Skip to content

kriging

Kriging

Bases: surrogates

Kriging surrogate.

Source code in spotpython/build/kriging.py
  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
 208
 209
 210
 211
 212
 213
 214
 215
 216
 217
 218
 219
 220
 221
 222
 223
 224
 225
 226
 227
 228
 229
 230
 231
 232
 233
 234
 235
 236
 237
 238
 239
 240
 241
 242
 243
 244
 245
 246
 247
 248
 249
 250
 251
 252
 253
 254
 255
 256
 257
 258
 259
 260
 261
 262
 263
 264
 265
 266
 267
 268
 269
 270
 271
 272
 273
 274
 275
 276
 277
 278
 279
 280
 281
 282
 283
 284
 285
 286
 287
 288
 289
 290
 291
 292
 293
 294
 295
 296
 297
 298
 299
 300
 301
 302
 303
 304
 305
 306
 307
 308
 309
 310
 311
 312
 313
 314
 315
 316
 317
 318
 319
 320
 321
 322
 323
 324
 325
 326
 327
 328
 329
 330
 331
 332
 333
 334
 335
 336
 337
 338
 339
 340
 341
 342
 343
 344
 345
 346
 347
 348
 349
 350
 351
 352
 353
 354
 355
 356
 357
 358
 359
 360
 361
 362
 363
 364
 365
 366
 367
 368
 369
 370
 371
 372
 373
 374
 375
 376
 377
 378
 379
 380
 381
 382
 383
 384
 385
 386
 387
 388
 389
 390
 391
 392
 393
 394
 395
 396
 397
 398
 399
 400
 401
 402
 403
 404
 405
 406
 407
 408
 409
 410
 411
 412
 413
 414
 415
 416
 417
 418
 419
 420
 421
 422
 423
 424
 425
 426
 427
 428
 429
 430
 431
 432
 433
 434
 435
 436
 437
 438
 439
 440
 441
 442
 443
 444
 445
 446
 447
 448
 449
 450
 451
 452
 453
 454
 455
 456
 457
 458
 459
 460
 461
 462
 463
 464
 465
 466
 467
 468
 469
 470
 471
 472
 473
 474
 475
 476
 477
 478
 479
 480
 481
 482
 483
 484
 485
 486
 487
 488
 489
 490
 491
 492
 493
 494
 495
 496
 497
 498
 499
 500
 501
 502
 503
 504
 505
 506
 507
 508
 509
 510
 511
 512
 513
 514
 515
 516
 517
 518
 519
 520
 521
 522
 523
 524
 525
 526
 527
 528
 529
 530
 531
 532
 533
 534
 535
 536
 537
 538
 539
 540
 541
 542
 543
 544
 545
 546
 547
 548
 549
 550
 551
 552
 553
 554
 555
 556
 557
 558
 559
 560
 561
 562
 563
 564
 565
 566
 567
 568
 569
 570
 571
 572
 573
 574
 575
 576
 577
 578
 579
 580
 581
 582
 583
 584
 585
 586
 587
 588
 589
 590
 591
 592
 593
 594
 595
 596
 597
 598
 599
 600
 601
 602
 603
 604
 605
 606
 607
 608
 609
 610
 611
 612
 613
 614
 615
 616
 617
 618
 619
 620
 621
 622
 623
 624
 625
 626
 627
 628
 629
 630
 631
 632
 633
 634
 635
 636
 637
 638
 639
 640
 641
 642
 643
 644
 645
 646
 647
 648
 649
 650
 651
 652
 653
 654
 655
 656
 657
 658
 659
 660
 661
 662
 663
 664
 665
 666
 667
 668
 669
 670
 671
 672
 673
 674
 675
 676
 677
 678
 679
 680
 681
 682
 683
 684
 685
 686
 687
 688
 689
 690
 691
 692
 693
 694
 695
 696
 697
 698
 699
 700
 701
 702
 703
 704
 705
 706
 707
 708
 709
 710
 711
 712
 713
 714
 715
 716
 717
 718
 719
 720
 721
 722
 723
 724
 725
 726
 727
 728
 729
 730
 731
 732
 733
 734
 735
 736
 737
 738
 739
 740
 741
 742
 743
 744
 745
 746
 747
 748
 749
 750
 751
 752
 753
 754
 755
 756
 757
 758
 759
 760
 761
 762
 763
 764
 765
 766
 767
 768
 769
 770
 771
 772
 773
 774
 775
 776
 777
 778
 779
 780
 781
 782
 783
 784
 785
 786
 787
 788
 789
 790
 791
 792
 793
 794
 795
 796
 797
 798
 799
 800
 801
 802
 803
 804
 805
 806
 807
 808
 809
 810
 811
 812
 813
 814
 815
 816
 817
 818
 819
 820
 821
 822
 823
 824
 825
 826
 827
 828
 829
 830
 831
 832
 833
 834
 835
 836
 837
 838
 839
 840
 841
 842
 843
 844
 845
 846
 847
 848
 849
 850
 851
 852
 853
 854
 855
 856
 857
 858
 859
 860
 861
 862
 863
 864
 865
 866
 867
 868
 869
 870
 871
 872
 873
 874
 875
 876
 877
 878
 879
 880
 881
 882
 883
 884
 885
 886
 887
 888
 889
 890
 891
 892
 893
 894
 895
 896
 897
 898
 899
 900
 901
 902
 903
 904
 905
 906
 907
 908
 909
 910
 911
 912
 913
 914
 915
 916
 917
 918
 919
 920
 921
 922
 923
 924
 925
 926
 927
 928
 929
 930
 931
 932
 933
 934
 935
 936
 937
 938
 939
 940
 941
 942
 943
 944
 945
 946
 947
 948
 949
 950
 951
 952
 953
 954
 955
 956
 957
 958
 959
 960
 961
 962
 963
 964
 965
 966
 967
 968
 969
 970
 971
 972
 973
 974
 975
 976
 977
 978
 979
 980
 981
 982
 983
 984
 985
 986
 987
 988
 989
 990
 991
 992
 993
 994
 995
 996
 997
 998
 999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
class Kriging(surrogates):
    """Kriging surrogate.
    """
    def __init__(
            self: object,
            noise: bool = False,
            var_type: List[str] = ["num"],
            name: str = "kriging",
            seed: int = 124,
            model_optimizer=None,
            model_fun_evals: Optional[int] = None,
            min_theta: float = -3.0,
            max_theta: float = 2.0,
            n_theta: int = 1,
            theta_init_zero: bool = True,
            p_val: float = 2.0,
            n_p: int = 1,
            optim_p: bool = False,
            min_Lambda: float = 1e-9,
            max_Lambda: float = 1.,
            log_level: int = 50,
            spot_writer=None,
            counter=None,
            metric_factorial="canberra",
            **kwargs
    ):
        """
        Initialize the Kriging surrogate.

        Args:
            noise (bool): Use regression instead of interpolation kriging. Defaults to False.
            var_type (List[str]):
                Variable type. Can be either "num" (numerical) or "factor" (factor).
                Defaults to ["num"].
            name (str):
                Surrogate name. Defaults to "kriging".
            seed (int):
                Random seed. Defaults to 124.
            model_optimizer (Optional[object]):
                Optimizer on the surrogate. If None, differential_evolution is selected.
            model_fun_evals (Optional[int]):
                Number of iterations used by the optimizer on the surrogate.
            min_theta (float):
                Min log10 theta value. Defaults to -3.
            max_theta (float):
                Max log10 theta value. Defaults to 2.
            n_theta (int):
                Number of theta values. Defaults to 1.
            theta_init_zero (bool):
                Initialize theta with zero. Defaults to True.
            p_val (float):
                p value. Used as an initial value if optim_p = True. Otherwise as a constant. Defaults to 2.
            n_p (int):
                Number of p values. Defaults to 1.
            optim_p (bool):
                Determines whether p should be optimized. Deafults to False.
            min_Lambda (float):
                Min Lambda value. Defaults to 1e-9.
            max_Lambda (float):
                Max Lambda value. Defaults to 1.
            log_level (int):
                Logging level, e.g., 20 is "INFO". Defaults to 50 ("CRITICAL").
            spot_writer (Optional[object]):
                Spot writer. Defaults to None.
            counter (Optional[int]):
                Counter. Defaults to None.
            metric_factorial (str):
                Metric for factorial. Defaults to "canberra". Can be "euclidean",
                "cityblock", seuclidean", "sqeuclidean", "cosine",
                "correlation", "hamming", "jaccard", "jensenshannon",
                "chebyshev", "canberra", "braycurtis", "mahalanobis", "matching".

        Examples:
            >>> from spotpython.build.kriging import Kriging
                import numpy as np
                import matplotlib.pyplot as plt
                from numpy import linspace, arange
                rng = np.random.RandomState(1)
                X = linspace(start=0, stop=10, num=1_000).reshape(-1, 1)
                y = np.squeeze(X * np.sin(X))
                training_indices = rng.choice(arange(y.size), size=6, replace=False)
                X_train, y_train = X[training_indices], y[training_indices]
                S = Kriging(name='kriging', seed=124)
                S.fit(X_train, y_train)
                mean_prediction, std_prediction, s_ei = S.predict(X, return_val="all")
                plt.plot(X, y, label=r"$f(x)$", linestyle="dotted")
                plt.scatter(X_train, y_train, label="Observations")
                plt.plot(X, mean_prediction, label="Mean prediction")
                plt.fill_between(
                    X.ravel(),
                    mean_prediction - 1.96 * std_prediction,
                    mean_prediction + 1.96 * std_prediction,
                    alpha=0.5,
                    label=r"95% confidence interval",
                    )
                plt.legend()
                plt.xlabel("$x$")
                plt.ylabel("$f(x)$")
                _ = plt.title("Gaussian process regression on noise-free dataset")
                plt.show()

        References:
            https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.distance.pdist.html
            [[1](https://scikit-learn.org/stable/auto_examples/gaussian_process/plot_gpr_noisy_targets.html)]
            scikit-learn: Gaussian Processes regression: basic introductory example

        """
        super().__init__(name, seed, log_level)

        self.noise = noise
        self.var_type = var_type
        self.name = name
        self.seed = seed
        self.log_level = log_level
        self.spot_writer = spot_writer
        self.counter = counter
        self.metric_factorial = metric_factorial

        self.sigma = 0
        self.eps = sqrt(spacing(1))
        self.min_theta = min_theta
        self.max_theta = max_theta
        self.min_p = 1
        self.max_p = 2
        self.min_Lambda = min_Lambda
        self.max_Lambda = max_Lambda
        self.n_theta = n_theta
        self.p_val = p_val
        self.n_p = n_p
        self.optim_p = optim_p
        self.theta_init_zero = theta_init_zero
        # Psi matrix condition:
        self.cnd_Psi = 0
        self.inf_Psi = False

        self.model_optimizer = model_optimizer
        if self.model_optimizer is None:
            self.model_optimizer = differential_evolution
        self.model_fun_evals = model_fun_evals
        # differential evolution uses maxiter = 1000
        # and sets the number of function evaluations to
        # (maxiter + 1) * popsize * N, which results in
        # 1000 * 15 * k, because the default popsize is 15 and
        # N is the number of parameters. This seems to be quite large:
        # for k=2 these are 30 000 iterations. Therefore we set this value to
        # 100
        if self.model_fun_evals is None:
            self.model_fun_evals = 100

        # Logging information
        self.log["negLnLike"] = []
        self.log["theta"] = []
        self.log["p"] = []
        self.log["Lambda"] = []
        # Logger
        logger.setLevel(self.log_level)
        logger.info(f"Starting the logger at level {self.log_level} for module {__name__}:")

    def exp_imp(self, y0: float, s0: float) -> float:
        """
        Calculates the expected improvement for a given function value and error in coded units.

        Args:
            self (object): The Kriging object.
            y0 (float): The function value in coded units.
            s0 (float): The error value.

        Returns:
            float: The expected improvement value.

        Examples:
            >>> from spotpython.build.kriging import Kriging
                S = Kriging(name='kriging', seed=124)
                S.aggregated_mean_y = [0.0, 0.0, 0.0, 0.0, 0.0]
                S.exp_imp(1.0, 0.0)
                0.0
            >>> from spotpython.build.kriging import Kriging
                S = Kriging(name='kriging', seed=124)
                S.aggregated_mean_y = [0.0, 0.0, 0.0, 0.0, 0.0]
                # assert S.exp_imp(0.0, 1.0) == 1/np.sqrt(2*np.pi)
                # which is approx. 0.3989422804014327
                S.exp_imp(0.0, 1.0)
                0.3989422804014327
        """
        # We do not use the min y values, but the aggragated mean values
        # y_min = min(self.nat_y)
        y_min = min(self.aggregated_mean_y)
        if s0 <= 0.0:
            EI = 0.0
        elif s0 > 0.0:
            EI_one = (y_min - y0) * (
                    0.5 + 0.5 * erf((1.0 / sqrt(2.0)) * ((y_min - y0) / s0))
            )
            EI_two = (s0 * (1.0 / sqrt(2.0 * pi))) * (
                exp(-(1.0 / 2.0) * ((y_min - y0) ** 2.0 / s0 ** 2.0))
            )
            EI = EI_one + EI_two
        return EI

    def set_de_bounds(self) -> None:
        """
        Determine search bounds for model_optimizer, e.g., differential evolution.

        This method sets the attribute `de_bounds` of the object to a list of lists,
        where each inner list represents the lower and upper bounds for a parameter
        being optimized. The number of inner lists is determined by the number of
        parameters being optimized (`n_theta` and `n_p`), as well as whether noise is
        being considered (`noise`).

        Args:
            self (object): The Kriging object.

        Examples:
            >>> from spotpython.build.kriging import Kriging
                S = Kriging(name='kriging', seed=124)
                S.set_de_bounds()
                print(S.de_bounds)

        Returns:
            None
        """
        logger.debug("In set_de_bounds(): self.min_theta: %s", self.min_theta)
        logger.debug("In set_de_bounds(): self.max_theta: %s", self.max_theta)
        logger.debug("In set_de_bounds(): self.n_theta: %s", self.n_theta)
        logger.debug("In set_de_bounds(): self.optim_p: %s", self.optim_p)
        logger.debug("In set_de_bounds(): self.min_p: %s", self.min_p)
        logger.debug("In set_de_bounds(): self.max_p: %s", self.max_p)
        logger.debug("In set_de_bounds(): self.n_p: %s", self.n_p)
        logger.debug("In set_de_bounds(): self.noise: %s", self.noise)
        logger.debug("In set_de_bounds(): self.min_Lambda: %s", self.min_Lambda)
        logger.debug("In set_de_bounds(): self.max_Lambda: %s", self.max_Lambda)

        de_bounds = [[self.min_theta, self.max_theta] for _ in range(self.n_theta)]
        if self.optim_p:
            de_bounds += [[self.min_p, self.max_p] for _ in range(self.n_p)]
            if self.noise:
                de_bounds.append([self.min_Lambda, self.max_Lambda])
        else:
            if self.noise:
                de_bounds.append([self.min_Lambda, self.max_Lambda])
        self.de_bounds = de_bounds
        logger.debug("In set_de_bounds(): self.de_bounds: %s", self.de_bounds)

    def extract_from_bounds(self, new_theta_p_Lambda: np.ndarray) -> None:
        """
        Extract `theta`, `p`, and `Lambda` from bounds. The kriging object stores
        `theta` as an array,  `p` as an array, and `Lambda` as a float.

        Args:
            self (object): The Kriging object.
            new_theta_p_Lambda (np.ndarray):
                1d-array with theta, p, and Lambda values. Order is important.

        Examples:
            >>> import numpy as np
                from spotpython.build.kriging import Kriging
                n=2
                p=4
                S = Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False)
                S.extract_from_bounds(np.array([1, 2, 3]))
                print(S.theta)
                print(S.p)
                [1 2]
                [3]

        Returns:
            None
        """
        logger.debug("In extract_from_bounds(): new_theta_p_Lambda: %s", new_theta_p_Lambda)
        self.theta = new_theta_p_Lambda[:self.n_theta]
        logger.debug("In extract_from_bounds(): self.n_theta: %s", self.n_theta)
        if self.optim_p:
            self.p = new_theta_p_Lambda[self.n_theta:self.n_theta + self.n_p]
            logger.debug("In extract_from_bounds(): self.p: %s", self.p)
            if self.noise:
                self.Lambda = new_theta_p_Lambda[self.n_theta + self.n_p]
                logger.debug("In extract_from_bounds(): self.Lambda: %s", self.Lambda)
        else:
            if self.noise:
                self.Lambda = new_theta_p_Lambda[self.n_theta]
                logger.debug("In extract_from_bounds(): self.Lambda: %s", self.Lambda)

    def optimize_model(self) -> Union[List[float], Tuple[float]]:
        """
        Optimize the model using the specified model_optimizer.

        This method uses the specified model_optimizer to optimize the
        likelihood function (`fun_likelihood`) with respect to the model parameters.
        The optimization is performed within the bounds specified by the attribute
        `de_bounds`.
        The result of the optimization is returned as a list or tuple of optimized parameter values.

