemblyCulture("")]
Note: See TracChangeset
for help on using the changeset viewer.