File size: 26,669 Bytes
7885a28 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 |
"""
Binding for libsvm_skl
----------------------
These are the bindings for libsvm_skl, which is a fork of libsvm[1]
that adds to libsvm some capabilities, like index of support vectors
and efficient representation of dense matrices.
These are low-level routines, but can be used for flexibility or
performance reasons. See sklearn.svm for a higher-level API.
Low-level memory management is done in libsvm_helper.c. If we happen
to run out of memory a MemoryError will be raised. In practice this is
not very helpful since high chances are malloc fails inside svm.cpp,
where no sort of memory checks are done.
[1] https://www.csie.ntu.edu.tw/~cjlin/libsvm/
Notes
-----
The signature mode='c' is somewhat superficial, since we already
check that arrays are C-contiguous in svm.py
Authors
-------
2010: Fabian Pedregosa <[email protected]>
Gael Varoquaux <[email protected]>
"""
import numpy as np
from libc.stdlib cimport free
from ..utils._cython_blas cimport _dot
from ..utils._typedefs cimport float64_t, int32_t, intp_t
include "_libsvm.pxi"
cdef extern from *:
ctypedef struct svm_parameter:
pass
################################################################################
# Internal variables
LIBSVM_KERNEL_TYPES = ['linear', 'poly', 'rbf', 'sigmoid', 'precomputed']
################################################################################
# Wrapper functions
def fit(
const float64_t[:, ::1] X,
const float64_t[::1] Y,
int svm_type=0,
kernel='rbf',
int degree=3,
double gamma=0.1,
double coef0=0.0,
double tol=1e-3,
double C=1.0,
double nu=0.5,
double epsilon=0.1,
const float64_t[::1] class_weight=np.empty(0),
const float64_t[::1] sample_weight=np.empty(0),
int shrinking=1,
int probability=0,
double cache_size=100.,
int max_iter=-1,
int random_seed=0,
):
"""
Train the model using libsvm (low-level method)
Parameters
----------
X : array-like, dtype=float64 of shape (n_samples, n_features)
Y : array, dtype=float64 of shape (n_samples,)
target vector
svm_type : {0, 1, 2, 3, 4}, default=0
Type of SVM: C_SVC, NuSVC, OneClassSVM, EpsilonSVR or NuSVR
respectively.
kernel : {'linear', 'rbf', 'poly', 'sigmoid', 'precomputed'}, default="rbf"
Kernel to use in the model: linear, polynomial, RBF, sigmoid
or precomputed.
degree : int32, default=3
Degree of the polynomial kernel (only relevant if kernel is
set to polynomial).
gamma : float64, default=0.1
Gamma parameter in rbf, poly and sigmoid kernels. Ignored by other
kernels.
coef0 : float64, default=0
Independent parameter in poly/sigmoid kernel.
tol : float64, default=1e-3
Numeric stopping criterion (WRITEME).
C : float64, default=1
C parameter in C-Support Vector Classification.
nu : float64, default=0.5
An upper bound on the fraction of training errors and a lower bound of
the fraction of support vectors. Should be in the interval (0, 1].
epsilon : double, default=0.1
Epsilon parameter in the epsilon-insensitive loss function.
class_weight : array, dtype=float64, shape (n_classes,), \
default=np.empty(0)
Set the parameter C of class i to class_weight[i]*C for
SVC. If not given, all classes are supposed to have
weight one.
sample_weight : array, dtype=float64, shape (n_samples,), \
default=np.empty(0)
Weights assigned to each sample.
shrinking : int, default=1
Whether to use the shrinking heuristic.
probability : int, default=0
Whether to enable probability estimates.
cache_size : float64, default=100
Cache size for gram matrix columns (in megabytes).
max_iter : int (-1 for no limit), default=-1
Stop solver after this many iterations regardless of accuracy
(XXX Currently there is no API to know whether this kicked in.)
random_seed : int, default=0
Seed for the random number generator used for probability estimates.
