- Timestamp:
- 01/02/18 17:45:15 (7 years ago)
- Location:
- stable
- Files:
-
- 4 edited
Legend:
- Unmodified
- Added
- Removed
-
stable
- Property svn:mergeinfo changed
/trunk/sources merged: 15532,15545,15548,15551,15556
- Property svn:mergeinfo changed
-
stable/HeuristicLab.Algorithms.DataAnalysis
- Property svn:mergeinfo changed
/trunk/sources/HeuristicLab.Algorithms.DataAnalysis merged: 15532,15545,15548,15551,15556
- Property svn:mergeinfo changed
-
stable/HeuristicLab.Algorithms.DataAnalysis/3.4
-
Property
svn:mergeinfo
set to
(toggle deleted branches)
/trunk/sources/HeuristicLab.Algorithms.DataAnalysis/3.4 merged eligible /branches/1721-RandomForestPersistence/HeuristicLab.Algorithms.DataAnalysis/3.4 10321-10322 /branches/Async/HeuristicLab.Algorithms.DataAnalysis/3.4 13329-15286 /branches/Benchmarking/sources/HeuristicLab.Algorithms.DataAnalysis/3.4 6917-7005 /branches/ClassificationModelComparison/HeuristicLab.Algorithms.DataAnalysis/3.4 9070-13099 /branches/CloningRefactoring/HeuristicLab.Algorithms.DataAnalysis/3.4 4656-4721 /branches/DataAnalysis Refactoring/HeuristicLab.Algorithms.DataAnalysis/3.4 5471-5808 /branches/DataAnalysis SolutionEnsembles/HeuristicLab.Algorithms.DataAnalysis/3.4 5815-6180 /branches/DataAnalysis/HeuristicLab.Algorithms.DataAnalysis/3.4 4458-4459,4462,4464 /branches/DataPreprocessing/HeuristicLab.Algorithms.DataAnalysis/3.4 10085-11101 /branches/GP.Grammar.Editor/HeuristicLab.Algorithms.DataAnalysis/3.4 6284-6795 /branches/GP.Symbols (TimeLag, Diff, Integral)/HeuristicLab.Algorithms.DataAnalysis/3.4 5060 /branches/HeuristicLab.DatasetRefactor/sources/HeuristicLab.Algorithms.DataAnalysis/3.4 11570-12508 /branches/HeuristicLab.Problems.Orienteering/HeuristicLab.Algorithms.DataAnalysis/3.4 11130-12721 /branches/HeuristicLab.RegressionSolutionGradientView/HeuristicLab.Algorithms.DataAnalysis/3.4 13819-14091 /branches/HeuristicLab.TimeSeries/HeuristicLab.Algorithms.DataAnalysis/3.4 8116-8789 /branches/LogResidualEvaluator/HeuristicLab.Algorithms.DataAnalysis/3.4 10202-10483 /branches/NET40/sources/HeuristicLab.Algorithms.DataAnalysis/3.4 5138-5162 /branches/ParallelEngine/HeuristicLab.Algorithms.DataAnalysis/3.4 5175-5192 /branches/ProblemInstancesRegressionAndClassification/HeuristicLab.Algorithms.DataAnalysis/3.4 7773-7810 /branches/QAPAlgorithms/HeuristicLab.Algorithms.DataAnalysis/3.4 6350-6627 /branches/Restructure trunk solution/HeuristicLab.Algorithms.DataAnalysis/3.4 6828 /branches/SpectralKernelForGaussianProcesses/HeuristicLab.Algorithms.DataAnalysis/3.4 10204-10479 /branches/SuccessProgressAnalysis/HeuristicLab.Algorithms.DataAnalysis/3.4 5370-5682 /branches/Trunk/HeuristicLab.Algorithms.DataAnalysis/3.4 6829-6865 /branches/VNS/HeuristicLab.Algorithms.DataAnalysis/3.4 5594-5752 /branches/Weighted TSNE/3.4 15451-15531 /branches/histogram/HeuristicLab.Algorithms.DataAnalysis/3.4 5959-6341 /branches/symbreg-factors-2650/HeuristicLab.Algorithms.DataAnalysis/3.4 14232-14825
-
Property
svn:mergeinfo
set to
(toggle deleted branches)
-
stable/HeuristicLab.Algorithms.DataAnalysis/3.4/TSNE/TSNEStatic.cs
