/usr/lib/python3/dist-packages/sklearn/decomposition/dict_learning.py is in python3-sklearn 0.17.0-4.
This file is owned by root:root, with mode 0o644.
The actual contents of the file can be viewed below.
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 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 | """ Dictionary learning
"""
from __future__ import print_function
# Author: Vlad Niculae, Gael Varoquaux, Alexandre Gramfort
# License: BSD 3 clause
import time
import sys
import itertools
from math import sqrt, ceil
import numpy as np
from scipy import linalg
from numpy.lib.stride_tricks import as_strided
from ..base import BaseEstimator, TransformerMixin
from ..externals.joblib import Parallel, delayed, cpu_count
from ..externals.six.moves import zip
from ..utils import (check_array, check_random_state, gen_even_slices,
gen_batches, _get_n_jobs)
from ..utils.extmath import randomized_svd, row_norms
from ..utils.validation import check_is_fitted
from ..linear_model import Lasso, orthogonal_mp_gram, LassoLars, Lars
def _sparse_encode(X, dictionary, gram, cov=None, algorithm='lasso_lars',
regularization=None, copy_cov=True,
init=None, max_iter=1000, check_input=True, verbose=0):
"""Generic sparse coding
Each column of the result is the solution to a Lasso problem.
Parameters
----------
X: array of shape (n_samples, n_features)
Data matrix.
dictionary: array of shape (n_components, n_features)
The dictionary matrix against which to solve the sparse coding of
the data. Some of the algorithms assume normalized rows.
gram: None | array, shape=(n_components, n_components)
Precomputed Gram matrix, dictionary * dictionary'
gram can be None if method is 'threshold'.
cov: array, shape=(n_components, n_samples)
Precomputed covariance, dictionary * X'
algorithm: {'lasso_lars', 'lasso_cd', 'lars', 'omp', 'threshold'}
lars: uses the least angle regression method (linear_model.lars_path)
lasso_lars: uses Lars to compute the Lasso solution
lasso_cd: uses the coordinate descent method to compute the
Lasso solution (linear_model.Lasso). lasso_lars will be faster if
the estimated components are sparse.
omp: uses orthogonal matching pursuit to estimate the sparse solution
threshold: squashes to zero all coefficients less than regularization
from the projection dictionary * data'
regularization : int | float
The regularization parameter. It corresponds to alpha when
algorithm is 'lasso_lars', 'lasso_cd' or 'threshold'.
Otherwise it corresponds to n_nonzero_coefs.
init: array of shape (n_samples, n_components)
Initialization value of the sparse code. Only used if
`algorithm='lasso_cd'`.
max_iter: int, 1000 by default
Maximum number of iterations to perform if `algorithm='lasso_cd'`.
copy_cov: boolean, optional
Whether to copy the precomputed covariance matrix; if False, it may be
overwritten.
check_input: boolean, optional
If False, the input arrays X and dictionary will not be checked.
verbose: int
Controls the verbosity; the higher, the more messages. Defaults to 0.
Returns
-------
code: array of shape (n_components, n_features)
The sparse codes
See also
--------
sklearn.linear_model.lars_path
sklearn.linear_model.orthogonal_mp
sklearn.linear_model.Lasso
SparseCoder
"""
if X.ndim == 1:
X = X[:, np.newaxis]
n_samples, n_features = X.shape
if cov is None and algorithm != 'lasso_cd':
# overwriting cov is safe
copy_cov = False
cov = np.dot(dictionary, X.T)
if algorithm == 'lasso_lars':
alpha = float(regularization) / n_features # account for scaling
try:
err_mgt = np.seterr(all='ignore')
# Not passing in verbose=max(0, verbose-1) because Lars.fit already
# corrects the verbosity level.
lasso_lars = LassoLars(alpha=alpha, fit_intercept=False,
verbose=verbose, normalize=False,
precompute=gram, fit_path=False)
lasso_lars.fit(dictionary.T, X.T, Xy=cov)
new_code = lasso_lars.coef_
finally:
np.seterr(**err_mgt)
elif algorithm == 'lasso_cd':
alpha = float(regularization) / n_features # account for scaling
# TODO: Make verbosity argument for Lasso?
# sklearn.linear_model.coordinate_descent.enet_path has a verbosity
# argument that we could pass in from Lasso.
clf = Lasso(alpha=alpha, fit_intercept=False, normalize=False,
precompute=gram, max_iter=max_iter, warm_start=True)
clf.coef_ = init
clf.fit(dictionary.T, X.T, check_input=check_input)
new_code = clf.coef_
elif algorithm == 'lars':
try:
err_mgt = np.seterr(all='ignore')
# Not passing in verbose=max(0, verbose-1) because Lars.fit already
# corrects the verbosity level.
lars = Lars(fit_intercept=False, verbose=verbose, normalize=False,
precompute=gram, n_nonzero_coefs=int(regularization),
fit_path=False)
lars.fit(dictionary.T, X.T, Xy=cov)
new_code = lars.coef_
finally:
np.seterr(**err_mgt)
elif algorithm == 'threshold':
new_code = ((np.sign(cov) *
np.maximum(np.abs(cov) - regularization, 0)).T)
elif algorithm == 'omp':
# TODO: Should verbose argument be passed to this?
new_code = orthogonal_mp_gram(
Gram=gram, Xy=cov, n_nonzero_coefs=int(regularization),
tol=None, norms_squared=row_norms(X, squared=True),
copy_Xy=copy_cov).T
else:
raise ValueError('Sparse coding method must be "lasso_lars" '
'"lasso_cd", "lasso", "threshold" or "omp", got %s.'