Returns
-------
support : array of shape (n_support,)
Index of support vectors.
support_vectors : array of shape (n_support, n_features)
Support vectors (equivalent to X[support]). Will return an
empty array in the case of precomputed kernel.
n_class_SV : array of shape (n_class,)
Number of support vectors in each class.
sv_coef : array of shape (n_class-1, n_support)
Coefficients of support vectors in decision function.
intercept : array of shape (n_class*(n_class-1)/2,)
Intercept in decision function.
probA, probB : array of shape (n_class*(n_class-1)/2,)
Probability estimates, empty array for probability=False.
n_iter : ndarray of shape (max(1, (n_class * (n_class - 1) // 2)),)
Number of iterations run by the optimization routine to fit the model.
"""
cdef svm_parameter param
cdef svm_problem problem
cdef svm_model *model
cdef const char *error_msg
cdef intp_t SV_len
if len(sample_weight) == 0:
sample_weight = np.ones(X.shape[0], dtype=np.float64)
else:
assert sample_weight.shape[0] == X.shape[0], (
f"sample_weight and X have incompatible shapes: sample_weight has "
f"{sample_weight.shape[0]} samples while X has {X.shape[0]}"
)
kernel_index = LIBSVM_KERNEL_TYPES.index(kernel)
set_problem(
&problem,
<char*> &X[0, 0],
<char*> &Y[0],
<char*> &sample_weight[0],
<intp_t*> X.shape,
kernel_index,
)
if problem.x == NULL:
raise MemoryError("Seems we've run out of memory")
cdef int32_t[::1] class_weight_label = np.arange(
class_weight.shape[0], dtype=np.int32
)
set_parameter(
¶m,
svm_type,
kernel_index,
degree,
gamma,
coef0,
nu,
cache_size,
C,
tol,
epsilon,
shrinking,
probability,
<int> class_weight.shape[0],
<char*> &class_weight_label[0] if class_weight_label.size > 0 else NULL,
<char*> &class_weight[0] if class_weight.size > 0 else NULL,
max_iter,
random_seed,
)
error_msg = svm_check_parameter(&problem, ¶m)
if error_msg:
# for SVR: epsilon is called p in libsvm
error_repl = error_msg.decode('utf-8').replace("p < 0", "epsilon < 0")
raise ValueError(error_repl)
cdef BlasFunctions blas_functions
blas_functions.dot = _dot[double]
# this does the real work
cdef int fit_status = 0
with nogil:
model = svm_train(&problem, ¶m, &fit_status, &blas_functions)
# from here until the end, we just copy the data returned by
# svm_train
SV_len = get_l(model)
n_class = get_nr(model)
cdef int[::1] n_iter = np.empty(max(1, n_class * (n_class - 1) // 2), dtype=np.intc)
copy_n_iter(<char*> &n_iter[0], model)
cdef float64_t[:, ::1] sv_coef = np.empty((n_class-1, SV_len), dtype=np.float64)
copy_sv_coef(<char*> &sv_coef[0, 0] if sv_coef.size > 0 else NULL, model)
# the intercept is just model.rho but with sign changed
cdef float64_t[::1] intercept = np.empty(
int((n_class*(n_class-1))/2), dtype=np.float64
)
copy_intercept(<char*> &intercept[0], model, <intp_t*> intercept.shape)
cdef int32_t[::1] support = np.empty(SV_len, dtype=np.int32)
copy_support(<char*> &support[0] if support.size > 0 else NULL, model)
# copy model.SV
cdef float64_t[:, ::1] support_vectors
if kernel_index == 4:
# precomputed kernel
support_vectors = np.empty((0, 0), dtype=np.float64)
else:
support_vectors = np.empty((SV_len, X.shape[1]), dtype=np.float64)
copy_SV(
<char*> &support_vectors[0, 0] if support_vectors.size > 0 else NULL,
model,
<intp_t*> support_vectors.shape,
)
cdef int32_t[::1] n_class_SV
if svm_type == 0 or svm_type == 1:
n_class_SV = np.empty(n_class, dtype=np.int32)
copy_nSV(<char*> &n_class_SV[0] if n_class_SV.size > 0 else NULL, model)
else:
# OneClass and SVR are considered to have 2 classes
n_class_SV = np.array([SV_len, SV_len], dtype=np.int32)
cdef float64_t[::1] probA
cdef float64_t[::1] probB
if probability != 0:
if svm_type < 2: # SVC and NuSVC
probA = np.empty(int(n_class*(n_class-1)/2), dtype=np.float64)
probB = np.empty(int(n_class*(n_class-1)/2), dtype=np.float64)
copy_probB(<char*> &probB[0], model, <intp_t*> probB.shape)
else:
probA = np.empty(1, dtype=np.float64)
probB = np.empty(0, dtype=np.float64)
copy_probA(<char*> &probA[0], model, <intp_t*> probA.shape)
else:
probA = np.empty(0, dtype=np.float64)
probB = np.empty(0, dtype=np.float64)
svm_free_and_destroy_model(&model)
free(problem.x)
return (
support.base,
support_vectors.base,
n_class_SV.base,
sv_coef.base,
intercept.base,
probA.base,
probB.base,
fit_status,
n_iter.base,
)
cdef void set_predict_params(
svm_parameter *param,
int svm_type,
kernel,
int degree,
double gamma,
double coef0,
double cache_size,
int probability,
int nr_weight,
char *weight_label,
char *weight,
) except *:
"""Fill param with prediction time-only parameters."""