        Args:
            self (object): The Kriging object.

        Examples:
            >>> from spotpython.build.kriging import Kriging
                import numpy as np
                nat_X = np.array([[1, 2], [3, 4]])
                nat_y = np.array([1, 2])
                n=2
                p=2
                S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=True)
                S.initialize_variables(nat_X, nat_y)
                S.set_variable_types()
                S.set_theta_values()
                S.initialize_matrices()
                S.set_de_bounds()
                new_theta_p_Lambda = S.optimize_model()
                print(new_theta_p_Lambda)

        Returns:
            result["x"] (Union[List[float], Tuple[float]]):
                A list or tuple of optimized parameter values.
        """
        logger.debug("In optimize_model(): self.de_bounds passed to optimizer: %s", self.de_bounds)
        if self.model_optimizer.__name__ == 'dual_annealing':
            result = self.model_optimizer(func=self.fun_likelihood,
                                          bounds=self.de_bounds)
        elif self.model_optimizer.__name__ == 'differential_evolution':
            result = self.model_optimizer(func=self.fun_likelihood,
                                          bounds=self.de_bounds,
                                          maxiter=self.model_fun_evals,
                                          seed=self.seed)
        elif self.model_optimizer.__name__ == 'direct':
            result = self.model_optimizer(func=self.fun_likelihood,
                                          bounds=self.de_bounds,
                                          # maxfun=self.model_fun_evals,
                                          eps=1e-2)
        elif self.model_optimizer.__name__ == 'shgo':
            result = self.model_optimizer(func=self.fun_likelihood,
                                          bounds=self.de_bounds)
        elif self.model_optimizer.__name__ == 'basinhopping':
            result = self.model_optimizer(func=self.fun_likelihood,
                                          x0=mean(self.de_bounds, axis=1))
        else:
            result = self.model_optimizer(func=self.fun_likelihood, bounds=self.de_bounds)
        logger.debug("In optimize_model(): result: %s", result)
        logger.debug('In optimize_model(): returned result["x"]: %s', result["x"])
        return result["x"]

    def update_log(self) -> None:
        """
        Update the log with the current values of negLnLike, theta, p, and Lambda.

        This method appends the current values of negLnLike, theta, p (if optim_p is True),
        and Lambda (if noise is True)
        to their respective lists in the log dictionary.
        It also updates the log_length attribute with the current length
        of the negLnLike list in the log.

        If spot_writer is not None, this method also writes the current values of
        negLnLike, theta, p (if optim_p is True),
        and Lambda (if noise is True) to the spot_writer object.

        Args:
            self (object): The Kriging object.

        Returns:
            None

        Examples:
            >>> from spotpython.build.kriging import Kriging
                import numpy as np
                nat_X = np.array([[1, 2], [3, 4]])
                nat_y = np.array([1, 2])
                n=2
                p=2
                S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=True)
                S.initialize_variables(nat_X, nat_y)
                S.set_variable_types()
                S.set_theta_values()
                S.initialize_matrices()
                S.set_de_bounds()
                new_theta_p_Lambda = S.optimize_model()
                S.update_log()
                print(S.log)
                {'negLnLike': array([-1.38629436]),
                 'theta': array([-1.14525993,  1.6123372 ]),
                  'p': array([1.84444406, 1.74590865]),
                  'Lambda': array([0.44268472])}

        """
        self.log["negLnLike"] = append(self.log["negLnLike"], self.negLnLike)
        self.log["theta"] = append(self.log["theta"], self.theta)
        if self.optim_p:
            self.log["p"] = append(self.log["p"], self.p)
        if self.noise:
            self.log["Lambda"] = append(self.log["Lambda"], self.Lambda)
        # get the length of the log
        self.log_length = len(self.log["negLnLike"])
        if self.spot_writer is not None:
            negLnLike = self.negLnLike.copy()
            self.spot_writer.add_scalar("spot_negLnLike", negLnLike, self.counter+self.log_length)
            # add the self.n_theta theta values to the writer with one key "theta",
            # i.e, the same key for all theta values
            theta = self.theta.copy()
            self.spot_writer.add_scalars("spot_theta", {f"theta_{i}": theta[i] for i in range(self.n_theta)},
                                         self.counter+self.log_length)
            if self.noise:
                Lambda = self.Lambda.copy()
                self.spot_writer.add_scalar("spot_Lambda", Lambda, self.counter+self.log_length)
            if self.optim_p:
                p = self.p.copy()
                self.spot_writer.add_scalars("spot_p", {f"p_{i}": p[i] for i in range(self.n_p)}, self.counter+self.log_length)
            self.spot_writer.flush()

    def fit(self, nat_X: np.ndarray, nat_y: np.ndarray) -> object:
        """
        Fits the hyperparameters (`theta`, `p`, `Lambda`) of the Kriging model.

        The function computes the following internal values:
        1. `theta`, `p`, and `Lambda` values via optimization of the function `fun_likelihood()`.
        2. Correlation matrix `Psi` via `rebuildPsi()`.

        Args:
            self (object): The Kriging object.
            nat_X (np.ndarray): Sample points.
            nat_y (np.ndarray): Function values.

        Returns:
            object: Fitted estimator.

        Attributes:
            theta (np.ndarray): Kriging theta values. Shape (k,).
            p (np.ndarray): Kriging p values. Shape (k,).
            LnDetPsi (np.float64): Determinant Psi matrix.
            Psi (np.matrix): Correlation matrix Psi. Shape (n,n).
            psi (np.ndarray): psi vector. Shape (n,).
            one (np.ndarray): vector of ones. Shape (n,).
            mu (np.float64): Kriging expected mean value mu.
            U (np.matrix): Kriging U matrix, Cholesky decomposition. Shape (n,n).
            SigmaSqr (np.float64): Sigma squared value.
            Lambda (float): lambda noise value.

        Examples:
            >>> from spotpython.build.kriging import Kriging
                import numpy as np
                nat_X = np.array([[1, 0], [1, 0]])
                nat_y = np.array([1, 2])
                S = Kriging()
                S.fit(nat_X, nat_y)
                print(S.Psi)
                [[1.00000001 1.        ]
                [1.         1.00000001]]

        """
        logger.debug("In fit(): nat_X: %s", nat_X)
        logger.debug("In fit(): nat_y: %s", nat_y)
        self.initialize_variables(nat_X, nat_y)
        self.set_variable_types()
        self.set_theta_values()
        self.initialize_matrices()
        # build_Psi() and build_U() are called in fun_likelihood
        self.set_de_bounds()
        # Finally, set new theta and p values and update the surrogate again
        # for new_theta_p_Lambda in de_results["x"]:
        new_theta_p_Lambda = self.optimize_model()
        self.extract_from_bounds(new_theta_p_Lambda)
        self.build_Psi()
        self.build_U()
        # TODO: check if the following line is necessary!
        self.likelihood()
        self.update_log()

    def initialize_variables(self, nat_X: np.ndarray, nat_y: np.ndarray) -> None:
        """
        Initialize variables for the class instance.

        This method takes in the independent and dependent variable data as input
        and initializes the class instance variables.
        It creates deep copies of the input data and stores them in the
        instance variables `nat_X` and `nat_y`.
        It also calculates the number of observations `n` and
        the number of independent variables `k` from the shape of `nat_X`.
        Finally, it creates empty arrays with the same shape as `nat_X`
        and `nat_y` and stores them in the instance variables `cod_X` and `cod_y`.

        Args:
            self (object): The Kriging object.
            nat_X (np.ndarray): The independent variable data.
            nat_y (np.ndarray): The dependent variable data.

        Returns:
            None

        Examples:
            >>> from spotpython.build.kriging import Kriging
                import numpy as np
                nat_X = np.array([[1, 2], [3, 4]])
                nat_y = np.array([1, 2])
                S = Kriging()
                S.initialize_variables(nat_X, nat_y)
                print(f"S.nat_X: {S.nat_X}")
                print(f"S.nat_y: {S.nat_y}")
                S.nat_X: [[1 2]
                          [3 4]]
                S.nat_y: [1 2]

        """
        self.nat_X = copy.deepcopy(nat_X)
        self.nat_y = copy.deepcopy(nat_y)
        self.n = self.nat_X.shape[0]
        self.k = self.nat_X.shape[1]

        self.min_X = min(self.nat_X, axis=0)
        self.max_X = max(self.nat_X, axis=0)

        Z = aggregate_mean_var(X=self.nat_X, y=self.nat_y)
        # aggregated y values:
        mu = Z[1]
        self.aggregated_mean_y = np.copy(mu)
        logger.debug("In initialize_variables(): self.nat_X: %s", self.nat_X)
        logger.debug("In initialize_variables(): self.nat_y: %s", self.nat_y)
        logger.debug("In initialize_variables(): self.aggregated_mean_y: %s", self.aggregated_mean_y)
        logger.debug("In initialize_variables(): self.min_X: %s", self.min_X)
        logger.debug("In initialize_variables(): self.max_X: %s", self.max_X)
        logger.debug("In initialize_variables(): self.n: %s", self.n)
        logger.debug("In initialize_variables(): self.k: %s", self.k)

    def set_variable_types(self) -> None:
        """
        Set the variable types for the class instance.

        This method sets the variable types for the class instance based
        on the `var_type` attribute. If the length of `var_type` is less
        than `k`, all variable types are forced to 'num' and a warning is logged.
        The method then creates Boolean masks for each variable
        type ('num', 'factor', 'int', 'ordered') using numpy arrays, e.g.,
        `num_mask = array([ True,  True])` if two numerical variables are present.

        Args:
            self (object): The Kriging object.

        Examples:
            >>> from spotpython.build.kriging import Kriging
                nat_X = np.array([[1, 2], [3, 4]])
                nat_y = np.array([1, 2])
                n=2
                p=2
                S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=True)
                S.initialize_variables(nat_X, nat_y)
                S.set_variable_types()
                assert S.var_type == ['num', 'num']
                assert S.var_type == ['num', 'num']
                assert S.num_mask.all() == True
                assert S.factor_mask.all() == False
                assert S.int_mask.all() == False
                assert S.ordered_mask.all() == True

        Returns:
            None
        """
        logger.debug("In set_variable_types(): self.k: %s", self.k)
        logger.debug("In set_variable_types(): self.var_type: %s", self.var_type)
        # assume all variable types are "num" if "num" is
        # specified once:
        if len(self.var_type) < self.k:
            self.var_type = self.var_type * self.k
            logger.warning("In set_variable_types(): All variable types forced to 'num'.")
            logger.debug("In set_variable_types(): self.var_type: %s", self.var_type)
        self.num_mask = np.array(list(map(lambda x: x == "num", self.var_type)))
        self.factor_mask = np.array(list(map(lambda x: x == "factor", self.var_type)))
        self.int_mask = np.array(list(map(lambda x: x == "int", self.var_type)))
        self.ordered_mask = np.array(list(map(lambda x: x == "int" or x == "num" or x == "float", self.var_type)))
        logger.debug("In set_variable_types(): self.num_mask: %s", self.num_mask)
        logger.debug("In set_variable_types(): self.factor_mask: %s", self.factor_mask)
        logger.debug("In set_variable_types(): self.int_mask: %s", self.int_mask)
        logger.debug("In set_variable_types(): self.ordered_mask: %s", self.ordered_mask)

    def set_theta_values(self) -> None:
        """
        Set the theta values for the class instance.

        This method sets the theta values for the class instance based
        on the `n_theta` and `k` attributes. If `n_theta` is greater than
        `k`, `n_theta` is set to `k` and a warning is logged.
        The method then initializes the `theta` attribute as a list
        of zeros with length `n_theta`.
        The `x0_theta` attribute is also initialized as a list of ones
        with length `n_theta`, multiplied by `n / (100 * k)`.

        Args:
            self (object): The Kriging object.
        Returns:
            None

        Examples:
            >>> from spotpython.build.kriging import Kriging
                import numpy as np
                from numpy import array
                nat_X = np.array([[1, 2], [3, 4]])
                nat_y = np.array([1, 2])
                n=2
                p=2
                S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=True)
                S.initialize_variables(nat_X, nat_y)
                S.set_variable_types()
                S.set_theta_values()
                assert S.theta.all() == array([0., 0.]).all()
        """
        logger.debug("In set_theta_values(): self.k: %s", self.k)
        logger.debug("In set_theta_values(): self.n_theta: %s", self.n_theta)
        if ((self.n_theta > 1) or (self.n_theta > self.k)) and (self.n_theta != self.k):
            self.n_theta = self.k
            logger.warning("Too few theta values or more theta values than dimensions. `n_theta` set to `k`.")
            logger.debug("In set_theta_values(): self.n_theta: %s", self.n_theta)
        if self.theta_init_zero:
            self.theta: List[float] = zeros(self.n_theta)
            logger.debug("In set_theta_values(): self.theta: %s", self.theta)
        else:
            logger.debug("In set_theta_values(): self.n: %s", self.n)
            self.theta: List[float] = ones((self.n_theta,)) * self.n / (100 * self.k)
            logger.debug("In set_theta_values(): self.theta: %s", self.theta)

    def initialize_matrices(self) -> None:
        """
        Initialize the matrices for the class instance.

        This method initializes several matrices and attributes for the class instance.
        The `p` attribute is initialized as a list of ones with length `n_p`, multiplied by 2.0.
        The `pen_val` attribute is initialized as the natural logarithm of the
        variance of `nat_y`, multiplied by `n`, plus 1e4.
        The `negLnLike`, `LnDetPsi`, `mu`, `U`, `SigmaSqr`, and `Lambda` attributes are all set to None.
        The `gen` attribute is initialized using the `spacefilling` function with arguments `k` and `seed`.
        The `Psi` attribute is initialized as a zero matrix with shape `(n, n)` and dtype `float64`.
        The `psi` attribute is initialized as a zero matrix with shape `(n, 1)`.
        The `one` attribute is initialized as a list of ones with length `n`.

        Args:
            self (object): The Kriging object.

        Examples:
            >>> from spotpython.build.kriging import Kriging
                import numpy as np
                from numpy import log, var
                nat_X = np.array([[1, 2], [3, 4], [5, 6]])
                nat_y = np.array([1, 2, 3])
                n=3
                p=1
                S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=True)
                S.initialize_variables(nat_X, nat_y)
                S.set_variable_types()
                S.set_theta_values()
                S.initialize_matrices()
                # if var(self.nat_y) is > 0, then self.pen_val = self.n * log(var(self.nat_y)) + 1e4
                # else self.pen_val = self.n * var(self.nat_y) + 1e4
                assert S.pen_val == nat_X.shape[0] * log(var(S.nat_y)) + 1e4
                assert S.Psi.shape == (n, n)

        Returns:
            None
        """
        logger.debug("In initialize_matrices(): self.n_p: %s", self.n_p)
        self.p = ones(self.n_p) * self.p_val
        logger.debug("In initialize_matrices(): self.p: %s", self.p)
        # if var(self.nat_y) is > 0, then self.pen_val = self.n * log(var(self.nat_y)) + 1e4
        # else self.pen_val = self.n * var(self.nat_y) + 1e4
        logger.debug("In initialize_matrices(): var(self.nat_y): %s", var(self.nat_y))
        logger.debug("In initialize_matrices(): self.n: %s", self.n)
        if var(self.nat_y) > 0:
            self.pen_val = self.n * log(var(self.nat_y)) + 1e4
        else:
            self.pen_val = self.n * var(self.nat_y) + 1e4
        logger.debug("In initialize_matrices(): self.pen_val: %s", self.pen_val)
        self.negLnLike = None
        logger.debug("In initialize_matrices(): self.k: %s", self.k)
        logger.debug("In initialize_matrices(): self.seed: %s", self.seed)
        self.gen = spacefilling(k=self.k, seed=self.seed)
        logger.debug("In initialize_matrices(): self.gen: %s", self.gen)
        self.LnDetPsi = None
        self.Psi = zeros((self.n, self.n), dtype=float64)
        logger.debug("In initialize_matrices(): self.Psi: %s", self.Psi)
        self.psi = zeros((self.n, 1))
        logger.debug("In initialize_matrices(): self.psi: %s", self.psi)
        self.one = ones(self.n)
        logger.debug("In initialize_matrices(): self.one: %s", self.one)
        self.mu = None
        self.U = None
        self.SigmaSqr = None
        self.Lambda = None

    def fun_likelihood(self, new_theta_p_Lambda: np.ndarray) -> float:
        """
        Compute log likelihood for a set of hyperparameters (theta, p, Lambda).

        This method computes the log likelihood for a set of hyperparameters
        (theta, p, Lambda) by performing the following steps:
        1. Extracts the hyperparameters from the input array using `extract_from_bounds()`.
        2. Checks if any element in `10^theta` is equal to 0. If so, logs a warning and
        returns the penalty value (`pen_val`).
        3. Builds the `Psi` matrix using `build_Psi()`.
        4. Checks if `Psi` is ill-conditioned or infinite. If so, logs a warning and returns
        the penalty value (`pen_val`).
        5. Builds the `U` matrix using `build_U()`. If an exception occurs, logs an error and
        returns the penalty value (`pen_val`).
        6. Computes the negative log likelihood using `likelihood()`.
        7. Returns the computed negative log likelihood (`negLnLike`).