% algorithm)
return new_code
# XXX : could be moved to the linear_model module
def sparse_encode(X, dictionary, gram=None, cov=None, algorithm='lasso_lars',
n_nonzero_coefs=None, alpha=None, copy_cov=True, init=None,
max_iter=1000, n_jobs=1, check_input=True, verbose=0):
"""Sparse coding
Each row of the result is the solution to a sparse coding problem.
The goal is to find a sparse array `code` such that::
X ~= code * dictionary
Read more in the :ref:`User Guide <SparseCoder>`.
Parameters
----------
X: array of shape (n_samples, n_features)
Data matrix
dictionary: array of shape (n_components, n_features)
The dictionary matrix against which to solve the sparse coding of
the data. Some of the algorithms assume normalized rows for meaningful
output.
gram: array, shape=(n_components, n_components)
Precomputed Gram matrix, dictionary * dictionary'
cov: array, shape=(n_components, n_samples)
Precomputed covariance, dictionary' * X
algorithm: {'lasso_lars', 'lasso_cd', 'lars', 'omp', 'threshold'}
lars: uses the least angle regression method (linear_model.lars_path)
lasso_lars: uses Lars to compute the Lasso solution
lasso_cd: uses the coordinate descent method to compute the
Lasso solution (linear_model.Lasso). lasso_lars will be faster if
the estimated components are sparse.
omp: uses orthogonal matching pursuit to estimate the sparse solution
threshold: squashes to zero all coefficients less than alpha from
the projection dictionary * X'
n_nonzero_coefs: int, 0.1 * n_features by default
Number of nonzero coefficients to target in each column of the
solution. This is only used by `algorithm='lars'` and `algorithm='omp'`
and is overridden by `alpha` in the `omp` case.
alpha: float, 1. by default
If `algorithm='lasso_lars'` or `algorithm='lasso_cd'`, `alpha` is the
penalty applied to the L1 norm.
If `algorithm='threhold'`, `alpha` is the absolute value of the
threshold below which coefficients will be squashed to zero.
If `algorithm='omp'`, `alpha` is the tolerance parameter: the value of
the reconstruction error targeted. In this case, it overrides
`n_nonzero_coefs`.
init: array of shape (n_samples, n_components)
Initialization value of the sparse codes. Only used if
`algorithm='lasso_cd'`.
max_iter: int, 1000 by default
Maximum number of iterations to perform if `algorithm='lasso_cd'`.
copy_cov: boolean, optional
Whether to copy the precomputed covariance matrix; if False, it may be
overwritten.
n_jobs: int, optional
Number of parallel jobs to run.
check_input: boolean, optional
If False, the input arrays X and dictionary will not be checked.
verbose : int, optional
Controls the verbosity; the higher, the more messages. Defaults to 0.
Returns
-------
code: array of shape (n_samples, n_components)
The sparse codes
See also
--------
sklearn.linear_model.lars_path
sklearn.linear_model.orthogonal_mp
sklearn.linear_model.Lasso
SparseCoder
"""
if check_input:
if algorithm == 'lasso_cd':
dictionary = check_array(dictionary, order='C', dtype='float64')
X = check_array(X, order='C', dtype='float64')
else:
dictionary = check_array(dictionary)
X = check_array(X)
n_samples, n_features = X.shape
n_components = dictionary.shape[0]
if gram is None and algorithm != 'threshold':
gram = np.dot(dictionary, dictionary.T)
if cov is None and algorithm != 'lasso_cd':
copy_cov = False
cov = np.dot(dictionary, X.T)
if algorithm in ('lars', 'omp'):
regularization = n_nonzero_coefs
if regularization is None:
regularization = min(max(n_features / 10, 1), n_components)
else:
regularization = alpha
if regularization is None:
regularization = 1.
if n_jobs == 1 or algorithm == 'threshold':
code = _sparse_encode(X,
dictionary, gram, cov=cov,
algorithm=algorithm,
regularization=regularization, copy_cov=copy_cov,
init=init,
max_iter=max_iter,
check_input=False,
verbose=verbose)
# This ensure that dimensionality of code is always 2,
# consistant with the case n_jobs > 1
if code.ndim == 1:
code = code[np.newaxis, :]
return code
# Enter parallel code block
code = np.empty((n_samples, n_components))
slices = list(gen_even_slices(n_samples, _get_n_jobs(n_jobs)))
code_views = Parallel(n_jobs=n_jobs, verbose=verbose)(
delayed(_sparse_encode)(
X[this_slice], dictionary, gram,
cov[:, this_slice] if cov is not None else None,
algorithm,
regularization=regularization, copy_cov=copy_cov,
init=init[this_slice] if init is not None else None,
max_iter=max_iter,
check_input=False)
for this_slice in slices)
for this_slice, this_view in zip(slices, code_views):
code[this_slice] = this_view
return code
def _update_dict(dictionary, Y, code, verbose=False, return_r2=False,
random_state=None):
"""Update the dense dictionary factor in place.