# training-time only parameters
cdef double C = 0.0
cdef double epsilon = 0.1
cdef int max_iter = 0
cdef double nu = 0.5
cdef int shrinking = 0
cdef double tol = 0.1
cdef int random_seed = -1
kernel_index = LIBSVM_KERNEL_TYPES.index(kernel)
set_parameter(
param,
svm_type,
kernel_index,
degree,
gamma,
coef0,
nu,
cache_size,
C,
tol,
epsilon,
shrinking,
probability,
nr_weight,
weight_label,
weight,
max_iter,
random_seed,
)
def predict(
const float64_t[:, ::1] X,
const int32_t[::1] support,
const float64_t[:, ::1] SV,
const int32_t[::1] nSV,
const float64_t[:, ::1] sv_coef,
const float64_t[::1] intercept,
const float64_t[::1] probA=np.empty(0),
const float64_t[::1] probB=np.empty(0),
int svm_type=0,
kernel='rbf',
int degree=3,
double gamma=0.1,
double coef0=0.0,
const float64_t[::1] class_weight=np.empty(0),
const float64_t[::1] sample_weight=np.empty(0),
double cache_size=100.0,
):
"""
Predict target values of X given a model (low-level method)
Parameters
----------
X : array-like, dtype=float of shape (n_samples, n_features)
support : array of shape (n_support,)
Index of support vectors in training set.
SV : array of shape (n_support, n_features)
Support vectors.
nSV : array of shape (n_class,)
Number of support vectors in each class.
sv_coef : array of shape (n_class-1, n_support)
Coefficients of support vectors in decision function.
intercept : array of shape (n_class*(n_class-1)/2)
Intercept in decision function.
probA, probB : array of shape (n_class*(n_class-1)/2,)
Probability estimates.
svm_type : {0, 1, 2, 3, 4}, default=0
Type of SVM: C_SVC, NuSVC, OneClassSVM, EpsilonSVR or NuSVR
respectively.
kernel : {'linear', 'rbf', 'poly', 'sigmoid', 'precomputed'}, default="rbf"
Kernel to use in the model: linear, polynomial, RBF, sigmoid
or precomputed.
degree : int32, default=3
Degree of the polynomial kernel (only relevant if kernel is
set to polynomial).
gamma : float64, default=0.1
Gamma parameter in rbf, poly and sigmoid kernels. Ignored by other
kernels.
coef0 : float64, default=0.0
Independent parameter in poly/sigmoid kernel.
Returns
-------
dec_values : array
Predicted values.