        Args:
            self (object): The Kriging object.
            new_theta_p_Lambda (np.ndarray):
                An array containing the `theta`, `p`, and `Lambda` values.

        Returns:
            float:
                The negative log likelihood of the surface at the specified hyperparameters.

        Attributes:
            theta (np.ndarray): Kriging theta values. Shape (k,).
            p (np.ndarray): Kriging p values. Shape (k,).
            Lambda (float): lambda noise value.
            Psi (np.matrix): Correlation matrix Psi. Shape (n,n).
            U (np.matrix): Kriging U matrix, Cholesky decomposition. Shape (n,n).
            negLnLike (float): Negative log likelihood of the surface at the specified hyperparameters.
            pen_val (float): Penalty value.

        Examples:
            >>> from spotpython.build.kriging import Kriging
                import numpy as np
                nat_X = np.array([[0], [1]])
                nat_y = np.array([0, 1])
                n=1
                p=1
                S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False)
                S.initialize_variables(nat_X, nat_y)
                S.set_variable_types()
                print(S.cod_X)
                print(S.cod_y)
                S.set_theta_values()
                print(f"S.theta: {S.theta}")
                S.initialize_matrices()
                S.set_de_bounds()
                new_theta_p_Lambda = S.optimize_model()
                S.extract_from_bounds(new_theta_p_Lambda)
                print(f"S.theta: {S.theta}")
                S.build_Psi()
                print(f"S.Psi: {S.Psi}")
                S.build_U()
                print(f"S.U:{S.U}")
                S.likelihood()
                S.negLnLike

        """
        self.extract_from_bounds(new_theta_p_Lambda)
        if self.__is_any__(power(10.0, self.theta), 0):
            logger.warning("Failure in fun_likelihood: 10^theta == 0. Setting negLnLike to %s", self.pen_val)
            return self.pen_val
        self.build_Psi()
        if (self.inf_Psi or self.cnd_Psi > 1e9):
            logger.warning("Failure in fun_likelihood: Psi is ill conditioned: %s", self.cnd_Psi)
            logger.warning("Setting negLnLike to: %s", self.pen_val)
            return self.pen_val

        try:
            self.build_U()
        except Exception as error:
            penalty_value = self.pen_val
            print("Error in fun_likelihood(). Call to build_U() failed.")
            print("error=%s, type(error)=%s" % (error, type(error)))
            print("Setting negLnLike to %.2f." % self.pen_val)
            return penalty_value
        self.likelihood()
        return self.negLnLike

    def __is_any__(self, x: Union[np.ndarray, Any], v: Any) -> bool:
        """
        Check if any element in `x` is equal to `v`.
        This method checks if any element in the input array `x` is equal to the value `v`.
        If `x` is not an instance of `ndarray`, it is first converted to a numpy array using
        the `array()` function.

        Args:
            self (object): The Kriging object.
            x (np.ndarray or array-like):
                The input array to check for the presence of value `v`.
            v (scalar):
                The value to check for in the input array `x`.

        Returns:
            bool:
                True if any element in `x` is equal to `v`, False otherwise.

        Examples:
            >>> from spotpython.build.kriging import Kriging
                from numpy import power
                import numpy as np
                nat_X = np.array([[0], [1]])
                nat_y = np.array([0, 1])
                n=1
                p=1
                S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False)
                S.initialize_variables(nat_X, nat_y)
                S.set_variable_types()
                S.set_theta_values()
                print(f"S.theta: {S.theta}")
                print(S.__is_any__(power(10.0, S.theta), 0))
                print(S.__is_any__(S.theta, 0))
                S.theta: [0.]
                False
                True

        """
        if not isinstance(x, ndarray):
            x = array([x])
        return any(x == v)

    def build_Psi(self) -> None:
        """
        Constructs a new (n x n) correlation matrix Psi to reflect new data
        or a change in hyperparameters.

        This method uses `theta`, `p`, and coded `X` values to construct the
        correlation matrix as described in [Forr08a, p.57].

        Args:
            self (object): The Kriging object.

        Returns:
            None

        Raises:
            LinAlgError: If building Psi fails.

        Attributes:
            Psi (np.matrix): Correlation matrix Psi. Shape (n,n).
            cnd_Psi (float): Condition number of Psi.
            inf_Psi (bool): True if Psi is infinite, False otherwise.

        Examples:
            >>> from spotpython.build.kriging import Kriging
                import numpy as np
                nat_X = np.array([[0], [1]])
                nat_y = np.array([0, 1])
                n=1
                p=1
                S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False)
                S.initialize_variables(nat_X, nat_y)
                S.set_variable_types()
                print(S.nat_X)
                print(S.nat_y)
                S.set_theta_values()
                print(f"S.theta: {S.theta}")
                S.initialize_matrices()
                S.set_de_bounds()
                new_theta_p_Lambda = S.optimize_model()
                S.extract_from_bounds(new_theta_p_Lambda)
                print(f"S.theta: {S.theta}")
                S.build_Psi()
                print(f"S.Psi: {S.Psi}")
                [[0]
                [1]]
                [0 1]
                S.theta: [0.]
                S.theta: [1.72284258]
                S.Psi: [[1.00000001e+00 1.14348852e-23]
                [1.14348852e-23 1.00000001e+00]]

        """
        self.Psi = zeros((self.n, self.n), dtype=float64)
        theta = power(10.0, self.theta)
        if self.n_theta == 1:
            theta = theta * ones(self.k)
        try:
            D = zeros((self.n, self.n))
            if self.ordered_mask.any():
                X_ordered = self.nat_X[:, self.ordered_mask]
                D = squareform(
                    pdist(
                        X_ordered, metric='sqeuclidean', out=None, w=theta[self.ordered_mask]))
            if self.factor_mask.any():
                X_factor = self.nat_X[:, self.factor_mask]
                D = (D + squareform(
                    pdist(X_factor,
                          metric=self.metric_factorial,
                          out=None,
                          w=theta[self.factor_mask])))
            self.Psi = exp(-D)
        except LinAlgError as err:
            print(f"Building Psi failed:\n {self.Psi}. {err=}, {type(err)=}")
        if self.noise:
            logger.debug("In build_Psi(): self.Lambda: %s", self.Lambda)
            self.Psi[diag_indices_from(self.Psi)] += self.Lambda
        else:
            self.Psi[diag_indices_from(self.Psi)] += self.eps
        if (isinf(self.Psi)).any():
            self.inf_Psi = True
        self.cnd_Psi = cond(self.Psi)

    def build_U(self, scipy: bool = True) -> None:
        """
        Performs Cholesky factorization of Psi as U as described in [Forr08a, p.57].
        This method uses either `scipy_cholesky` or numpy's `cholesky` to perform the Cholesky factorization of Psi.

        Args:
            self (object):
                The Kriging object.
            scipy (bool):
                If True, use `scipy_cholesky`.
                If False, use numpy's `cholesky`.
                Defaults to True.

        Returns:
            None

        Raises:
            LinAlgError:
                If Cholesky factorization fails for Psi.

        Attributes:
            U (np.matrix): Kriging U matrix, Cholesky decomposition. Shape (n,n).

        Examples:
            >>> from spotpython.build.kriging import Kriging
                import numpy as np
                nat_X = np.array([[0], [1]])
                nat_y = np.array([0, 1])
                n=1
                p=1
                S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False)
                S.initialize_variables(nat_X, nat_y)
                S.set_variable_types()
                print(S.nat_X)
                print(S.nat_y)
                S.set_theta_values()
                print(f"S.theta: {S.theta}")
                S.initialize_matrices()
                S.set_de_bounds()
                new_theta_p_Lambda = S.optimize_model()
                S.extract_from_bounds(new_theta_p_Lambda)
                print(f"S.theta: {S.theta}")
                S.build_Psi()
                print(f"S.Psi: {S.Psi}")
                S.build_U()
                print(f"S.U:{S.U}")
                [[0]
                [1]]
                [0 1]
                S.theta: [0.]
                S.theta: [1.72284258]
                S.Psi: [[1.00000001e+00 1.14348852e-23]
                [1.14348852e-23 1.00000001e+00]]
                S.U:[[1.00000001e+00 1.14348851e-23]
                [0.00000000e+00 1.00000001e+00]]
        """
        try:
            self.U = scipy_cholesky(self.Psi, lower=True) if scipy else cholesky(self.Psi)
            self.U = self.U.T
        except LinAlgError as err:
            print(f"build_U() Cholesky failed for Psi:\n {self.Psi}. {err=}, {type(err)=}")

    def likelihood(self) -> None:
        """
        Calculates the negative of the concentrated log-likelihood.

        This method implements equation (2.32) in [Forr08a] to calculate
        the negative of the concentrated log-likelihood. It also modifies `mu`,
        `SigmaSqr`, `LnDetPsi`, and `negLnLike`.

        Note:
            `build_Psi` and `build_U` should be called first.

        Args:
            self (object):
                The Kriging object.

        Returns:
            None

        Attributes:
            mu (np.float64): Kriging expected mean value mu.
            SigmaSqr (np.float64): Sigma squared value.
            LnDetPsi (np.float64): Determinant Psi matrix.
            negLnLike (float): Negative log likelihood of the surface at the specified hyperparameters.

        Examples:
            >>> from spotpython.build.kriging import Kriging
                import numpy as np
                nat_X = np.array([[1], [2]])
                nat_y = np.array([5, 10])
                n=2
                p=1
                S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False, theta_init_zero=True)
                S.initialize_variables(nat_X, nat_y)
                S.set_variable_types()
                S.set_theta_values()
                S.initialize_matrices()
                S.build_Psi()
                S.build_U()
                S.likelihood()
                # assert S.mu is close to 7.5 with a tolerance of 1e-6
                assert np.allclose(S.mu, 7.5, atol=1e-6)
                E = np.exp(1)
                sigma2 = E/(E**2 -1) * (25/4 + 25/4*E)
                # asssert S.SigmaSqr is close to sigma2 with a tolerance of 1e-6
                assert np.allclose(S.SigmaSqr, sigma2, atol=1e-6)
                print(f"S.LnDetPsi:{S.LnDetPsi}")
                print(f"S.self.negLnLike:{S.negLnLike}")
        """
        # (2.20) in [Forr08a]:
        U_T_inv_one = solve(self.U.T, self.one)
        U_T_inv_cod_y = solve(self.U.T, self.nat_y)
        mu = self.one.T.dot(solve(self.U, U_T_inv_cod_y)) / self.one.T.dot(solve(self.U, U_T_inv_one))
        self.mu = mu
        # (2.31) in [Forr08a]
        cod_y_minus_mu = self.nat_y - self.one.dot(self.mu)
        self.SigmaSqr = cod_y_minus_mu.T.dot(solve(self.U, solve(self.U.T, cod_y_minus_mu))) / self.n
        # (2.32) in [Forr08a]
        self.LnDetPsi = 2.0 * sum(log(abs(diag(self.U))))
        self.negLnLike = -1.0 * (-(self.n / 2.0) * log(self.SigmaSqr) - 0.5 * self.LnDetPsi)

    def plot(self, show: Optional[bool] = True) -> None:
        """
        This function plots 1D and 2D surrogates.

        Args:
            self (object):
                The Kriging object.
            show (bool):
                If `True`, the plots are displayed.
                If `False`, `plt.show()` should be called outside this function.

        Returns:
            None

        Note:
            * This method provides only a basic plot. For more advanced plots,
                use the `plot_contour()` method of the `Spot` class.

        Examples:
            >>> import numpy as np
                from spotpython.fun.objectivefunctions import analytical
                from spotpython.spot import spot
                # 1-dimensional example
                fun = analytical().fun_sphere
                lower = np.array([-1])
                upper = np.array([1])
                design_control={"init_size": 10}
                S = spot.Spot(fun=fun,
                            noise=False,
                            lower = lower,
                            upper= upper,
                            design_control=design_control,)
                S.initialize_design()
                S.update_stats()
                S.fit_surrogate()
                S.surrogate.plot()
                # 2-dimensional example
                fun = analytical().fun_sphere
                lower = np.array([-1, -1])
                upper = np.array([1, 1])
                design_control={"init_size": 10}
                S = spot.Spot(fun=fun,
                            noise=False,
                            lower = lower,
                            upper= upper,
                            design_control=design_control,)
                S.initialize_design()
                S.update_stats()
                S.fit_surrogate()
                S.surrogate.plot()
        """
        if self.k == 1:
            # TODO: Improve plot (add conf. interval etc.)
            fig = pylab.figure(figsize=(9, 6))
            n_grid = 100
            x = linspace(
                self.min_X[0], self.max_X[0], num=n_grid
            )
            y = self.predict(x)
            plt.figure()
            plt.plot(x, y, "k")
            if show:
                plt.show()

        if self.k == 2:
            fig = pylab.figure(figsize=(9, 6))
            n_grid = 100
            x = linspace(
                self.min_X[0], self.max_X[0], num=n_grid
            )
            y = linspace(
                self.min_X[1], self.max_X[1], num=n_grid
            )
            X, Y = meshgrid(x, y)
            # Predict based on the optimized results
            zz = array(
                [self.predict(array([x, y]), return_val="all") for x, y in zip(ravel(X), ravel(Y))]
            )
            zs = zz[:, 0, :]
            zse = zz[:, 1, :]
            Z = zs.reshape(X.shape)
            Ze = zse.reshape(X.shape)

            nat_point_X = self.nat_X[:, 0]
            nat_point_Y = self.nat_X[:, 1]
            contour_levels = 30
            ax = fig.add_subplot(224)
            # plot predicted values:
            pylab.contourf(X, Y, Ze, contour_levels, cmap="jet")
            pylab.title("Error")
            pylab.colorbar()
            # plot observed points:
            pylab.plot(nat_point_X, nat_point_Y, "ow")
            #
            ax = fig.add_subplot(223)
            # plot predicted values:
            plt.contourf(X, Y, Z, contour_levels, zorder=1, cmap="jet")
            plt.title("Surrogate")
            # plot observed points:
            pylab.plot(nat_point_X, nat_point_Y, "ow", zorder=3)
            pylab.colorbar()
            #
            ax = fig.add_subplot(221, projection="3d")
            ax.plot_surface(X, Y, Z, rstride=3, cstride=3, alpha=0.9, cmap="jet")
            #
            ax = fig.add_subplot(222, projection="3d")
            ax.plot_surface(X, Y, Ze, rstride=3, cstride=3, alpha=0.9, cmap="jet")
            #
            pylab.show()

    def predict(self, nat_X: ndarray, return_val: str = "y") -> Union[float,
                                                                      Tuple[float, float]]:
        """
        This function returns the prediction (in natural units) of the surrogate at the natural coordinates of X.

        Args:
            self (object):
                The Kriging object.
            nat_X (ndarray):
                Design variable to evaluate in natural units.
            return_val (str):
                whether `y`, `s`, neg. `ei` (negative expected improvement),
                or all three values are returned.
                Default is (for compatibility with sklearn) "y". To return `s`, select "s",
                to return neg. `ei`, select "ei".
                To return the tuple `(y, s, ei)`, select "all".

        Returns:
            float:
                The predicted value in natural units if return_val is "y".
            float:
                predicted error if return_val is "s".
            float:
                expected improvement if return_val is "ei".
            Tuple[float, float, float]:
                The predicted value in natural units, predicted error
                and expected improvement if return_val is "all".

        Examples:
            >>> from spotpython.build.kriging import Kriging
                import numpy as np
                import matplotlib.pyplot as plt
                from numpy import linspace, arange
                rng = np.random.RandomState(1)
                X = linspace(start=0, stop=10, num=1_0).reshape(-1, 1)
                y = np.squeeze(X * np.sin(X))
                training_indices = rng.choice(arange(y.size), size=6, replace=False)
                X_train, y_train = X[training_indices], y[training_indices]
                S = Kriging(name='kriging', seed=124)
                S.fit(X_train, y_train)
                mean_prediction, std_prediction, s_ei = S.predict(X, return_val="all")
                print(f"mean_prediction: {mean_prediction}")
                print(f"std_prediction: {std_prediction}")
                print(f"s_ei: {s_ei}")
                mean_prediction: [-1.41991225e-08  6.48310037e-01  1.76715565e+00 -6.35226564e-01
                                  -4.28585379e+00 -1.22301198e+00  2.49434148e+00  5.61900501e-01
                                  -3.04558205e+00 -5.44021104e+00]
                std_prediction: [3.69706811e-04 2.07958787e+00 3.69706810e-04 3.69706807e-04
                                3.69706809e-04 2.07958584e+00 3.69706811e-04 2.60615408e+00
                                2.60837033e+00 3.69706811e-04]
                s_ei: [-0.00000000e+00 -1.02341235e-03 -0.00000000e+00 -0.00000000e+00
                       -0.00000000e+00 -1.63799181e-02 -0.00000000e+00 -9.45766290e-03
                       -2.53405666e-01 -1.47459347e-04]

        """
        # Check for the shape and the type of the Input
        if isinstance(nat_X, ndarray):
            try:
                X = nat_X.reshape(-1, self.nat_X.shape[1])
                X = repair_non_numeric(X, self.var_type)
            except Exception:
                raise TypeError("13.1: Input to predict was not convertible to the size of X")
        else:
            raise TypeError(f"type of the given input is an {type(nat_X)} instead of an ndarray")
        n = X.shape[0]
        y = empty(n, dtype=float)
        s = empty(n, dtype=float)
        ei = empty(n, dtype=float)
        for i in range(n):
            x = X[i, :]
            y[i], s[i], ei[i] = self.predict_coded(x)
        if return_val == "y":
            return y
        elif return_val == "s":
            return s
        elif return_val == "ei":
            return -1.0 * ei
        else:
            return y, s, -1.0 * ei

    def predict_coded(self, cod_x: np.ndarray) -> Tuple[float, float, float]:
        """
        Kriging prediction of one point in the coded units as described in (2.20) in [Forr08a].
        The error is returned as well.