r15249 r15571 65 65 [StorableClass] 66 66 public class TSNEStatic<T> { 67 68 67 [StorableClass] 69 68 public sealed class TSNEState : DeepCloneable { … … 170 169 [StorableConstructor] 171 170 public TSNEState(bool deserializing) { } 172 public TSNEState(T[] data, IDistance<T> distance, IRandom random, int newDimensions, double perplexity, double theta, int stopLyingIter, int momSwitchIter, double momentum, double finalMomentum, double eta) { 171 172 public TSNEState(IReadOnlyList<T> data, IDistance<T> distance, IRandom random, int newDimensions, double perplexity, 173 double theta, int stopLyingIter, int momSwitchIter, double momentum, double finalMomentum, double eta, bool randomInit) { 173 174 this.distance = distance; 174 175 this.random = random; … … 183 184 184 185 // initialize 185 noDatapoints = data. Length;186 noDatapoints = data.Count; 186 187 if (noDatapoints - 1 < 3 * perplexity) 187 188 throw new ArgumentException("Perplexity too large for the number of data points!"); … … 193 194 gains = new double[noDatapoints, newDimensions]; 194 195 for (var i = 0; i < noDatapoints; i++) 195 196 196 for (var j = 0; j < newDimensions; j++) 197 gains[i, j] = 1.0; 197 198 198 199 p = null; … … 212 213 var rand = new NormalDistributedRandom(random, 0, 1); 213 214 for (var i = 0; i < noDatapoints; i++) 214 for (var j = 0; j < newDimensions; j++) 215 newData[i, j] = rand.NextDouble() * .0001; 215 for (var j = 0; j < newDimensions; j++) 216 newData[i, j] = rand.NextDouble() * .0001; 217 218 if (!(data[0] is IReadOnlyList<double>) || randomInit) return; 219 for (var i = 0; i < noDatapoints; i++) 220 for (var j = 0; j < newDimensions; j++) { 221 var row = (IReadOnlyList<double>) data[i]; 222 newData[i, j] = row[j % row.Count]; 223 } 216 224 } 217 225 #endregion 218 226 219 227 public double EvaluateError() { 220 return exact ? 221 EvaluateErrorExact(p, newData, noDatapoints, newDimensions) : 222 EvaluateErrorApproximate(rowP, colP, valP, newData, theta); 228 return exact ? EvaluateErrorExact(p, newData, noDatapoints, newDimensions) : EvaluateErrorApproximate(rowP, colP, valP, newData, theta); 223 229 } 224 230 225 231 #region Helpers 226 private static void CalculateApproximateSimilarities( T[]data, IDistance<T> distance, double perplexity, out int[] rowP, out int[] colP, out double[] valP) {232 private static void CalculateApproximateSimilarities(IReadOnlyList<T> data, IDistance<T> distance, double perplexity, out int[] rowP, out int[] colP, out double[] valP) { 227 233 // Compute asymmetric pairwise input similarities 228 ComputeGaussianPerplexity(data, distance, out rowP, out colP, out valP, perplexity, (int) (3 * perplexity));234 ComputeGaussianPerplexity(data, distance, out rowP, out colP, out valP, perplexity, (int) (3 * perplexity)); 229 235 // Symmetrize input similarities 230 236 int[] sRowP, symColP; … … 235 241 valP = sValP; 236 242 var sumP = .0; 237 for (var i = 0; i < rowP[data.Length]; i++) sumP += valP[i]; 238 for (var i = 0; i < rowP[data.Length]; i++) valP[i] /= sumP; 239 } 240 241 private static double[,] CalculateExactSimilarites(T[] data, IDistance<T> distance, double perplexity) { 243 for (var i = 0; i < rowP[data.Count]; i++) sumP += valP[i]; 244 for (var i = 0; i < rowP[data.Count]; i++) valP[i] /= sumP; 245 } 246 private static double[,] CalculateExactSimilarites(IReadOnlyList<T> data, IDistance<T> distance, double perplexity) { 242 247 // Compute similarities 243 var p = new double[data. Length, data.Length];248 var p = new double[data.Count, data.Count]; 244 249 ComputeGaussianPerplexity(data, distance, p, perplexity); 245 250 // Symmetrize input similarities 246 for (var n = 0; n < data. Length; n++) {247 for (var m = n + 1; m < data. Length; m++) {251 for (var n = 0; n < data.Count; n++) { 252 for (var m = n + 1; m < data.Count; m++) { 248 253 p[n, m] += p[m, n]; 249 254 p[m, n] = p[n, m]; … … 251 256 } 252 257 var sumP = .0; 253 for (var i = 0; i < data.Length; i++) for (var j = 0; j < data.Length; j++) sumP += p[i, j]; 254 for (var i = 0; i < data.Length; i++) for (var j = 0; j < data.Length; j++) p[i, j] /= sumP; 258 for (var i = 0; i < data.Count; i++) { 259 for (var j = 0; j < data.Count; j++) { 260 sumP += p[i, j]; 261 } 262 } 263 for (var i = 0; i < data.Count; i++) { 264 for (var j = 0; j < data.Count; j++) { 265 p[i, j] /= sumP; 266 } 267 } 255 268 return p; 256 269 } 257 258 270 private static void ComputeGaussianPerplexity(IReadOnlyList<T> x, IDistance<T> distance, out int[] rowP, out int[] colP, out double[] valP, double perplexity, int k) { 259 271 if (perplexity > k) throw new ArgumentException("Perplexity should be lower than k!"); … … 290 302 291 303 // Iterate until we found a good perplexity 292 var iter = 0; double sumP = 0; 304 var iter = 0; 305 double sumP = 0; 293 306 while (!found && iter < 200) { 294 295 307 // Compute Gaussian kernel row 296 308 for (var m = 0; m < k; m++) curP[m] = Math.Exp(-beta * distances[m + 1]); … … 307 319 if (hdiff < tol && -hdiff < tol) { 308 320 found = true; 309 } else { 321 } 322 else { 310 323 if (hdiff > 0) { 311 324 minBeta = beta; … … 314 327 else 315 328 beta = (beta + maxBeta) / 2.0; 316 } else { 329 } 330 else { 317 331 maxBeta = beta; 318 332 if (minBeta.IsAlmost(double.MinValue) || minBeta.IsAlmost(double.MaxValue)) … … 335 349 } 336 350 } 337 private static void ComputeGaussianPerplexity( T[]x, IDistance<T> distance, double[,] p, double perplexity) {351 private static void ComputeGaussianPerplexity(IReadOnlyList<T> x, IDistance<T> distance, double[,] p, double perplexity) { 338 352 // Compute the distance matrix 339 353 var dd = ComputeDistances(x, distance); 340 354 341 var n = x. Length;355 var n = x.Count; 342 356 // Compute the Gaussian kernel row by row 343 357 for (var i = 0; i < n; i++) { … … 352 366 // Iterate until we found a good perplexity 353 367 var iter = 0; 354 while (!found && iter < 200) { 368 while (!found && iter < 200) { // 200 iterations as in tSNE implementation by van der Maarten 355 369 356 370 // Compute Gaussian kernel row … … 369 383 if (hdiff < tol && -hdiff < tol) { 370 384 found = true; 371 } else { 385 } 386 else { 372 387 if (hdiff > 0) { 373 388 minBeta = beta; … … 376 391 else 377 392 beta = (beta + maxBeta) / 2.0; 378 } else { 393 } 394 else { 379 395 maxBeta = beta; 380 396 if (minBeta.IsAlmost(double.MinValue) || minBeta.IsAlmost(double.MaxValue)) … … 393 409 } 394 410 } 395 396 private static double[][] ComputeDistances(T[] x, IDistance<T> distance) { 397 var res = new double[x.Length][]; 398 for (var r = 0; r < x.Length; r++) { 399 var rowV = new double[x.Length]; 411 private static double[][] ComputeDistances(IReadOnlyList<T> x, IDistance<T> distance) { 412 var res = new double[x.Count][]; 413 for (var r = 0; r < x.Count; r++) { 414 var rowV = new double[x.Count]; 400 415 // all distances must be symmetric 401 416 for (var c = 0; c < r; c++) { … … 403 418 } 404 419 rowV[r] = 0.0; // distance to self is zero for all distances 405 for (var c = r + 1; c < x. Length; c++) {420 for (var c = r + 1; c < x.Count; c++) { 406 421 rowV[c] = distance.Get(x[r], x[c]); 407 422 } … … 411 426 // return x.Select(m => x.Select(n => distance.Get(m, n)).ToArray()).ToArray(); 412 427 } 413 414 428 private static double EvaluateErrorExact(double[,] p, double[,] y, int n, int d) { 415 429 // Compute the squared Euclidean distance matrix … … 425 439 q[n1, m] = 1 / (1 + dd[n1, m]); 426 440 sumQ += q[n1, m]; 427 } else q[n1, m] = double.Epsilon; 441 } 442 else q[n1, m] = double.Epsilon; 428 443 } 429 444 } … … 433 448 var c = .0; 434 449 for (var i = 0; i < n; i++) 435 436 437 450 for (var j = 0; j < n; j++) { 451 c += p[i, j] * Math.Log((p[i, j] + float.Epsilon) / (q[i, j] + float.Epsilon)); 452 } 438 453 