"""
cdef float64_t[::1] dec_values
cdef svm_parameter param
cdef svm_model *model
cdef int rv
cdef int32_t[::1] class_weight_label = np.arange(
class_weight.shape[0], dtype=np.int32
)
set_predict_params(
¶m,
svm_type,
kernel,
degree,
gamma,
coef0,
cache_size,
0,
<int>class_weight.shape[0],
<char*> &class_weight_label[0] if class_weight_label.size > 0 else NULL,
<char*> &class_weight[0] if class_weight.size > 0 else NULL,
)
model = set_model(
¶m,
<int> nSV.shape[0],
<char*> &SV[0, 0] if SV.size > 0 else NULL,
<intp_t*> SV.shape,
<char*> &support[0] if support.size > 0 else NULL,
<intp_t*> support.shape,
<intp_t*> sv_coef.strides,
<char*> &sv_coef[0, 0] if sv_coef.size > 0 else NULL,
<char*> &intercept[0],
<char*> &nSV[0],
<char*> &probA[0] if probA.size > 0 else NULL,
<char*> &probB[0] if probB.size > 0 else NULL,
)
cdef BlasFunctions blas_functions
blas_functions.dot = _dot[double]
# TODO: use check_model
try:
dec_values = np.empty(X.shape[0])
with nogil:
rv = copy_predict(
<char*> &X[0, 0],
model,
<intp_t*> X.shape,
<char*> &dec_values[0],
&blas_functions,
)
if rv < 0:
raise MemoryError("We've run out of memory")
finally:
free_model(model)
return dec_values.base
def predict_proba(
const float64_t[:, ::1] X,
const int32_t[::1] support,
const float64_t[:, ::1] SV,
const int32_t[::1] nSV,
float64_t[:, ::1] sv_coef,
float64_t[::1] intercept,
float64_t[::1] probA=np.empty(0),
float64_t[::1] probB=np.empty(0),
int svm_type=0,
kernel='rbf',
int degree=3,
double gamma=0.1,
double coef0=0.0,
float64_t[::1] class_weight=np.empty(0),
float64_t[::1] sample_weight=np.empty(0),
double cache_size=100.0,
):
"""
Predict probabilities
svm_model stores all parameters needed to predict a given value.
For speed, all real work is done at the C level in function
copy_predict (libsvm_helper.c).
We have to reconstruct model and parameters to make sure we stay
in sync with the python object.
See sklearn.svm.predict for a complete list of parameters.
Parameters
----------
X : array-like, dtype=float of shape (n_samples, n_features)
support : array of shape (n_support,)
Index of support vectors in training set.
SV : array of shape (n_support, n_features)
Support vectors.
nSV : array of shape (n_class,)
Number of support vectors in each class.
sv_coef : array of shape (n_class-1, n_support)
Coefficients of support vectors in decision function.
intercept : array of shape (n_class*(n_class-1)/2,)
Intercept in decision function.
probA, probB : array of shape (n_class*(n_class-1)/2,)
Probability estimates.
svm_type : {0, 1, 2, 3, 4}, default=0
Type of SVM: C_SVC, NuSVC, OneClassSVM, EpsilonSVR or NuSVR
respectively.
kernel : {'linear', 'rbf', 'poly', 'sigmoid', 'precomputed'}, default="rbf"
Kernel to use in the model: linear, polynomial, RBF, sigmoid
or precomputed.
degree : int32, default=3
Degree of the polynomial kernel (only relevant if kernel is
set to polynomial).
gamma : float64, default=0.1
Gamma parameter in rbf, poly and sigmoid kernels. Ignored by other
kernels.
coef0 : float64, default=0.0
Independent parameter in poly/sigmoid kernel.
Returns
-------
dec_values : array
Predicted values.
"""
cdef float64_t[:, ::1] dec_values
cdef svm_parameter param
cdef svm_model *model
cdef int32_t[::1] class_weight_label = np.arange(
class_weight.shape[0], dtype=np.int32
)
cdef int rv
set_predict_params(
¶m,
svm_type,
kernel,
degree,
gamma,
coef0,
cache_size,
1,
<int> class_weight.shape[0],
<char*> &class_weight_label[0] if class_weight_label.size > 0 else NULL,
<char*> &class_weight[0] if class_weight.size > 0 else NULL,
)
model = set_model(
¶m,
<int> nSV.shape[0],
<char*> &SV[0, 0] if SV.size > 0 else NULL,
<intp_t*> SV.shape,
<char*> &support[0],
<intp_t*> support.shape,
<intp_t*> sv_coef.strides,
<char*> &sv_coef[0, 0],
<char*> &intercept[0],
<char*> &nSV[0],
<char*> &probA[0] if probA.size > 0 else NULL,
<char*> &probB[0] if probB.size > 0 else NULL,
)
cdef intp_t n_class = get_nr(model)
cdef BlasFunctions blas_functions
blas_functions.dot = _dot[double]
try:
dec_values = np.empty((X.shape[0], n_class), dtype=np.float64)
with nogil:
rv = copy_predict_proba(
<char*> &X[0, 0],
model,
<intp_t*> X.shape,
<char*> &dec_values[0, 0],
&blas_functions,
)
if rv < 0:
raise MemoryError("We've run out of memory")
finally:
free_model(model)
return dec_values.base
def decision_function(
const float64_t[:, ::1] X,
const int32_t[::1] support,
const float64_t[:, ::1] SV,
const int32_t[::1] nSV,
const float64_t[:, ::1] sv_coef,
const float64_t[::1] intercept,
const float64_t[::1] probA=np.empty(0),
const float64_t[::1] probB=np.empty(0),
int svm_type=0,
kernel='rbf',
int degree=3,
double gamma=0.1,
double coef0=0.0,
const float64_t[::1] class_weight=np.empty(0),
const float64_t[::1] sample_weight=np.empty(0),
double cache_size=100.0,
):
"""
Predict margin (libsvm name for this is predict_values)
We have to reconstruct model and parameters to make sure we stay
in sync with the python object.