        Args:
            self (object):
                The Kriging object.
            cod_x (np.ndarray):
                Point in coded units to make prediction at.

        Returns:
            f (float): Predicted value in coded units.
            SSqr (float): Predicted error.
            EI (float): Expected improvement.

        Note:
            `self.mu` and `self.SigmaSqr` are computed in `likelihood`, not here.
            See also [Forr08a, p.60].
        """
        self.build_psi_vec(cod_x)
        U_T_inv = solve(self.U.T, self.nat_y - self.one.dot(self.mu))
        f = self.mu + self.psi.T.dot(solve(self.U, U_T_inv))
        if self.noise:
            Lambda = self.Lambda
        else:
            Lambda = 0.0
        # Error in [Forr08a, p.87]:
        SSqr = self.SigmaSqr * (1 + Lambda - self.psi.T.dot(solve(self.U, solve(self.U.T, self.psi))))
        SSqr = power(abs(SSqr[0]), 0.5)[0]
        EI = self.exp_imp(y0=f[0], s0=SSqr)
        return f[0], SSqr, EI

    def build_psi_vec(self, cod_x: ndarray) -> None:
        """
        Build the psi vector. Needed by `predict_cod`, `predict_err_coded`,
        `regression_predict_coded`. Modifies `self.psi`.

        Args:
            self (object):
                The Kriging object.
            cod_x (ndarray):
                point to calculate psi

        Returns:
            None

        Attributes:
            self.psi (ndarray):
                psi vector

        Examples:
            >>> import numpy as np
                from spotpython.build.kriging import Kriging
                X_train = np.array([[1., 2.],
                                    [2., 4.],
                                    [3., 6.]])
                y_train = np.array([1., 2., 3.])
                S = Kriging(name='kriging',
                            seed=123,
                            log_level=50,
                            n_theta=1,
                            noise=False,
                            cod_type="norm")
                S.fit(X_train, y_train)
                # force theta to simple values:
                S.theta = np.array([0.0])
                nat_X = np.array([1., 0.])
                S.psi = np.zeros((S.n, 1))
                S.build_psi_vec(nat_X)
                res = np.array([[np.exp(-4)],
                    [np.exp(-17)],
                    [np.exp(-40)]])
                assert np.array_equal(S.psi, res)
                print(f"S.psi: {S.psi}")
                print(f"Control value res: {res}")
                S.psi:
                [[1.83156389e-02]
                [4.13993772e-08]
                [4.24835426e-18]]
                Control value res:
                [[1.83156389e-02]
                [4.13993772e-08]
                [4.24835426e-18]]
        """
        self.psi = zeros((self.n))
        theta = power(10.0, self.theta)
        if self.n_theta == 1:
            theta = theta * ones(self.k)
        try:
            D = zeros((self.n))
            if self.ordered_mask.any():
                X_ordered = self.nat_X[:, self.ordered_mask]
                x_ordered = cod_x[self.ordered_mask]
                D = cdist(x_ordered.reshape(-1, sum(self.ordered_mask)),
                          X_ordered.reshape(-1, sum(self.ordered_mask)),
                          metric='sqeuclidean',
                          out=None,
                          w=theta[self.ordered_mask])
            if self.factor_mask.any():
                X_factor = self.nat_X[:, self.factor_mask]
                x_factor = cod_x[self.factor_mask]
                D = (D + cdist(x_factor.reshape(-1, sum(self.factor_mask)),
                               X_factor.reshape(-1, sum(self.factor_mask)),
                               metric=self.metric_factorial,
                               out=None,
                               w=theta[self.factor_mask]))
            self.psi = exp(-D).T
        except LinAlgError as err:
            print(f"Building psi failed:\n {self.psi}. {err=}, {type(err)=}")

    def weighted_exp_imp(self, cod_x: np.ndarray, w: float) -> float:
        """
        Weighted expected improvement.

        Args:
            self (object): The Kriging object.
            cod_x (np.ndarray): A coded design vector.
            w (float): Weight.

        Returns:
            EI (float): Weighted expected improvement.

        References:
            [Sobester et al. 2005].
        """
        y0, s0 = self.predict_coded(cod_x)
        y_min = min(self.nat_y)
        if s0 <= 0.0:
            EI = 0.0
        else:
            y_min_y0 = y_min - y0
            EI_one = w * (
                    y_min_y0
                    * (0.5 + 0.5 * erf((1.0 / sqrt(2.0)) * (y_min_y0 / s0)))
            )
            EI_two = (
                    (1.0 - w)
                    * (s0 * (1.0 / sqrt(2.0 * pi)))
                    * (exp(-(1.0 / 2.0) * ((y_min_y0) ** 2.0 / s0 ** 2.0)))
            )
            EI = EI_one + EI_two
        return EI

__init__(noise=False, var_type=['num'], name='kriging', seed=124, model_optimizer=None, model_fun_evals=None, min_theta=-3.0, max_theta=2.0, n_theta=1, theta_init_zero=True, p_val=2.0, n_p=1, optim_p=False, min_Lambda=1e-09, max_Lambda=1.0, log_level=50, spot_writer=None, counter=None, metric_factorial='canberra', **kwargs)

Initialize the Kriging surrogate.

Parameters:

Name Type Description Default
noise bool

Use regression instead of interpolation kriging. Defaults to False.

False
var_type List[str]

Variable type. Can be either “num” (numerical) or “factor” (factor). Defaults to [“num”].

['num']
name str

Surrogate name. Defaults to “kriging”.

'kriging'
seed int

Random seed. Defaults to 124.

124
model_optimizer Optional[object]

Optimizer on the surrogate. If None, differential_evolution is selected.

None
model_fun_evals Optional[int]

Number of iterations used by the optimizer on the surrogate.

None
min_theta float

Min log10 theta value. Defaults to -3.

-3.0
max_theta float

Max log10 theta value. Defaults to 2.

2.0
n_theta int

Number of theta values. Defaults to 1.

1
theta_init_zero bool

Initialize theta with zero. Defaults to True.

True
p_val float

p value. Used as an initial value if optim_p = True. Otherwise as a constant. Defaults to 2.

2.0
n_p int

Number of p values. Defaults to 1.

1
optim_p bool

Determines whether p should be optimized. Deafults to False.

False
min_Lambda float

Min Lambda value. Defaults to 1e-9.

1e-09
max_Lambda float

Max Lambda value. Defaults to 1.

1.0
log_level int

Logging level, e.g., 20 is “INFO”. Defaults to 50 (“CRITICAL”).

50
spot_writer Optional[object]

Spot writer. Defaults to None.

None
counter Optional[int]

Counter. Defaults to None.

None
metric_factorial str

Metric for factorial. Defaults to “canberra”. Can be “euclidean”, “cityblock”, seuclidean”, “sqeuclidean”, “cosine”, “correlation”, “hamming”, “jaccard”, “jensenshannon”, “chebyshev”, “canberra”, “braycurtis”, “mahalanobis”, “matching”.

'canberra'

Examples:

>>> from spotpython.build.kriging import Kriging
    import numpy as np
    import matplotlib.pyplot as plt
    from numpy import linspace, arange
    rng = np.random.RandomState(1)
    X = linspace(start=0, stop=10, num=1_000).reshape(-1, 1)
    y = np.squeeze(X * np.sin(X))
    training_indices = rng.choice(arange(y.size), size=6, replace=False)
    X_train, y_train = X[training_indices], y[training_indices]
    S = Kriging(name='kriging', seed=124)
    S.fit(X_train, y_train)
    mean_prediction, std_prediction, s_ei = S.predict(X, return_val="all")
    plt.plot(X, y, label=r"$f(x)$", linestyle="dotted")
    plt.scatter(X_train, y_train, label="Observations")
    plt.plot(X, mean_prediction, label="Mean prediction")
    plt.fill_between(
        X.ravel(),
        mean_prediction - 1.96 * std_prediction,
        mean_prediction + 1.96 * std_prediction,
        alpha=0.5,
        label=r"95% confidence interval",
        )
    plt.legend()
    plt.xlabel("$x$")
    plt.ylabel("$f(x)$")
    _ = plt.title("Gaussian process regression on noise-free dataset")
    plt.show()
References

https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.distance.pdist.html [1] scikit-learn: Gaussian Processes regression: basic introductory example

Source code in spotpython/build/kriging.py
 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
def __init__(
        self: object,
        noise: bool = False,
        var_type: List[str] = ["num"],
        name: str = "kriging",
        seed: int = 124,
        model_optimizer=None,
        model_fun_evals: Optional[int] = None,
        min_theta: float = -3.0,
        max_theta: float = 2.0,
        n_theta: int = 1,
        theta_init_zero: bool = True,
        p_val: float = 2.0,
        n_p: int = 1,
        optim_p: bool = False,
        min_Lambda: float = 1e-9,
        max_Lambda: float = 1.,
        log_level: int = 50,
        spot_writer=None,
        counter=None,
        metric_factorial="canberra",
        **kwargs
):
    """
    Initialize the Kriging surrogate.

    Args:
        noise (bool): Use regression instead of interpolation kriging. Defaults to False.
        var_type (List[str]):
            Variable type. Can be either "num" (numerical) or "factor" (factor).
            Defaults to ["num"].
        name (str):
            Surrogate name. Defaults to "kriging".
        seed (int):
            Random seed. Defaults to 124.
        model_optimizer (Optional[object]):
            Optimizer on the surrogate. If None, differential_evolution is selected.
        model_fun_evals (Optional[int]):
            Number of iterations used by the optimizer on the surrogate.
        min_theta (float):
            Min log10 theta value. Defaults to -3.
        max_theta (float):
            Max log10 theta value. Defaults to 2.
        n_theta (int):
            Number of theta values. Defaults to 1.
        theta_init_zero (bool):
            Initialize theta with zero. Defaults to True.
        p_val (float):
            p value. Used as an initial value if optim_p = True. Otherwise as a constant. Defaults to 2.
        n_p (int):
            Number of p values. Defaults to 1.
        optim_p (bool):
            Determines whether p should be optimized. Deafults to False.
        min_Lambda (float):
            Min Lambda value. Defaults to 1e-9.
        max_Lambda (float):
            Max Lambda value. Defaults to 1.
        log_level (int):
            Logging level, e.g., 20 is "INFO". Defaults to 50 ("CRITICAL").
        spot_writer (Optional[object]):
            Spot writer. Defaults to None.
        counter (Optional[int]):
            Counter. Defaults to None.
        metric_factorial (str):
            Metric for factorial. Defaults to "canberra". Can be "euclidean",
            "cityblock", seuclidean", "sqeuclidean", "cosine",
            "correlation", "hamming", "jaccard", "jensenshannon",
            "chebyshev", "canberra", "braycurtis", "mahalanobis", "matching".

    Examples:
        >>> from spotpython.build.kriging import Kriging
            import numpy as np
            import matplotlib.pyplot as plt
            from numpy import linspace, arange
            rng = np.random.RandomState(1)
            X = linspace(start=0, stop=10, num=1_000).reshape(-1, 1)
            y = np.squeeze(X * np.sin(X))
            training_indices = rng.choice(arange(y.size), size=6, replace=False)
            X_train, y_train = X[training_indices], y[training_indices]
            S = Kriging(name='kriging', seed=124)
            S.fit(X_train, y_train)
            mean_prediction, std_prediction, s_ei = S.predict(X, return_val="all")
            plt.plot(X, y, label=r"$f(x)$", linestyle="dotted")
            plt.scatter(X_train, y_train, label="Observations")
            plt.plot(X, mean_prediction, label="Mean prediction")
            plt.fill_between(
                X.ravel(),
                mean_prediction - 1.96 * std_prediction,
                mean_prediction + 1.96 * std_prediction,
                alpha=0.5,
                label=r"95% confidence interval",
                )
            plt.legend()
            plt.xlabel("$x$")
            plt.ylabel("$f(x)$")
            _ = plt.title("Gaussian process regression on noise-free dataset")
            plt.show()

    References:
        https://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.distance.pdist.html
        [[1](https://scikit-learn.org/stable/auto_examples/gaussian_process/plot_gpr_noisy_targets.html)]
        scikit-learn: Gaussian Processes regression: basic introductory example

    """
    super().__init__(name, seed, log_level)

    self.noise = noise
    self.var_type = var_type
    self.name = name
    self.seed = seed
    self.log_level = log_level
    self.spot_writer = spot_writer
    self.counter = counter
    self.metric_factorial = metric_factorial

    self.sigma = 0
    self.eps = sqrt(spacing(1))
    self.min_theta = min_theta
    self.max_theta = max_theta
    self.min_p = 1
    self.max_p = 2
    self.min_Lambda = min_Lambda
    self.max_Lambda = max_Lambda
    self.n_theta = n_theta
    self.p_val = p_val
    self.n_p = n_p
    self.optim_p = optim_p
    self.theta_init_zero = theta_init_zero
    # Psi matrix condition:
    self.cnd_Psi = 0
    self.inf_Psi = False

    self.model_optimizer = model_optimizer
    if self.model_optimizer is None:
        self.model_optimizer = differential_evolution
    self.model_fun_evals = model_fun_evals
    # differential evolution uses maxiter = 1000
    # and sets the number of function evaluations to
    # (maxiter + 1) * popsize * N, which results in
    # 1000 * 15 * k, because the default popsize is 15 and
    # N is the number of parameters. This seems to be quite large:
    # for k=2 these are 30 000 iterations. Therefore we set this value to
    # 100
    if self.model_fun_evals is None:
        self.model_fun_evals = 100

    # Logging information
    self.log["negLnLike"] = []
    self.log["theta"] = []
    self.log["p"] = []
    self.log["Lambda"] = []
    # Logger
    logger.setLevel(self.log_level)
    logger.info(f"Starting the logger at level {self.log_level} for module {__name__}:")

__is_any__(x, v)

Check if any element in x is equal to v. This method checks if any element in the input array x is equal to the value v. If x is not an instance of ndarray, it is first converted to a numpy array using the array() function.

Parameters:

Name Type Description Default
self object

The Kriging object.

required
x ndarray or array - like

The input array to check for the presence of value v.

required
v scalar

The value to check for in the input array x.

required

Returns:

Name Type Description
bool bool

True if any element in x is equal to v, False otherwise.

Examples:

>>> from spotpython.build.kriging import Kriging
    from numpy import power
    import numpy as np
    nat_X = np.array([[0], [1]])
    nat_y = np.array([0, 1])
    n=1
    p=1
    S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False)
    S.initialize_variables(nat_X, nat_y)
    S.set_variable_types()
    S.set_theta_values()
    print(f"S.theta: {S.theta}")
    print(S.__is_any__(power(10.0, S.theta), 0))
    print(S.__is_any__(S.theta, 0))
    S.theta: [0.]
    False
    True
Source code in spotpython/build/kriging.py
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
def __is_any__(self, x: Union[np.ndarray, Any], v: Any) -> bool:
    """
    Check if any element in `x` is equal to `v`.
    This method checks if any element in the input array `x` is equal to the value `v`.
    If `x` is not an instance of `ndarray`, it is first converted to a numpy array using
    the `array()` function.

    Args:
        self (object): The Kriging object.
        x (np.ndarray or array-like):
            The input array to check for the presence of value `v`.
        v (scalar):
            The value to check for in the input array `x`.

    Returns:
        bool:
            True if any element in `x` is equal to `v`, False otherwise.

    Examples:
        >>> from spotpython.build.kriging import Kriging
            from numpy import power
            import numpy as np
            nat_X = np.array([[0], [1]])
            nat_y = np.array([0, 1])
            n=1
            p=1
            S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False)
            S.initialize_variables(nat_X, nat_y)
            S.set_variable_types()
            S.set_theta_values()
            print(f"S.theta: {S.theta}")
            print(S.__is_any__(power(10.0, S.theta), 0))
            print(S.__is_any__(S.theta, 0))
            S.theta: [0.]
            False
            True

    """
    if not isinstance(x, ndarray):
        x = array([x])
    return any(x == v)

build_Psi()

Constructs a new (n x n) correlation matrix Psi to reflect new data or a change in hyperparameters.