return c; 439 454 } 440 441 455 private static double EvaluateErrorApproximate(IReadOnlyList<int> rowP, IReadOnlyList<int> colP, IReadOnlyList<double> valP, double[,] y, double theta) { 442 456 // Get estimate of normalization term … … 463 477 } 464 478 private static void SymmetrizeMatrix(IReadOnlyList<int> rowP, IReadOnlyList<int> colP, IReadOnlyList<double> valP, out int[] symRowP, out int[] symColP, out double[] symValP) { 465 466 479 // Count number of elements and row counts of symmetric matrix 467 480 var n = rowP.Count - 1; … … 469 482 for (var j = 0; j < n; j++) { 470 483 for (var i = rowP[j]; i < rowP[j + 1]; i++) { 471 472 484 // Check whether element (col_P[i], n) is present 473 485 var present = false; … … 497 509 var offset = new int[n]; 498 510 for (var j = 0; j < n; j++) { 499 for (var i = rowP[j]; i < rowP[j + 1]; i++) { 511 for (var i = rowP[j]; i < rowP[j + 1]; i++) { // considering element(n, colP[i]) 500 512 501 513 // Check whether element (col_P[i], n) is present … … 549 561 public static double[,] Run(T[] data, IDistance<T> distance, IRandom random, 550 562 int newDimensions = 2, double perplexity = 25, int iterations = 1000, 551 double theta = 0, 552 int stopLyingIter = 0, int momSwitchIter = 0, double momentum = .5, 563 double theta = 0, int stopLyingIter = 0, int momSwitchIter = 0, double momentum = .5, 553 564 double finalMomentum = .8, double eta = 10.0 554 565 ) { 555 566 var state = CreateState(data, distance, random, newDimensions, perplexity, 556 567 theta, stopLyingIter, momSwitchIter, momentum, finalMomentum, eta); … … 565 576 int newDimensions = 2, double perplexity = 25, double theta = 0, 566 577 int stopLyingIter = 0, int momSwitchIter = 0, double momentum = .5, 567 double finalMomentum = .8, double eta = 10.0 568 569 return new TSNEState(data, distance, random, newDimensions, perplexity, theta, stopLyingIter, momSwitchIter, momentum, finalMomentum, eta );578 double finalMomentum = .8, double eta = 10.0, bool randomInit = true 579 ) { 580 return new TSNEState(data, distance, random, newDimensions, perplexity, theta, stopLyingIter, momSwitchIter, momentum, finalMomentum, eta, randomInit); 570 581 } 571 582 … … 580 591 for (var j = 0; j < state.newDimensions; j++) { 581 592 state.gains[i, j] = Math.Sign(state.dY[i, j]) != Math.Sign(state.uY[i, j]) 582 ? state.gains[i, j] + .2 593 ? state.gains[i, j] + .2 // +0.2 nd *0.8 are used in two separate implementations of tSNE -> seems to be correct 583 594 : state.gains[i, j] * .8; 584 585 595 if (state.gains[i, j] < .01) state.gains[i, j] = .01; 586 596 } 587 597 } 588 589 598 590 599 // Perform gradient update (with momentum and gains) 591 600 for (var i = 0; i < state.noDatapoints; i++) 592 593 601 for (var j = 0; j < state.newDimensions; j++) 602 state.uY[i, j] = state.currentMomentum * state.uY[i, j] - state.eta * state.gains[i, j] * state.dY[i, j]; 594 603 595 604 for (var i = 0; i < state.noDatapoints; i++) 596 597 605 for (var j = 0; j < state.newDimensions; j++) 606 state.newData[i, j] = state.newData[i, j] + state.uY[i, j]; 598 607 599 608 // Make solution zero-mean … … 604 613 if (state.exact) 605 614 for (var i = 0; i < state.noDatapoints; i++) 606 607 615 for (var j = 0; j < state.noDatapoints; j++) 616 state.p[i, j] /= 12.0; 608 617 else 609 618 for (var i = 0; i < state.rowP[state.noDatapoints]; i++) … … 634 643 // Compute final t-SNE gradient 635 644 for (var i = 0; i < n; i++) 636 637 638 645 for (var j = 0; j < d; j++) { 646 dC[i, j] = posF[i, j] - negF[i, j] / sumQ; 647 } 639 648 } 640 649
Note: See TracChangeset
for help on using the changeset viewer.