Parameters
----------
X : array-like, dtype=float, size=[n_samples, n_features]
support : array, shape=[n_support]
Index of support vectors in training set.
SV : array, shape=[n_support, n_features]
Support vectors.
nSV : array, shape=[n_class]
Number of support vectors in each class.
sv_coef : array, shape=[n_class-1, n_support]
Coefficients of support vectors in decision function.
intercept : array, shape=[n_class*(n_class-1)/2]
Intercept in decision function.
probA, probB : array, shape=[n_class*(n_class-1)/2]
Probability estimates.
svm_type : {0, 1, 2, 3, 4}, optional
Type of SVM: C_SVC, NuSVC, OneClassSVM, EpsilonSVR or NuSVR
respectively. 0 by default.
kernel : {'linear', 'rbf', 'poly', 'sigmoid', 'precomputed'}, optional
Kernel to use in the model: linear, polynomial, RBF, sigmoid
or precomputed. 'rbf' by default.
degree : int32, optional
Degree of the polynomial kernel (only relevant if kernel is
set to polynomial), 3 by default.
gamma : float64, optional
Gamma parameter in rbf, poly and sigmoid kernels. Ignored by other
kernels. 0.1 by default.
coef0 : float64, optional
Independent parameter in poly/sigmoid kernel. 0 by default.
Returns
-------
dec_values : array
Predicted values.
"""
cdef float64_t[:, ::1] dec_values
cdef svm_parameter param
cdef svm_model *model
cdef intp_t n_class
cdef int32_t[::1] class_weight_label = np.arange(
class_weight.shape[0], dtype=np.int32
)
cdef int rv
set_predict_params(
¶m,
svm_type,
kernel,
degree,
gamma,
coef0,
cache_size,
0,
<int> class_weight.shape[0],
<char*> &class_weight_label[0] if class_weight_label.size > 0 else NULL,
<char*> &class_weight[0] if class_weight.size > 0 else NULL,
)
model = set_model(
¶m,
<int> nSV.shape[0],
<char*> &SV[0, 0] if SV.size > 0 else NULL,
<intp_t*> SV.shape,
<char*> &support[0],
<intp_t*> support.shape,
<intp_t*> sv_coef.strides,
<char*> &sv_coef[0, 0],
<char*> &intercept[0],
<char*> &nSV[0],
<char*> &probA[0] if probA.size > 0 else NULL,
<char*> &probB[0] if probB.size > 0 else NULL,
)
if svm_type > 1:
n_class = 1
else:
n_class = get_nr(model)
n_class = n_class * (n_class - 1) // 2
cdef BlasFunctions blas_functions
blas_functions.dot = _dot[double]
try:
dec_values = np.empty((X.shape[0], n_class), dtype=np.float64)
with nogil:
rv = copy_predict_values(
<char*> &X[0, 0],
model,
<intp_t*> X.shape,
<char*> &dec_values[0, 0],
n_class,
&blas_functions,
)
if rv < 0:
raise MemoryError("We've run out of memory")
finally:
free_model(model)
return dec_values.base
def cross_validation(
const float64_t[:, ::1] X,
const float64_t[::1] Y,
int n_fold,
int svm_type=0,
kernel='rbf',
int degree=3,
double gamma=0.1,
double coef0=0.0,
double tol=1e-3,
double C=1.0,
double nu=0.5,
double epsilon=0.1,
float64_t[::1] class_weight=np.empty(0),
float64_t[::1] sample_weight=np.empty(0),
int shrinking=0,
int probability=0,
double cache_size=100.0,
int max_iter=-1,
int random_seed=0,
):
"""
Binding of the cross-validation routine (low-level routine)
Parameters
----------
X : array-like, dtype=float of shape (n_samples, n_features)
Y : array, dtype=float of shape (n_samples,)
target vector
n_fold : int32
Number of folds for cross validation.