This method uses theta, p, and coded X values to construct the correlation matrix as described in [Forr08a, p.57].

Parameters:

Name Type Description Default
self object

The Kriging object.

required

Returns:

Type Description
None

None

Raises:

Type Description
LinAlgError

If building Psi fails.

Attributes:

Name Type Description
Psi matrix

Correlation matrix Psi. Shape (n,n).

cnd_Psi float

Condition number of Psi.

inf_Psi bool

True if Psi is infinite, False otherwise.

Examples:

>>> from spotpython.build.kriging import Kriging
    import numpy as np
    nat_X = np.array([[0], [1]])
    nat_y = np.array([0, 1])
    n=1
    p=1
    S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False)
    S.initialize_variables(nat_X, nat_y)
    S.set_variable_types()
    print(S.nat_X)
    print(S.nat_y)
    S.set_theta_values()
    print(f"S.theta: {S.theta}")
    S.initialize_matrices()
    S.set_de_bounds()
    new_theta_p_Lambda = S.optimize_model()
    S.extract_from_bounds(new_theta_p_Lambda)
    print(f"S.theta: {S.theta}")
    S.build_Psi()
    print(f"S.Psi: {S.Psi}")
    [[0]
    [1]]
    [0 1]
    S.theta: [0.]
    S.theta: [1.72284258]
    S.Psi: [[1.00000001e+00 1.14348852e-23]
    [1.14348852e-23 1.00000001e+00]]
Source code in spotpython/build/kriging.py
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
def build_Psi(self) -> None:
    """
    Constructs a new (n x n) correlation matrix Psi to reflect new data
    or a change in hyperparameters.

    This method uses `theta`, `p`, and coded `X` values to construct the
    correlation matrix as described in [Forr08a, p.57].

    Args:
        self (object): The Kriging object.

    Returns:
        None

    Raises:
        LinAlgError: If building Psi fails.

    Attributes:
        Psi (np.matrix): Correlation matrix Psi. Shape (n,n).
        cnd_Psi (float): Condition number of Psi.
        inf_Psi (bool): True if Psi is infinite, False otherwise.

    Examples:
        >>> from spotpython.build.kriging import Kriging
            import numpy as np
            nat_X = np.array([[0], [1]])
            nat_y = np.array([0, 1])
            n=1
            p=1
            S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False)
            S.initialize_variables(nat_X, nat_y)
            S.set_variable_types()
            print(S.nat_X)
            print(S.nat_y)
            S.set_theta_values()
            print(f"S.theta: {S.theta}")
            S.initialize_matrices()
            S.set_de_bounds()
            new_theta_p_Lambda = S.optimize_model()
            S.extract_from_bounds(new_theta_p_Lambda)
            print(f"S.theta: {S.theta}")
            S.build_Psi()
            print(f"S.Psi: {S.Psi}")
            [[0]
            [1]]
            [0 1]
            S.theta: [0.]
            S.theta: [1.72284258]
            S.Psi: [[1.00000001e+00 1.14348852e-23]
            [1.14348852e-23 1.00000001e+00]]

    """
    self.Psi = zeros((self.n, self.n), dtype=float64)
    theta = power(10.0, self.theta)
    if self.n_theta == 1:
        theta = theta * ones(self.k)
    try:
        D = zeros((self.n, self.n))
        if self.ordered_mask.any():
            X_ordered = self.nat_X[:, self.ordered_mask]
            D = squareform(
                pdist(
                    X_ordered, metric='sqeuclidean', out=None, w=theta[self.ordered_mask]))
        if self.factor_mask.any():
            X_factor = self.nat_X[:, self.factor_mask]
            D = (D + squareform(
                pdist(X_factor,
                      metric=self.metric_factorial,
                      out=None,
                      w=theta[self.factor_mask])))
        self.Psi = exp(-D)
    except LinAlgError as err:
        print(f"Building Psi failed:\n {self.Psi}. {err=}, {type(err)=}")
    if self.noise:
        logger.debug("In build_Psi(): self.Lambda: %s", self.Lambda)
        self.Psi[diag_indices_from(self.Psi)] += self.Lambda
    else:
        self.Psi[diag_indices_from(self.Psi)] += self.eps
    if (isinf(self.Psi)).any():
        self.inf_Psi = True
    self.cnd_Psi = cond(self.Psi)

build_U(scipy=True)

Performs Cholesky factorization of Psi as U as described in [Forr08a, p.57]. This method uses either scipy_cholesky or numpy’s cholesky to perform the Cholesky factorization of Psi.

Parameters:

Name Type Description Default
self object

The Kriging object.

required
scipy bool

If True, use scipy_cholesky. If False, use numpy’s cholesky. Defaults to True.

True

Returns:

Type Description
None

None

Raises:

Type Description
LinAlgError

If Cholesky factorization fails for Psi.

Attributes:

Name Type Description
U matrix

Kriging U matrix, Cholesky decomposition. Shape (n,n).

Examples:

>>> from spotpython.build.kriging import Kriging
    import numpy as np
    nat_X = np.array([[0], [1]])
    nat_y = np.array([0, 1])
    n=1
    p=1
    S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False)
    S.initialize_variables(nat_X, nat_y)
    S.set_variable_types()
    print(S.nat_X)
    print(S.nat_y)
    S.set_theta_values()
    print(f"S.theta: {S.theta}")
    S.initialize_matrices()
    S.set_de_bounds()
    new_theta_p_Lambda = S.optimize_model()
    S.extract_from_bounds(new_theta_p_Lambda)
    print(f"S.theta: {S.theta}")
    S.build_Psi()
    print(f"S.Psi: {S.Psi}")
    S.build_U()
    print(f"S.U:{S.U}")
    [[0]
    [1]]
    [0 1]
    S.theta: [0.]
    S.theta: [1.72284258]
    S.Psi: [[1.00000001e+00 1.14348852e-23]
    [1.14348852e-23 1.00000001e+00]]
    S.U:[[1.00000001e+00 1.14348851e-23]
    [0.00000000e+00 1.00000001e+00]]
Source code in spotpython/build/kriging.py
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
def build_U(self, scipy: bool = True) -> None:
    """
    Performs Cholesky factorization of Psi as U as described in [Forr08a, p.57].
    This method uses either `scipy_cholesky` or numpy's `cholesky` to perform the Cholesky factorization of Psi.

    Args:
        self (object):
            The Kriging object.
        scipy (bool):
            If True, use `scipy_cholesky`.
            If False, use numpy's `cholesky`.
            Defaults to True.

    Returns:
        None

    Raises:
        LinAlgError:
            If Cholesky factorization fails for Psi.

    Attributes:
        U (np.matrix): Kriging U matrix, Cholesky decomposition. Shape (n,n).

    Examples:
        >>> from spotpython.build.kriging import Kriging
            import numpy as np
            nat_X = np.array([[0], [1]])
            nat_y = np.array([0, 1])
            n=1
            p=1
            S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False)
            S.initialize_variables(nat_X, nat_y)
            S.set_variable_types()
            print(S.nat_X)
            print(S.nat_y)
            S.set_theta_values()
            print(f"S.theta: {S.theta}")
            S.initialize_matrices()
            S.set_de_bounds()
            new_theta_p_Lambda = S.optimize_model()
            S.extract_from_bounds(new_theta_p_Lambda)
            print(f"S.theta: {S.theta}")
            S.build_Psi()
            print(f"S.Psi: {S.Psi}")
            S.build_U()
            print(f"S.U:{S.U}")
            [[0]
            [1]]
            [0 1]
            S.theta: [0.]
            S.theta: [1.72284258]
            S.Psi: [[1.00000001e+00 1.14348852e-23]
            [1.14348852e-23 1.00000001e+00]]
            S.U:[[1.00000001e+00 1.14348851e-23]
            [0.00000000e+00 1.00000001e+00]]
    """
    try:
        self.U = scipy_cholesky(self.Psi, lower=True) if scipy else cholesky(self.Psi)
        self.U = self.U.T
    except LinAlgError as err:
        print(f"build_U() Cholesky failed for Psi:\n {self.Psi}. {err=}, {type(err)=}")

build_psi_vec(cod_x)

Build the psi vector. Needed by predict_cod, predict_err_coded, regression_predict_coded. Modifies self.psi.

Parameters:

Name Type Description Default
self object

The Kriging object.

required
cod_x ndarray

point to calculate psi

required

Returns:

Type Description
None

None

Attributes:

Name Type Description
self.psi ndarray

psi vector

Examples:

>>> import numpy as np
    from spotpython.build.kriging import Kriging
    X_train = np.array([[1., 2.],
                        [2., 4.],
                        [3., 6.]])
    y_train = np.array([1., 2., 3.])
    S = Kriging(name='kriging',
                seed=123,
                log_level=50,
                n_theta=1,
                noise=False,
                cod_type="norm")
    S.fit(X_train, y_train)
    # force theta to simple values:
    S.theta = np.array([0.0])
    nat_X = np.array([1., 0.])
    S.psi = np.zeros((S.n, 1))
    S.build_psi_vec(nat_X)
    res = np.array([[np.exp(-4)],
        [np.exp(-17)],
        [np.exp(-40)]])
    assert np.array_equal(S.psi, res)
    print(f"S.psi: {S.psi}")
    print(f"Control value res: {res}")
    S.psi:
    [[1.83156389e-02]
    [4.13993772e-08]
    [4.24835426e-18]]
    Control value res:
    [[1.83156389e-02]
    [4.13993772e-08]
    [4.24835426e-18]]
Source code in spotpython/build/kriging.py
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
def build_psi_vec(self, cod_x: ndarray) -> None:
    """
    Build the psi vector. Needed by `predict_cod`, `predict_err_coded`,
    `regression_predict_coded`. Modifies `self.psi`.

    Args:
        self (object):
            The Kriging object.
        cod_x (ndarray):
            point to calculate psi

    Returns:
        None

    Attributes:
        self.psi (ndarray):
            psi vector

    Examples:
        >>> import numpy as np
            from spotpython.build.kriging import Kriging
            X_train = np.array([[1., 2.],
                                [2., 4.],
                                [3., 6.]])
            y_train = np.array([1., 2., 3.])
            S = Kriging(name='kriging',
                        seed=123,
                        log_level=50,
                        n_theta=1,
                        noise=False,
                        cod_type="norm")
            S.fit(X_train, y_train)
            # force theta to simple values:
            S.theta = np.array([0.0])
            nat_X = np.array([1., 0.])
            S.psi = np.zeros((S.n, 1))
            S.build_psi_vec(nat_X)
            res = np.array([[np.exp(-4)],
                [np.exp(-17)],
                [np.exp(-40)]])
            assert np.array_equal(S.psi, res)
            print(f"S.psi: {S.psi}")
            print(f"Control value res: {res}")
            S.psi:
            [[1.83156389e-02]
            [4.13993772e-08]
            [4.24835426e-18]]
            Control value res:
            [[1.83156389e-02]
            [4.13993772e-08]
            [4.24835426e-18]]
    """
    self.psi = zeros((self.n))
    theta = power(10.0, self.theta)
    if self.n_theta == 1:
        theta = theta * ones(self.k)
    try:
        D = zeros((self.n))
        if self.ordered_mask.any():
            X_ordered = self.nat_X[:, self.ordered_mask]
            x_ordered = cod_x[self.ordered_mask]
            D = cdist(x_ordered.reshape(-1, sum(self.ordered_mask)),
                      X_ordered.reshape(-1, sum(self.ordered_mask)),
                      metric='sqeuclidean',
                      out=None,
                      w=theta[self.ordered_mask])
        if self.factor_mask.any():
            X_factor = self.nat_X[:, self.factor_mask]
            x_factor = cod_x[self.factor_mask]
            D = (D + cdist(x_factor.reshape(-1, sum(self.factor_mask)),
                           X_factor.reshape(-1, sum(self.factor_mask)),
                           metric=self.metric_factorial,
                           out=None,
                           w=theta[self.factor_mask]))
        self.psi = exp(-D).T
    except LinAlgError as err:
        print(f"Building psi failed:\n {self.psi}. {err=}, {type(err)=}")

exp_imp(y0, s0)

Calculates the expected improvement for a given function value and error in coded units.

Parameters:

Name Type Description Default
self object

The Kriging object.

required
y0 float

The function value in coded units.

required
s0 float

The error value.

required

Returns:

Name Type Description
float float

The expected improvement value.

Examples:

>>> from spotpython.build.kriging import Kriging
    S = Kriging(name='kriging', seed=124)
    S.aggregated_mean_y = [0.0, 0.0, 0.0, 0.0, 0.0]
    S.exp_imp(1.0, 0.0)
    0.0
>>> from spotpython.build.kriging import Kriging
    S = Kriging(name='kriging', seed=124)
    S.aggregated_mean_y = [0.0, 0.0, 0.0, 0.0, 0.0]
    # assert S.exp_imp(0.0, 1.0) == 1/np.sqrt(2*np.pi)
    # which is approx. 0.3989422804014327
    S.exp_imp(0.0, 1.0)
    0.3989422804014327
Source code in spotpython/build/kriging.py
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
def exp_imp(self, y0: float, s0: float) -> float:
    """
    Calculates the expected improvement for a given function value and error in coded units.

    Args:
        self (object): The Kriging object.
        y0 (float): The function value in coded units.
        s0 (float): The error value.

    Returns:
        float: The expected improvement value.

    Examples:
        >>> from spotpython.build.kriging import Kriging
            S = Kriging(name='kriging', seed=124)
            S.aggregated_mean_y = [0.0, 0.0, 0.0, 0.0, 0.0]
            S.exp_imp(1.0, 0.0)
            0.0
        >>> from spotpython.build.kriging import Kriging
            S = Kriging(name='kriging', seed=124)
            S.aggregated_mean_y = [0.0, 0.0, 0.0, 0.0, 0.0]
            # assert S.exp_imp(0.0, 1.0) == 1/np.sqrt(2*np.pi)
            # which is approx. 0.3989422804014327
            S.exp_imp(0.0, 1.0)
            0.3989422804014327
    """
    # We do not use the min y values, but the aggragated mean values
    # y_min = min(self.nat_y)
    y_min = min(self.aggregated_mean_y)
    if s0 <= 0.0:
        EI = 0.0
    elif s0 > 0.0:
        EI_one = (y_min - y0) * (
                0.5 + 0.5 * erf((1.0 / sqrt(2.0)) * ((y_min - y0) / s0))
        )
        EI_two = (s0 * (1.0 / sqrt(2.0 * pi))) * (
            exp(-(1.0 / 2.0) * ((y_min - y0) ** 2.0 / s0 ** 2.0))
        )
        EI = EI_one + EI_two
    return EI

extract_from_bounds(new_theta_p_Lambda)

Extract theta, p, and Lambda from bounds. The kriging object stores theta as an array, p as an array, and Lambda as a float.

Parameters:

Name Type Description Default
self object

The Kriging object.

required
new_theta_p_Lambda ndarray

1d-array with theta, p, and Lambda values. Order is important.

required

Examples:

>>> import numpy as np
    from spotpython.build.kriging import Kriging
    n=2
    p=4
    S = Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False)
    S.extract_from_bounds(np.array([1, 2, 3]))
    print(S.theta)
    print(S.p)
    [1 2]
    [3]

Returns:

Type Description
None

None

Source code in spotpython/build/kriging.py
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
def extract_from_bounds(self, new_theta_p_Lambda: np.ndarray) -> None:
    """
    Extract `theta`, `p`, and `Lambda` from bounds. The kriging object stores
    `theta` as an array,  `p` as an array, and `Lambda` as a float.

    Args:
        self (object): The Kriging object.
        new_theta_p_Lambda (np.ndarray):
            1d-array with theta, p, and Lambda values. Order is important.

    Examples:
        >>> import numpy as np
            from spotpython.build.kriging import Kriging
            n=2
            p=4
            S = Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False)
            S.extract_from_bounds(np.array([1, 2, 3]))
            print(S.theta)
            print(S.p)
            [1 2]
            [3]

    Returns:
        None
    """
    logger.debug("In extract_from_bounds(): new_theta_p_Lambda: %s", new_theta_p_Lambda)
    self.theta = new_theta_p_Lambda[:self.n_theta]
    logger.debug("In extract_from_bounds(): self.n_theta: %s", self.n_theta)
    if self.optim_p:
        self.p = new_theta_p_Lambda[self.n_theta:self.n_theta + self.n_p]
        logger.debug("In extract_from_bounds(): self.p: %s", self.p)
        if self.noise:
            self.Lambda = new_theta_p_Lambda[self.n_theta + self.n_p]
            logger.debug("In extract_from_bounds(): self.Lambda: %s", self.Lambda)
    else:
        if self.noise:
            self.Lambda = new_theta_p_Lambda[self.n_theta]
            logger.debug("In extract_from_bounds(): self.Lambda: %s", self.Lambda)

fit(nat_X, nat_y)

Fits the hyperparameters (theta, p, Lambda) of the Kriging model.

The function computes the following internal values: 1. theta, p, and Lambda values via optimization of the function fun_likelihood(). 2. Correlation matrix Psi via rebuildPsi().

Parameters:

Name Type Description Default
self object

The Kriging object.

required
nat_X ndarray

Sample points.

required
nat_y ndarray

Function values.

required

Returns:

Name Type Description
object object

Fitted estimator.

Attributes:

Name Type Description
theta ndarray

Kriging theta values. Shape (k,).

p ndarray

Kriging p values. Shape (k,).

LnDetPsi float64

Determinant Psi matrix.

Psi matrix

Correlation matrix Psi. Shape (n,n).

psi ndarray

psi vector. Shape (n,).

one ndarray

vector of ones. Shape (n,).

mu float64

Kriging expected mean value mu.

U matrix

Kriging U matrix, Cholesky decomposition. Shape (n,n).

SigmaSqr float64

Sigma squared value.

Lambda float

lambda noise value.

Examples:

>>> from spotpython.build.kriging import Kriging
    import numpy as np
    nat_X = np.array([[1, 0], [1, 0]])
    nat_y = np.array([1, 2])
    S = Kriging()
    S.fit(nat_X, nat_y)
    print(S.Psi)
    [[1.00000001 1.        ]
    [1.         1.00000001]]
Source code in spotpython/build/kriging.py
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
def fit(self, nat_X: np.ndarray, nat_y: np.ndarray) -> object:
    """
    Fits the hyperparameters (`theta`, `p`, `Lambda`) of the Kriging model.

    The function computes the following internal values:
    1. `theta`, `p`, and `Lambda` values via optimization of the function `fun_likelihood()`.
    2. Correlation matrix `Psi` via `rebuildPsi()`.

    Args:
        self (object): The Kriging object.
        nat_X (np.ndarray): Sample points.
        nat_y (np.ndarray): Function values.

    Returns:
        object: Fitted estimator.

    Attributes:
        theta (np.ndarray): Kriging theta values. Shape (k,).
        p (np.ndarray): Kriging p values. Shape (k,).
        LnDetPsi (np.float64): Determinant Psi matrix.
        Psi (np.matrix): Correlation matrix Psi. Shape (n,n).
        psi (np.ndarray): psi vector. Shape (n,).
        one (np.ndarray): vector of ones. Shape (n,).
        mu (np.float64): Kriging expected mean value mu.
        U (np.matrix): Kriging U matrix, Cholesky decomposition. Shape (n,n).
        SigmaSqr (np.float64): Sigma squared value.
        Lambda (float): lambda noise value.

    Examples:
        >>> from spotpython.build.kriging import Kriging
            import numpy as np
            nat_X = np.array([[1, 0], [1, 0]])
            nat_y = np.array([1, 2])
            S = Kriging()
            S.fit(nat_X, nat_y)
            print(S.Psi)
            [[1.00000001 1.        ]
            [1.         1.00000001]]

    """
    logger.debug("In fit(): nat_X: %s", nat_X)
    logger.debug("In fit(): nat_y: %s", nat_y)
    self.initialize_variables(nat_X, nat_y)
    self.set_variable_types()
    self.set_theta_values()
    self.initialize_matrices()
    # build_Psi() and build_U() are called in fun_likelihood
    self.set_de_bounds()
    # Finally, set new theta and p values and update the surrogate again
    # for new_theta_p_Lambda in de_results["x"]:
    new_theta_p_Lambda = self.optimize_model()
    self.extract_from_bounds(new_theta_p_Lambda)
    self.build_Psi()
    self.build_U()
    # TODO: check if the following line is necessary!
    self.likelihood()
    self.update_log()

fun_likelihood(new_theta_p_Lambda)

Compute log likelihood for a set of hyperparameters (theta, p, Lambda).

This method computes the log likelihood for a set of hyperparameters (theta, p, Lambda) by performing the following steps: 1. Extracts the hyperparameters from the input array using extract_from_bounds(). 2. Checks if any element in 10^theta is equal to 0. If so, logs a warning and returns the penalty value (pen_val). 3. Builds the Psi matrix using build_Psi(). 4. Checks if Psi is ill-conditioned or infinite. If so, logs a warning and returns the penalty value (pen_val). 5. Builds the U matrix using build_U(). If an exception occurs, logs an error and returns the penalty value (pen_val). 6. Computes the negative log likelihood using likelihood(). 7. Returns the computed negative log likelihood (negLnLike).

Parameters:

Name Type Description Default
self object

The Kriging object.

required
new_theta_p_Lambda ndarray

An array containing the theta, p, and Lambda values.

required

Returns:

Name Type Description
float float

The negative log likelihood of the surface at the specified hyperparameters.

Attributes:

Name Type Description
theta ndarray

Kriging theta values. Shape (k,).

p ndarray

Kriging p values. Shape (k,).

Lambda float

lambda noise value.

Psi matrix

Correlation matrix Psi. Shape (n,n).

U matrix

Kriging U matrix, Cholesky decomposition. Shape (n,n).

negLnLike float

Negative log likelihood of the surface at the specified hyperparameters.

pen_val float

Penalty value.

Examples:

>>> from spotpython.build.kriging import Kriging
    import numpy as np
    nat_X = np.array([[0], [1]])
    nat_y = np.array([0, 1])
    n=1
    p=1
    S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False)
    S.initialize_variables(nat_X, nat_y)
    S.set_variable_types()
    print(S.cod_X)
    print(S.cod_y)
    S.set_theta_values()
    print(f"S.theta: {S.theta}")
    S.initialize_matrices()
    S.set_de_bounds()
    new_theta_p_Lambda = S.optimize_model()
    S.extract_from_bounds(new_theta_p_Lambda)
    print(f"S.theta: {S.theta}")
    S.build_Psi()
    print(f"S.Psi: {S.Psi}")
    S.build_U()
    print(f"S.U:{S.U}")
    S.likelihood()
    S.negLnLike
Source code in spotpython/build/kriging.py
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
def fun_likelihood(self, new_theta_p_Lambda: np.ndarray) -> float:
    """
    Compute log likelihood for a set of hyperparameters (theta, p, Lambda).

    This method computes the log likelihood for a set of hyperparameters
    (theta, p, Lambda) by performing the following steps:
    1. Extracts the hyperparameters from the input array using `extract_from_bounds()`.
    2. Checks if any element in `10^theta` is equal to 0. If so, logs a warning and
    returns the penalty value (`pen_val`).
    3. Builds the `Psi` matrix using `build_Psi()`.
    4. Checks if `Psi` is ill-conditioned or infinite. If so, logs a warning and returns
    the penalty value (`pen_val`).
    5. Builds the `U` matrix using `build_U()`. If an exception occurs, logs an error and
    returns the penalty value (`pen_val`).
    6. Computes the negative log likelihood using `likelihood()`.
    7. Returns the computed negative log likelihood (`negLnLike`).

    Args:
        self (object): The Kriging object.
        new_theta_p_Lambda (np.ndarray):
            An array containing the `theta`, `p`, and `Lambda` values.

    Returns:
        float:
            The negative log likelihood of the surface at the specified hyperparameters.

    Attributes:
        theta (np.ndarray): Kriging theta values. Shape (k,).
        p (np.ndarray): Kriging p values. Shape (k,).
        Lambda (float): lambda noise value.
        Psi (np.matrix): Correlation matrix Psi. Shape (n,n).
        U (np.matrix): Kriging U matrix, Cholesky decomposition. Shape (n,n).
        negLnLike (float): Negative log likelihood of the surface at the specified hyperparameters.
        pen_val (float): Penalty value.

    Examples:
        >>> from spotpython.build.kriging import Kriging
            import numpy as np
            nat_X = np.array([[0], [1]])
            nat_y = np.array([0, 1])
            n=1
            p=1
            S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False)
            S.initialize_variables(nat_X, nat_y)
            S.set_variable_types()
            print(S.cod_X)
            print(S.cod_y)
            S.set_theta_values()
            print(f"S.theta: {S.theta}")
            S.initialize_matrices()
            S.set_de_bounds()
            new_theta_p_Lambda = S.optimize_model()
            S.extract_from_bounds(new_theta_p_Lambda)
            print(f"S.theta: {S.theta}")
            S.build_Psi()
            print(f"S.Psi: {S.Psi}")
            S.build_U()
            print(f"S.U:{S.U}")
            S.likelihood()
            S.negLnLike

    """
    self.extract_from_bounds(new_theta_p_Lambda)
    if self.__is_any__(power(10.0, self.theta), 0):
        logger.warning("Failure in fun_likelihood: 10^theta == 0. Setting negLnLike to %s", self.pen_val)
        return self.pen_val
    self.build_Psi()
    if (self.inf_Psi or self.cnd_Psi > 1e9):
        logger.warning("Failure in fun_likelihood: Psi is ill conditioned: %s", self.cnd_Psi)
        logger.warning("Setting negLnLike to: %s", self.pen_val)
        return self.pen_val

    try:
        self.build_U()
    except Exception as error:
        penalty_value = self.pen_val
        print("Error in fun_likelihood(). Call to build_U() failed.")
        print("error=%s, type(error)=%s" % (error, type(error)))
        print("Setting negLnLike to %.2f." % self.pen_val)
        return penalty_value
    self.likelihood()
    return self.negLnLike

initialize_matrices()

Initialize the matrices for the class instance.

This method initializes several matrices and attributes for the class instance. The p attribute is initialized as a list of ones with length n_p, multiplied by 2.0. The pen_val attribute is initialized as the natural logarithm of the variance of nat_y, multiplied by n, plus 1e4. The negLnLike, LnDetPsi, mu, U, SigmaSqr, and Lambda attributes are all set to None. The gen attribute is initialized using the spacefilling function with arguments k and seed. The Psi attribute is initialized as a zero matrix with shape (n, n) and dtype float64. The psi attribute is initialized as a zero matrix with shape (n, 1). The one attribute is initialized as a list of ones with length n.

Parameters:

Name Type Description Default
self object

The Kriging object.

required

Examples:

>>> from spotpython.build.kriging import Kriging
    import numpy as np
    from numpy import log, var
    nat_X = np.array([[1, 2], [3, 4], [5, 6]])
    nat_y = np.array([1, 2, 3])
    n=3
    p=1
    S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=True)
    S.initialize_variables(nat_X, nat_y)
    S.set_variable_types()
    S.set_theta_values()
    S.initialize_matrices()
    # if var(self.nat_y) is > 0, then self.pen_val = self.n * log(var(self.nat_y)) + 1e4
    # else self.pen_val = self.n * var(self.nat_y) + 1e4
    assert S.pen_val == nat_X.shape[0] * log(var(S.nat_y)) + 1e4
    assert S.Psi.shape == (n, n)

Returns:

Type Description
None

None

Source code in spotpython/build/kriging.py
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
def initialize_matrices(self) -> None:
    """
    Initialize the matrices for the class instance.

    This method initializes several matrices and attributes for the class instance.
    The `p` attribute is initialized as a list of ones with length `n_p`, multiplied by 2.0.
    The `pen_val` attribute is initialized as the natural logarithm of the
    variance of `nat_y`, multiplied by `n`, plus 1e4.
    The `negLnLike`, `LnDetPsi`, `mu`, `U`, `SigmaSqr`, and `Lambda` attributes are all set to None.
    The `gen` attribute is initialized using the `spacefilling` function with arguments `k` and `seed`.
    The `Psi` attribute is initialized as a zero matrix with shape `(n, n)` and dtype `float64`.
    The `psi` attribute is initialized as a zero matrix with shape `(n, 1)`.
    The `one` attribute is initialized as a list of ones with length `n`.

    Args:
        self (object): The Kriging object.

    Examples:
        >>> from spotpython.build.kriging import Kriging
            import numpy as np
            from numpy import log, var
            nat_X = np.array([[1, 2], [3, 4], [5, 6]])
            nat_y = np.array([1, 2, 3])
            n=3
            p=1
            S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=True)
            S.initialize_variables(nat_X, nat_y)
            S.set_variable_types()
            S.set_theta_values()
            S.initialize_matrices()
            # if var(self.nat_y) is > 0, then self.pen_val = self.n * log(var(self.nat_y)) + 1e4
            # else self.pen_val = self.n * var(self.nat_y) + 1e4
            assert S.pen_val == nat_X.shape[0] * log(var(S.nat_y)) + 1e4
            assert S.Psi.shape == (n, n)

    Returns:
        None
    """
    logger.debug("In initialize_matrices(): self.n_p: %s", self.n_p)
    self.p = ones(self.n_p) * self.p_val
    logger.debug("In initialize_matrices(): self.p: %s", self.p)
    # if var(self.nat_y) is > 0, then self.pen_val = self.n * log(var(self.nat_y)) + 1e4
    # else self.pen_val = self.n * var(self.nat_y) + 1e4
    logger.debug("In initialize_matrices(): var(self.nat_y): %s", var(self.nat_y))
    logger.debug("In initialize_matrices(): self.n: %s", self.n)
    if var(self.nat_y) > 0:
        self.pen_val = self.n * log(var(self.nat_y)) + 1e4
    else:
        self.pen_val = self.n * var(self.nat_y) + 1e4
    logger.debug("In initialize_matrices(): self.pen_val: %s", self.pen_val)
    self.negLnLike = None
    logger.debug("In initialize_matrices(): self.k: %s", self.k)
    logger.debug("In initialize_matrices(): self.seed: %s", self.seed)
    self.gen = spacefilling(k=self.k, seed=self.seed)
    logger.debug("In initialize_matrices(): self.gen: %s", self.gen)
    self.LnDetPsi = None
    self.Psi = zeros((self.n, self.n), dtype=float64)
    logger.debug("In initialize_matrices(): self.Psi: %s", self.Psi)
    self.psi = zeros((self.n, 1))
    logger.debug("In initialize_matrices(): self.psi: %s", self.psi)
    self.one = ones(self.n)
    logger.debug("In initialize_matrices(): self.one: %s", self.one)
    self.mu = None
    self.U = None
    self.SigmaSqr = None
    self.Lambda = None

initialize_variables(nat_X, nat_y)

Initialize variables for the class instance.

This method takes in the independent and dependent variable data as input and initializes the class instance variables. It creates deep copies of the input data and stores them in the instance variables nat_X and nat_y. It also calculates the number of observations n and the number of independent variables k from the shape of nat_X. Finally, it creates empty arrays with the same shape as nat_X and nat_y and stores them in the instance variables cod_X and cod_y.

Parameters:

Name Type Description Default
self object

The Kriging object.

required
nat_X ndarray

The independent variable data.

required
nat_y ndarray

The dependent variable data.

required

Returns:

Type Description
None

None

Examples:

>>> from spotpython.build.kriging import Kriging
    import numpy as np
    nat_X = np.array([[1, 2], [3, 4]])
    nat_y = np.array([1, 2])
    S = Kriging()
    S.initialize_variables(nat_X, nat_y)
    print(f"S.nat_X: {S.nat_X}")
    print(f"S.nat_y: {S.nat_y}")
    S.nat_X: [[1 2]
              [3 4]]
    S.nat_y: [1 2]
Source code in spotpython/build/kriging.py
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
def initialize_variables(self, nat_X: np.ndarray, nat_y: np.ndarray) -> None:
    """
    Initialize variables for the class instance.

    This method takes in the independent and dependent variable data as input
    and initializes the class instance variables.
    It creates deep copies of the input data and stores them in the
    instance variables `nat_X` and `nat_y`.
    It also calculates the number of observations `n` and
    the number of independent variables `k` from the shape of `nat_X`.
    Finally, it creates empty arrays with the same shape as `nat_X`
    and `nat_y` and stores them in the instance variables `cod_X` and `cod_y`.

    Args:
        self (object): The Kriging object.
        nat_X (np.ndarray): The independent variable data.
        nat_y (np.ndarray): The dependent variable data.

    Returns:
        None

    Examples:
        >>> from spotpython.build.kriging import Kriging
            import numpy as np
            nat_X = np.array([[1, 2], [3, 4]])
            nat_y = np.array([1, 2])
            S = Kriging()
            S.initialize_variables(nat_X, nat_y)
            print(f"S.nat_X: {S.nat_X}")
            print(f"S.nat_y: {S.nat_y}")
            S.nat_X: [[1 2]
                      [3 4]]
            S.nat_y: [1 2]

    """
    self.nat_X = copy.deepcopy(nat_X)
    self.nat_y = copy.deepcopy(nat_y)
    self.n = self.nat_X.shape[0]
    self.k = self.nat_X.shape[1]

    self.min_X = min(self.nat_X, axis=0)
    self.max_X = max(self.nat_X, axis=0)

    Z = aggregate_mean_var(X=self.nat_X, y=self.nat_y)
    # aggregated y values:
    mu = Z[1]
    self.aggregated_mean_y = np.copy(mu)
    logger.debug("In initialize_variables(): self.nat_X: %s", self.nat_X)
    logger.debug("In initialize_variables(): self.nat_y: %s", self.nat_y)
    logger.debug("In initialize_variables(): self.aggregated_mean_y: %s", self.aggregated_mean_y)
    logger.debug("In initialize_variables(): self.min_X: %s", self.min_X)
    logger.debug("In initialize_variables(): self.max_X: %s", self.max_X)
    logger.debug("In initialize_variables(): self.n: %s", self.n)
    logger.debug("In initialize_variables(): self.k: %s", self.k)

likelihood()

Calculates the negative of the concentrated log-likelihood.