svm_type : {0, 1, 2, 3, 4}, default=0
Type of SVM: C_SVC, NuSVC, OneClassSVM, EpsilonSVR or NuSVR
respectively.
kernel : {'linear', 'rbf', 'poly', 'sigmoid', 'precomputed'}, default='rbf'
Kernel to use in the model: linear, polynomial, RBF, sigmoid
or precomputed.
degree : int32, default=3
Degree of the polynomial kernel (only relevant if kernel is
set to polynomial).
gamma : float64, default=0.1
Gamma parameter in rbf, poly and sigmoid kernels. Ignored by other
kernels.
coef0 : float64, default=0.0
Independent parameter in poly/sigmoid kernel.
tol : float64, default=1e-3
Numeric stopping criterion (WRITEME).
C : float64, default=1
C parameter in C-Support Vector Classification.
nu : float64, default=0.5
An upper bound on the fraction of training errors and a lower bound of
the fraction of support vectors. Should be in the interval (0, 1].
epsilon : double, default=0.1
Epsilon parameter in the epsilon-insensitive loss function.
class_weight : array, dtype=float64, shape (n_classes,), \
default=np.empty(0)
Set the parameter C of class i to class_weight[i]*C for
SVC. If not given, all classes are supposed to have
weight one.
sample_weight : array, dtype=float64, shape (n_samples,), \
default=np.empty(0)
Weights assigned to each sample.
shrinking : int, default=1
Whether to use the shrinking heuristic.
probability : int, default=0
Whether to enable probability estimates.
cache_size : float64, default=100
Cache size for gram matrix columns (in megabytes).
max_iter : int (-1 for no limit), default=-1
Stop solver after this many iterations regardless of accuracy
(XXX Currently there is no API to know whether this kicked in.)
random_seed : int, default=0
Seed for the random number generator used for probability estimates.
Returns
-------
target : array, float
"""
cdef svm_parameter param
cdef svm_problem problem
cdef const char *error_msg
if len(sample_weight) == 0:
sample_weight = np.ones(X.shape[0], dtype=np.float64)
else:
assert sample_weight.shape[0] == X.shape[0], (
f"sample_weight and X have incompatible shapes: sample_weight has "
f"{sample_weight.shape[0]} samples while X has {X.shape[0]}"
)
if X.shape[0] < n_fold:
raise ValueError("Number of samples is less than number of folds")
# set problem
kernel_index = LIBSVM_KERNEL_TYPES.index(kernel)
set_problem(
&problem,
<char*> &X[0, 0],
<char*> &Y[0],
<char*> &sample_weight[0] if sample_weight.size > 0 else NULL,
<intp_t*> X.shape,
kernel_index,
)
if problem.x == NULL:
raise MemoryError("Seems we've run out of memory")
cdef int32_t[::1] class_weight_label = np.arange(
class_weight.shape[0], dtype=np.int32
)
# set parameters
set_parameter(
¶m,
svm_type,
kernel_index,
degree,
gamma,
coef0,
nu,
cache_size,
C,
tol,
tol,
shrinking,
probability,
<int> class_weight.shape[0],
<char*> &class_weight_label[0] if class_weight_label.size > 0 else NULL,
<char*> &class_weight[0] if class_weight.size > 0 else NULL,
max_iter,
random_seed,
)
error_msg = svm_check_parameter(&problem, ¶m)
if error_msg:
raise ValueError(error_msg)
cdef float64_t[::1] target
cdef BlasFunctions blas_functions
blas_functions.dot = _dot[double]
try:
target = np.empty((X.shape[0]), dtype=np.float64)
with nogil:
svm_cross_validation(
&problem,
¶m,
n_fold,
<double *> &target[0],
&blas_functions,
)
finally:
free(problem.x)
return target.base
def set_verbosity_wrap(int verbosity):
"""
Control verbosity of libsvm library
"""
set_verbosity(verbosity)
|