This method implements equation (2.32) in [Forr08a] to calculate the negative of the concentrated log-likelihood. It also modifies mu, SigmaSqr, LnDetPsi, and negLnLike.

Note

build_Psi and build_U should be called first.

Parameters:

Name Type Description Default
self object

The Kriging object.

required

Returns:

Type Description
None

None

Attributes:

Name Type Description
mu float64

Kriging expected mean value mu.

SigmaSqr float64

Sigma squared value.

LnDetPsi float64

Determinant Psi matrix.

negLnLike float

Negative log likelihood of the surface at the specified hyperparameters.

Examples:

>>> from spotpython.build.kriging import Kriging
    import numpy as np
    nat_X = np.array([[1], [2]])
    nat_y = np.array([5, 10])
    n=2
    p=1
    S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False, theta_init_zero=True)
    S.initialize_variables(nat_X, nat_y)
    S.set_variable_types()
    S.set_theta_values()
    S.initialize_matrices()
    S.build_Psi()
    S.build_U()
    S.likelihood()
    # assert S.mu is close to 7.5 with a tolerance of 1e-6
    assert np.allclose(S.mu, 7.5, atol=1e-6)
    E = np.exp(1)
    sigma2 = E/(E**2 -1) * (25/4 + 25/4*E)
    # asssert S.SigmaSqr is close to sigma2 with a tolerance of 1e-6
    assert np.allclose(S.SigmaSqr, sigma2, atol=1e-6)
    print(f"S.LnDetPsi:{S.LnDetPsi}")
    print(f"S.self.negLnLike:{S.negLnLike}")
Source code in spotpython/build/kriging.py
 995
 996
 997
 998
 999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
def likelihood(self) -> None:
    """
    Calculates the negative of the concentrated log-likelihood.

    This method implements equation (2.32) in [Forr08a] to calculate
    the negative of the concentrated log-likelihood. It also modifies `mu`,
    `SigmaSqr`, `LnDetPsi`, and `negLnLike`.

    Note:
        `build_Psi` and `build_U` should be called first.

    Args:
        self (object):
            The Kriging object.

    Returns:
        None

    Attributes:
        mu (np.float64): Kriging expected mean value mu.
        SigmaSqr (np.float64): Sigma squared value.
        LnDetPsi (np.float64): Determinant Psi matrix.
        negLnLike (float): Negative log likelihood of the surface at the specified hyperparameters.

    Examples:
        >>> from spotpython.build.kriging import Kriging
            import numpy as np
            nat_X = np.array([[1], [2]])
            nat_y = np.array([5, 10])
            n=2
            p=1
            S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=False, theta_init_zero=True)
            S.initialize_variables(nat_X, nat_y)
            S.set_variable_types()
            S.set_theta_values()
            S.initialize_matrices()
            S.build_Psi()
            S.build_U()
            S.likelihood()
            # assert S.mu is close to 7.5 with a tolerance of 1e-6
            assert np.allclose(S.mu, 7.5, atol=1e-6)
            E = np.exp(1)
            sigma2 = E/(E**2 -1) * (25/4 + 25/4*E)
            # asssert S.SigmaSqr is close to sigma2 with a tolerance of 1e-6
            assert np.allclose(S.SigmaSqr, sigma2, atol=1e-6)
            print(f"S.LnDetPsi:{S.LnDetPsi}")
            print(f"S.self.negLnLike:{S.negLnLike}")
    """
    # (2.20) in [Forr08a]:
    U_T_inv_one = solve(self.U.T, self.one)
    U_T_inv_cod_y = solve(self.U.T, self.nat_y)
    mu = self.one.T.dot(solve(self.U, U_T_inv_cod_y)) / self.one.T.dot(solve(self.U, U_T_inv_one))
    self.mu = mu
    # (2.31) in [Forr08a]
    cod_y_minus_mu = self.nat_y - self.one.dot(self.mu)
    self.SigmaSqr = cod_y_minus_mu.T.dot(solve(self.U, solve(self.U.T, cod_y_minus_mu))) / self.n
    # (2.32) in [Forr08a]
    self.LnDetPsi = 2.0 * sum(log(abs(diag(self.U))))
    self.negLnLike = -1.0 * (-(self.n / 2.0) * log(self.SigmaSqr) - 0.5 * self.LnDetPsi)

optimize_model()

Optimize the model using the specified model_optimizer.

This method uses the specified model_optimizer to optimize the likelihood function (fun_likelihood) with respect to the model parameters. The optimization is performed within the bounds specified by the attribute de_bounds. The result of the optimization is returned as a list or tuple of optimized parameter values.

Parameters:

Name Type Description Default
self object

The Kriging object.

required

Examples:

>>> from spotpython.build.kriging import Kriging
    import numpy as np
    nat_X = np.array([[1, 2], [3, 4]])
    nat_y = np.array([1, 2])
    n=2
    p=2
    S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=True)
    S.initialize_variables(nat_X, nat_y)
    S.set_variable_types()
    S.set_theta_values()
    S.initialize_matrices()
    S.set_de_bounds()
    new_theta_p_Lambda = S.optimize_model()
    print(new_theta_p_Lambda)

Returns:

Type Description
Union[List[float], Tuple[float]]

result[“x”] (Union[List[float], Tuple[float]]): A list or tuple of optimized parameter values.

Source code in spotpython/build/kriging.py
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
def optimize_model(self) -> Union[List[float], Tuple[float]]:
    """
    Optimize the model using the specified model_optimizer.

    This method uses the specified model_optimizer to optimize the
    likelihood function (`fun_likelihood`) with respect to the model parameters.
    The optimization is performed within the bounds specified by the attribute
    `de_bounds`.
    The result of the optimization is returned as a list or tuple of optimized parameter values.

    Args:
        self (object): The Kriging object.

    Examples:
        >>> from spotpython.build.kriging import Kriging
            import numpy as np
            nat_X = np.array([[1, 2], [3, 4]])
            nat_y = np.array([1, 2])
            n=2
            p=2
            S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=True)
            S.initialize_variables(nat_X, nat_y)
            S.set_variable_types()
            S.set_theta_values()
            S.initialize_matrices()
            S.set_de_bounds()
            new_theta_p_Lambda = S.optimize_model()
            print(new_theta_p_Lambda)

    Returns:
        result["x"] (Union[List[float], Tuple[float]]):
            A list or tuple of optimized parameter values.
    """
    logger.debug("In optimize_model(): self.de_bounds passed to optimizer: %s", self.de_bounds)
    if self.model_optimizer.__name__ == 'dual_annealing':
        result = self.model_optimizer(func=self.fun_likelihood,
                                      bounds=self.de_bounds)
    elif self.model_optimizer.__name__ == 'differential_evolution':
        result = self.model_optimizer(func=self.fun_likelihood,
                                      bounds=self.de_bounds,
                                      maxiter=self.model_fun_evals,
                                      seed=self.seed)
    elif self.model_optimizer.__name__ == 'direct':
        result = self.model_optimizer(func=self.fun_likelihood,
                                      bounds=self.de_bounds,
                                      # maxfun=self.model_fun_evals,
                                      eps=1e-2)
    elif self.model_optimizer.__name__ == 'shgo':
        result = self.model_optimizer(func=self.fun_likelihood,
                                      bounds=self.de_bounds)
    elif self.model_optimizer.__name__ == 'basinhopping':
        result = self.model_optimizer(func=self.fun_likelihood,
                                      x0=mean(self.de_bounds, axis=1))
    else:
        result = self.model_optimizer(func=self.fun_likelihood, bounds=self.de_bounds)
    logger.debug("In optimize_model(): result: %s", result)
    logger.debug('In optimize_model(): returned result["x"]: %s', result["x"])
    return result["x"]

plot(show=True)

This function plots 1D and 2D surrogates.

Parameters:

Name Type Description Default
self object

The Kriging object.

required
show bool

If True, the plots are displayed. If False, plt.show() should be called outside this function.

True

Returns:

Type Description
None

None

Note
  • This method provides only a basic plot. For more advanced plots, use the plot_contour() method of the Spot class.

Examples:

>>> import numpy as np
    from spotpython.fun.objectivefunctions import analytical
    from spotpython.spot import spot
    # 1-dimensional example
    fun = analytical().fun_sphere
    lower = np.array([-1])
    upper = np.array([1])
    design_control={"init_size": 10}
    S = spot.Spot(fun=fun,
                noise=False,
                lower = lower,
                upper= upper,
                design_control=design_control,)
    S.initialize_design()
    S.update_stats()
    S.fit_surrogate()
    S.surrogate.plot()
    # 2-dimensional example
    fun = analytical().fun_sphere
    lower = np.array([-1, -1])
    upper = np.array([1, 1])
    design_control={"init_size": 10}
    S = spot.Spot(fun=fun,
                noise=False,
                lower = lower,
                upper= upper,
                design_control=design_control,)
    S.initialize_design()
    S.update_stats()
    S.fit_surrogate()
    S.surrogate.plot()
Source code in spotpython/build/kriging.py
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
def plot(self, show: Optional[bool] = True) -> None:
    """
    This function plots 1D and 2D surrogates.

    Args:
        self (object):
            The Kriging object.
        show (bool):
            If `True`, the plots are displayed.
            If `False`, `plt.show()` should be called outside this function.

    Returns:
        None

    Note:
        * This method provides only a basic plot. For more advanced plots,
            use the `plot_contour()` method of the `Spot` class.

    Examples:
        >>> import numpy as np
            from spotpython.fun.objectivefunctions import analytical
            from spotpython.spot import spot
            # 1-dimensional example
            fun = analytical().fun_sphere
            lower = np.array([-1])
            upper = np.array([1])
            design_control={"init_size": 10}
            S = spot.Spot(fun=fun,
                        noise=False,
                        lower = lower,
                        upper= upper,
                        design_control=design_control,)
            S.initialize_design()
            S.update_stats()
            S.fit_surrogate()
            S.surrogate.plot()
            # 2-dimensional example
            fun = analytical().fun_sphere
            lower = np.array([-1, -1])
            upper = np.array([1, 1])
            design_control={"init_size": 10}
            S = spot.Spot(fun=fun,
                        noise=False,
                        lower = lower,
                        upper= upper,
                        design_control=design_control,)
            S.initialize_design()
            S.update_stats()
            S.fit_surrogate()
            S.surrogate.plot()
    """
    if self.k == 1:
        # TODO: Improve plot (add conf. interval etc.)
        fig = pylab.figure(figsize=(9, 6))
        n_grid = 100
        x = linspace(
            self.min_X[0], self.max_X[0], num=n_grid
        )
        y = self.predict(x)
        plt.figure()
        plt.plot(x, y, "k")
        if show:
            plt.show()

    if self.k == 2:
        fig = pylab.figure(figsize=(9, 6))
        n_grid = 100
        x = linspace(
            self.min_X[0], self.max_X[0], num=n_grid
        )
        y = linspace(
            self.min_X[1], self.max_X[1], num=n_grid
        )
        X, Y = meshgrid(x, y)
        # Predict based on the optimized results
        zz = array(
            [self.predict(array([x, y]), return_val="all") for x, y in zip(ravel(X), ravel(Y))]
        )
        zs = zz[:, 0, :]
        zse = zz[:, 1, :]
        Z = zs.reshape(X.shape)
        Ze = zse.reshape(X.shape)

        nat_point_X = self.nat_X[:, 0]
        nat_point_Y = self.nat_X[:, 1]
        contour_levels = 30
        ax = fig.add_subplot(224)
        # plot predicted values:
        pylab.contourf(X, Y, Ze, contour_levels, cmap="jet")
        pylab.title("Error")
        pylab.colorbar()
        # plot observed points:
        pylab.plot(nat_point_X, nat_point_Y, "ow")
        #
        ax = fig.add_subplot(223)
        # plot predicted values:
        plt.contourf(X, Y, Z, contour_levels, zorder=1, cmap="jet")
        plt.title("Surrogate")
        # plot observed points:
        pylab.plot(nat_point_X, nat_point_Y, "ow", zorder=3)
        pylab.colorbar()
        #
        ax = fig.add_subplot(221, projection="3d")
        ax.plot_surface(X, Y, Z, rstride=3, cstride=3, alpha=0.9, cmap="jet")
        #
        ax = fig.add_subplot(222, projection="3d")
        ax.plot_surface(X, Y, Ze, rstride=3, cstride=3, alpha=0.9, cmap="jet")
        #
        pylab.show()

predict(nat_X, return_val='y')

This function returns the prediction (in natural units) of the surrogate at the natural coordinates of X.

Parameters:

Name Type Description Default
self object

The Kriging object.

required
nat_X ndarray

Design variable to evaluate in natural units.

required
return_val str

whether y, s, neg. ei (negative expected improvement), or all three values are returned. Default is (for compatibility with sklearn) “y”. To return s, select “s”, to return neg. ei, select “ei”. To return the tuple (y, s, ei), select “all”.

'y'

Returns:

Name Type Description
float Union[float, Tuple[float, float]]

The predicted value in natural units if return_val is “y”.

float Union[float, Tuple[float, float]]

predicted error if return_val is “s”.

float Union[float, Tuple[float, float]]

expected improvement if return_val is “ei”.

Union[float, Tuple[float, float]]

Tuple[float, float, float]: The predicted value in natural units, predicted error and expected improvement if return_val is “all”.

Examples:

>>> from spotpython.build.kriging import Kriging
    import numpy as np
    import matplotlib.pyplot as plt
    from numpy import linspace, arange
    rng = np.random.RandomState(1)
    X = linspace(start=0, stop=10, num=1_0).reshape(-1, 1)
    y = np.squeeze(X * np.sin(X))
    training_indices = rng.choice(arange(y.size), size=6, replace=False)
    X_train, y_train = X[training_indices], y[training_indices]
    S = Kriging(name='kriging', seed=124)
    S.fit(X_train, y_train)
    mean_prediction, std_prediction, s_ei = S.predict(X, return_val="all")
    print(f"mean_prediction: {mean_prediction}")
    print(f"std_prediction: {std_prediction}")
    print(f"s_ei: {s_ei}")
    mean_prediction: [-1.41991225e-08  6.48310037e-01  1.76715565e+00 -6.35226564e-01
                      -4.28585379e+00 -1.22301198e+00  2.49434148e+00  5.61900501e-01
                      -3.04558205e+00 -5.44021104e+00]
    std_prediction: [3.69706811e-04 2.07958787e+00 3.69706810e-04 3.69706807e-04
                    3.69706809e-04 2.07958584e+00 3.69706811e-04 2.60615408e+00
                    2.60837033e+00 3.69706811e-04]
    s_ei: [-0.00000000e+00 -1.02341235e-03 -0.00000000e+00 -0.00000000e+00
           -0.00000000e+00 -1.63799181e-02 -0.00000000e+00 -9.45766290e-03
           -2.53405666e-01 -1.47459347e-04]
Source code in spotpython/build/kriging.py
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
def predict(self, nat_X: ndarray, return_val: str = "y") -> Union[float,
                                                                  Tuple[float, float]]:
    """
    This function returns the prediction (in natural units) of the surrogate at the natural coordinates of X.

    Args:
        self (object):
            The Kriging object.
        nat_X (ndarray):
            Design variable to evaluate in natural units.
        return_val (str):
            whether `y`, `s`, neg. `ei` (negative expected improvement),
            or all three values are returned.
            Default is (for compatibility with sklearn) "y". To return `s`, select "s",
            to return neg. `ei`, select "ei".
            To return the tuple `(y, s, ei)`, select "all".

    Returns:
        float:
            The predicted value in natural units if return_val is "y".
        float:
            predicted error if return_val is "s".
        float:
            expected improvement if return_val is "ei".
        Tuple[float, float, float]:
            The predicted value in natural units, predicted error
            and expected improvement if return_val is "all".

    Examples:
        >>> from spotpython.build.kriging import Kriging
            import numpy as np
            import matplotlib.pyplot as plt
            from numpy import linspace, arange
            rng = np.random.RandomState(1)
            X = linspace(start=0, stop=10, num=1_0).reshape(-1, 1)
            y = np.squeeze(X * np.sin(X))
            training_indices = rng.choice(arange(y.size), size=6, replace=False)
            X_train, y_train = X[training_indices], y[training_indices]
            S = Kriging(name='kriging', seed=124)
            S.fit(X_train, y_train)
            mean_prediction, std_prediction, s_ei = S.predict(X, return_val="all")
            print(f"mean_prediction: {mean_prediction}")
            print(f"std_prediction: {std_prediction}")
            print(f"s_ei: {s_ei}")
            mean_prediction: [-1.41991225e-08  6.48310037e-01  1.76715565e+00 -6.35226564e-01
                              -4.28585379e+00 -1.22301198e+00  2.49434148e+00  5.61900501e-01
                              -3.04558205e+00 -5.44021104e+00]
            std_prediction: [3.69706811e-04 2.07958787e+00 3.69706810e-04 3.69706807e-04
                            3.69706809e-04 2.07958584e+00 3.69706811e-04 2.60615408e+00
                            2.60837033e+00 3.69706811e-04]
            s_ei: [-0.00000000e+00 -1.02341235e-03 -0.00000000e+00 -0.00000000e+00
                   -0.00000000e+00 -1.63799181e-02 -0.00000000e+00 -9.45766290e-03
                   -2.53405666e-01 -1.47459347e-04]

    """
    # Check for the shape and the type of the Input
    if isinstance(nat_X, ndarray):
        try:
            X = nat_X.reshape(-1, self.nat_X.shape[1])
            X = repair_non_numeric(X, self.var_type)
        except Exception:
            raise TypeError("13.1: Input to predict was not convertible to the size of X")
    else:
        raise TypeError(f"type of the given input is an {type(nat_X)} instead of an ndarray")
    n = X.shape[0]
    y = empty(n, dtype=float)
    s = empty(n, dtype=float)
    ei = empty(n, dtype=float)
    for i in range(n):
        x = X[i, :]
        y[i], s[i], ei[i] = self.predict_coded(x)
    if return_val == "y":
        return y
    elif return_val == "s":
        return s
    elif return_val == "ei":
        return -1.0 * ei
    else:
        return y, s, -1.0 * ei

predict_coded(cod_x)

Kriging prediction of one point in the coded units as described in (2.20) in [Forr08a]. The error is returned as well.

Parameters:

Name Type Description Default
self object

The Kriging object.

required
cod_x ndarray

Point in coded units to make prediction at.

required

Returns:

Name Type Description
f float

Predicted value in coded units.

SSqr float

Predicted error.

EI float

Expected improvement.

Note

self.mu and self.SigmaSqr are computed in likelihood, not here. See also [Forr08a, p.60].

Source code in spotpython/build/kriging.py
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
def predict_coded(self, cod_x: np.ndarray) -> Tuple[float, float, float]:
    """
    Kriging prediction of one point in the coded units as described in (2.20) in [Forr08a].
    The error is returned as well.

    Args:
        self (object):
            The Kriging object.
        cod_x (np.ndarray):
            Point in coded units to make prediction at.

    Returns:
        f (float): Predicted value in coded units.
        SSqr (float): Predicted error.
        EI (float): Expected improvement.

    Note:
        `self.mu` and `self.SigmaSqr` are computed in `likelihood`, not here.
        See also [Forr08a, p.60].
    """
    self.build_psi_vec(cod_x)
    U_T_inv = solve(self.U.T, self.nat_y - self.one.dot(self.mu))
    f = self.mu + self.psi.T.dot(solve(self.U, U_T_inv))
    if self.noise:
        Lambda = self.Lambda
    else:
        Lambda = 0.0
    # Error in [Forr08a, p.87]:
    SSqr = self.SigmaSqr * (1 + Lambda - self.psi.T.dot(solve(self.U, solve(self.U.T, self.psi))))
    SSqr = power(abs(SSqr[0]), 0.5)[0]
    EI = self.exp_imp(y0=f[0], s0=SSqr)
    return f[0], SSqr, EI

set_de_bounds()

Determine search bounds for model_optimizer, e.g., differential evolution.

This method sets the attribute de_bounds of the object to a list of lists, where each inner list represents the lower and upper bounds for a parameter being optimized. The number of inner lists is determined by the number of parameters being optimized (n_theta and n_p), as well as whether noise is being considered (noise).

Parameters:

Name Type Description Default
self object

The Kriging object.

required

Examples:

>>> from spotpython.build.kriging import Kriging
    S = Kriging(name='kriging', seed=124)
    S.set_de_bounds()
    print(S.de_bounds)

Returns:

Type Description
None

None

Source code in spotpython/build/kriging.py
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
def set_de_bounds(self) -> None:
    """
    Determine search bounds for model_optimizer, e.g., differential evolution.

    This method sets the attribute `de_bounds` of the object to a list of lists,
    where each inner list represents the lower and upper bounds for a parameter
    being optimized. The number of inner lists is determined by the number of
    parameters being optimized (`n_theta` and `n_p`), as well as whether noise is
    being considered (`noise`).

    Args:
        self (object): The Kriging object.

    Examples:
        >>> from spotpython.build.kriging import Kriging
            S = Kriging(name='kriging', seed=124)
            S.set_de_bounds()
            print(S.de_bounds)

    Returns:
        None
    """
    logger.debug("In set_de_bounds(): self.min_theta: %s", self.min_theta)
    logger.debug("In set_de_bounds(): self.max_theta: %s", self.max_theta)
    logger.debug("In set_de_bounds(): self.n_theta: %s", self.n_theta)
    logger.debug("In set_de_bounds(): self.optim_p: %s", self.optim_p)
    logger.debug("In set_de_bounds(): self.min_p: %s", self.min_p)
    logger.debug("In set_de_bounds(): self.max_p: %s", self.max_p)
    logger.debug("In set_de_bounds(): self.n_p: %s", self.n_p)
    logger.debug("In set_de_bounds(): self.noise: %s", self.noise)
    logger.debug("In set_de_bounds(): self.min_Lambda: %s", self.min_Lambda)
    logger.debug("In set_de_bounds(): self.max_Lambda: %s", self.max_Lambda)

    de_bounds = [[self.min_theta, self.max_theta] for _ in range(self.n_theta)]
    if self.optim_p:
        de_bounds += [[self.min_p, self.max_p] for _ in range(self.n_p)]
        if self.noise:
            de_bounds.append([self.min_Lambda, self.max_Lambda])
    else:
        if self.noise:
            de_bounds.append([self.min_Lambda, self.max_Lambda])
    self.de_bounds = de_bounds
    logger.debug("In set_de_bounds(): self.de_bounds: %s", self.de_bounds)

set_theta_values()

Set the theta values for the class instance.

This method sets the theta values for the class instance based on the n_theta and k attributes. If n_theta is greater than k, n_theta is set to k and a warning is logged. The method then initializes the theta attribute as a list of zeros with length n_theta. The x0_theta attribute is also initialized as a list of ones with length n_theta, multiplied by n / (100 * k).

Parameters:

Name Type Description Default
self object

The Kriging object.

required

Returns: None

Examples:

>>> from spotpython.build.kriging import Kriging
    import numpy as np
    from numpy import array
    nat_X = np.array([[1, 2], [3, 4]])
    nat_y = np.array([1, 2])
    n=2
    p=2
    S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=True)
    S.initialize_variables(nat_X, nat_y)
    S.set_variable_types()
    S.set_theta_values()
    assert S.theta.all() == array([0., 0.]).all()
Source code in spotpython/build/kriging.py
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
def set_theta_values(self) -> None:
    """
    Set the theta values for the class instance.

    This method sets the theta values for the class instance based
    on the `n_theta` and `k` attributes. If `n_theta` is greater than
    `k`, `n_theta` is set to `k` and a warning is logged.
    The method then initializes the `theta` attribute as a list
    of zeros with length `n_theta`.
    The `x0_theta` attribute is also initialized as a list of ones
    with length `n_theta`, multiplied by `n / (100 * k)`.

    Args:
        self (object): The Kriging object.
    Returns:
        None

    Examples:
        >>> from spotpython.build.kriging import Kriging
            import numpy as np
            from numpy import array
            nat_X = np.array([[1, 2], [3, 4]])
            nat_y = np.array([1, 2])
            n=2
            p=2
            S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=True)
            S.initialize_variables(nat_X, nat_y)
            S.set_variable_types()
            S.set_theta_values()
            assert S.theta.all() == array([0., 0.]).all()
    """
    logger.debug("In set_theta_values(): self.k: %s", self.k)
    logger.debug("In set_theta_values(): self.n_theta: %s", self.n_theta)
    if ((self.n_theta > 1) or (self.n_theta > self.k)) and (self.n_theta != self.k):
        self.n_theta = self.k
        logger.warning("Too few theta values or more theta values than dimensions. `n_theta` set to `k`.")
        logger.debug("In set_theta_values(): self.n_theta: %s", self.n_theta)
    if self.theta_init_zero:
        self.theta: List[float] = zeros(self.n_theta)
        logger.debug("In set_theta_values(): self.theta: %s", self.theta)
    else:
        logger.debug("In set_theta_values(): self.n: %s", self.n)
        self.theta: List[float] = ones((self.n_theta,)) * self.n / (100 * self.k)
        logger.debug("In set_theta_values(): self.theta: %s", self.theta)

set_variable_types()

Set the variable types for the class instance.

This method sets the variable types for the class instance based on the var_type attribute. If the length of var_type is less than k, all variable types are forced to ‘num’ and a warning is logged. The method then creates Boolean masks for each variable type (‘num’, ‘factor’, ‘int’, ‘ordered’) using numpy arrays, e.g., num_mask = array([ True, True]) if two numerical variables are present.

Parameters:

Name Type Description Default
self object

The Kriging object.

required

Examples:

>>> from spotpython.build.kriging import Kriging
    nat_X = np.array([[1, 2], [3, 4]])
    nat_y = np.array([1, 2])
    n=2
    p=2
    S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=True)
    S.initialize_variables(nat_X, nat_y)
    S.set_variable_types()
    assert S.var_type == ['num', 'num']
    assert S.var_type == ['num', 'num']
    assert S.num_mask.all() == True
    assert S.factor_mask.all() == False
    assert S.int_mask.all() == False
    assert S.ordered_mask.all() == True

Returns:

Type Description
None

None

Source code in spotpython/build/kriging.py
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
def set_variable_types(self) -> None:
    """
    Set the variable types for the class instance.

    This method sets the variable types for the class instance based
    on the `var_type` attribute. If the length of `var_type` is less
    than `k`, all variable types are forced to 'num' and a warning is logged.
    The method then creates Boolean masks for each variable
    type ('num', 'factor', 'int', 'ordered') using numpy arrays, e.g.,
    `num_mask = array([ True,  True])` if two numerical variables are present.

    Args:
        self (object): The Kriging object.

    Examples:
        >>> from spotpython.build.kriging import Kriging
            nat_X = np.array([[1, 2], [3, 4]])
            nat_y = np.array([1, 2])
            n=2
            p=2
            S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=True)
            S.initialize_variables(nat_X, nat_y)
            S.set_variable_types()
            assert S.var_type == ['num', 'num']
            assert S.var_type == ['num', 'num']
            assert S.num_mask.all() == True
            assert S.factor_mask.all() == False
            assert S.int_mask.all() == False
            assert S.ordered_mask.all() == True

    Returns:
        None
    """
    logger.debug("In set_variable_types(): self.k: %s", self.k)
    logger.debug("In set_variable_types(): self.var_type: %s", self.var_type)
    # assume all variable types are "num" if "num" is
    # specified once:
    if len(self.var_type) < self.k:
        self.var_type = self.var_type * self.k
        logger.warning("In set_variable_types(): All variable types forced to 'num'.")
        logger.debug("In set_variable_types(): self.var_type: %s", self.var_type)
    self.num_mask = np.array(list(map(lambda x: x == "num", self.var_type)))
    self.factor_mask = np.array(list(map(lambda x: x == "factor", self.var_type)))
    self.int_mask = np.array(list(map(lambda x: x == "int", self.var_type)))
    self.ordered_mask = np.array(list(map(lambda x: x == "int" or x == "num" or x == "float", self.var_type)))
    logger.debug("In set_variable_types(): self.num_mask: %s", self.num_mask)
    logger.debug("In set_variable_types(): self.factor_mask: %s", self.factor_mask)
    logger.debug("In set_variable_types(): self.int_mask: %s", self.int_mask)
    logger.debug("In set_variable_types(): self.ordered_mask: %s", self.ordered_mask)

update_log()

Update the log with the current values of negLnLike, theta, p, and Lambda.

This method appends the current values of negLnLike, theta, p (if optim_p is True), and Lambda (if noise is True) to their respective lists in the log dictionary. It also updates the log_length attribute with the current length of the negLnLike list in the log.

If spot_writer is not None, this method also writes the current values of negLnLike, theta, p (if optim_p is True), and Lambda (if noise is True) to the spot_writer object.

Parameters:

Name Type Description Default
self object

The Kriging object.

required

Returns:

Type Description
None

None

Examples:

>>> from spotpython.build.kriging import Kriging
    import numpy as np
    nat_X = np.array([[1, 2], [3, 4]])
    nat_y = np.array([1, 2])
    n=2
    p=2
    S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=True)
    S.initialize_variables(nat_X, nat_y)
    S.set_variable_types()
    S.set_theta_values()
    S.initialize_matrices()
    S.set_de_bounds()
    new_theta_p_Lambda = S.optimize_model()
    S.update_log()
    print(S.log)
    {'negLnLike': array([-1.38629436]),
     'theta': array([-1.14525993,  1.6123372 ]),
      'p': array([1.84444406, 1.74590865]),
      'Lambda': array([0.44268472])}
Source code in spotpython/build/kriging.py
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
def update_log(self) -> None:
    """
    Update the log with the current values of negLnLike, theta, p, and Lambda.

    This method appends the current values of negLnLike, theta, p (if optim_p is True),
    and Lambda (if noise is True)
    to their respective lists in the log dictionary.
    It also updates the log_length attribute with the current length
    of the negLnLike list in the log.

    If spot_writer is not None, this method also writes the current values of
    negLnLike, theta, p (if optim_p is True),
    and Lambda (if noise is True) to the spot_writer object.

    Args:
        self (object): The Kriging object.

    Returns:
        None

    Examples:
        >>> from spotpython.build.kriging import Kriging
            import numpy as np
            nat_X = np.array([[1, 2], [3, 4]])
            nat_y = np.array([1, 2])
            n=2
            p=2
            S=Kriging(name='kriging', seed=124, n_theta=n, n_p=p, optim_p=True, noise=True)
            S.initialize_variables(nat_X, nat_y)
            S.set_variable_types()
            S.set_theta_values()
            S.initialize_matrices()
            S.set_de_bounds()
            new_theta_p_Lambda = S.optimize_model()
            S.update_log()
            print(S.log)
            {'negLnLike': array([-1.38629436]),
             'theta': array([-1.14525993,  1.6123372 ]),
              'p': array([1.84444406, 1.74590865]),
              'Lambda': array([0.44268472])}

    """
    self.log["negLnLike"] = append(self.log["negLnLike"], self.negLnLike)
    self.log["theta"] = append(self.log["theta"], self.theta)
    if self.optim_p:
        self.log["p"] = append(self.log["p"], self.p)
    if self.noise:
        self.log["Lambda"] = append(self.log["Lambda"], self.Lambda)
    # get the length of the log
    self.log_length = len(self.log["negLnLike"])
    if self.spot_writer is not None:
        negLnLike = self.negLnLike.copy()
        self.spot_writer.add_scalar("spot_negLnLike", negLnLike, self.counter+self.log_length)
        # add the self.n_theta theta values to the writer with one key "theta",
        # i.e, the same key for all theta values
        theta = self.theta.copy()
        self.spot_writer.add_scalars("spot_theta", {f"theta_{i}": theta[i] for i in range(self.n_theta)},
                                     self.counter+self.log_length)
        if self.noise:
            Lambda = self.Lambda.copy()
            self.spot_writer.add_scalar("spot_Lambda", Lambda, self.counter+self.log_length)
        if self.optim_p:
            p = self.p.copy()
            self.spot_writer.add_scalars("spot_p", {f"p_{i}": p[i] for i in range(self.n_p)}, self.counter+self.log_length)
        self.spot_writer.flush()

weighted_exp_imp(cod_x, w)

Weighted expected improvement.

Parameters:

Name Type Description Default
self object

The Kriging object.

required
cod_x ndarray

A coded design vector.

required
w float

Weight.

required

Returns:

Name Type Description
EI float

Weighted expected improvement.

References

[Sobester et al. 2005].

Source code in spotpython/build/kriging.py
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
def weighted_exp_imp(self, cod_x: np.ndarray, w: float) -> float:
    """
    Weighted expected improvement.

    Args:
        self (object): The Kriging object.
        cod_x (np.ndarray): A coded design vector.
        w (float): Weight.

    Returns:
        EI (float): Weighted expected improvement.

    References:
        [Sobester et al. 2005].
    """
    y0, s0 = self.predict_coded(cod_x)
    y_min = min(self.nat_y)
    if s0 <= 0.0:
        EI = 0.0
    else:
        y_min_y0 = y_min - y0
        EI_one = w * (
                y_min_y0
                * (0.5 + 0.5 * erf((1.0 / sqrt(2.0)) * (y_min_y0 / s0)))
        )
        EI_two = (
                (1.0 - w)
                * (s0 * (1.0 / sqrt(2.0 * pi)))
                * (exp(-(1.0 / 2.0) * ((y_min_y0) ** 2.0 / s0 ** 2.0)))
        )
        EI = EI_one + EI_two
    return EI