Parameters
----------
dictionary: array of shape (n_features, n_components)
Value of the dictionary at the previous iteration.
Y: array of shape (n_features, n_samples)
Data matrix.
code: array of shape (n_components, n_samples)
Sparse coding of the data against which to optimize the dictionary.
verbose:
Degree of output the procedure will print.
return_r2: bool
Whether to compute and return the residual sum of squares corresponding
to the computed solution.
random_state: int or RandomState
Pseudo number generator state used for random sampling.
Returns
-------
dictionary: array of shape (n_features, n_components)
Updated dictionary.
"""
n_components = len(code)
n_samples = Y.shape[0]
random_state = check_random_state(random_state)
# Residuals, computed 'in-place' for efficiency
R = -np.dot(dictionary, code)
R += Y
R = np.asfortranarray(R)
ger, = linalg.get_blas_funcs(('ger',), (dictionary, code))
for k in range(n_components):
# R <- 1.0 * U_k * V_k^T + R
R = ger(1.0, dictionary[:, k], code[k, :], a=R, overwrite_a=True)
dictionary[:, k] = np.dot(R, code[k, :].T)
# Scale k'th atom
atom_norm_square = np.dot(dictionary[:, k], dictionary[:, k])
if atom_norm_square < 1e-20:
if verbose == 1:
sys.stdout.write("+")
sys.stdout.flush()
elif verbose:
print("Adding new random atom")
dictionary[:, k] = random_state.randn(n_samples)
# Setting corresponding coefs to 0
code[k, :] = 0.0
dictionary[:, k] /= sqrt(np.dot(dictionary[:, k],
dictionary[:, k]))
else:
dictionary[:, k] /= sqrt(atom_norm_square)
# R <- -1.0 * U_k * V_k^T + R
R = ger(-1.0, dictionary[:, k], code[k, :], a=R, overwrite_a=True)
if return_r2:
R **= 2
# R is fortran-ordered. For numpy version < 1.6, sum does not
# follow the quick striding first, and is thus inefficient on
# fortran ordered data. We take a flat view of the data with no
# striding
R = as_strided(R, shape=(R.size, ), strides=(R.dtype.itemsize,))
R = np.sum(R)
return dictionary, R
return dictionary
def dict_learning(X, n_components, alpha, max_iter=100, tol=1e-8,
method='lars', n_jobs=1, dict_init=None, code_init=None,
callback=None, verbose=False, random_state=None,
return_n_iter=False):
"""Solves a dictionary learning matrix factorization problem.
Finds the best dictionary and the corresponding sparse code for
approximating the data matrix X by solving::
(U^*, V^*) = argmin 0.5 || X - U V ||_2^2 + alpha * || U ||_1
(U,V)
with || V_k ||_2 = 1 for all 0 <= k < n_components
where V is the dictionary and U is the sparse code.
Read more in the :ref:`User Guide <DictionaryLearning>`.
Parameters
----------
X: array of shape (n_samples, n_features)
Data matrix.
n_components: int,
Number of dictionary atoms to extract.
alpha: int,
Sparsity controlling parameter.
max_iter: int,
Maximum number of iterations to perform.
tol: float,
Tolerance for the stopping condition.
method: {'lars', 'cd'}
lars: uses the least angle regression method to solve the lasso problem
(linear_model.lars_path)
cd: uses the coordinate descent method to compute the
Lasso solution (linear_model.Lasso). Lars will be faster if
the estimated components are sparse.
n_jobs: int,
Number of parallel jobs to run, or -1 to autodetect.
dict_init: array of shape (n_components, n_features),
Initial value for the dictionary for warm restart scenarios.
code_init: array of shape (n_samples, n_components),
Initial value for the sparse code for warm restart scenarios.
callback:
Callable that gets invoked every five iterations.
verbose:
Degree of output the procedure will print.
random_state: int or RandomState
Pseudo number generator state used for random sampling.
return_n_iter : bool
Whether or not to return the number of iterations.
Returns
-------
code: array of shape (n_samples, n_components)
The sparse code factor in the matrix factorization.
dictionary: array of shape (n_components, n_features),
The dictionary factor in the matrix factorization.
errors: array
Vector of errors at each iteration.
n_iter : int
Number of iterations run. Returned only if `return_n_iter` is
set to True.
See also
--------
dict_learning_online
DictionaryLearning
MiniBatchDictionaryLearning
SparsePCA
MiniBatchSparsePCA
"""
if method not in ('lars', 'cd'):
raise ValueError('Coding method %r not supported as a fit algorithm.'
% method)
method = 'lasso_' + method
t0 = time.time()
# Avoid integer division problems
alpha = float(alpha)
random_state = check_random_state(random_state)
if n_jobs == -1:
n_jobs = cpu_count()
# Init the code and the dictionary with SVD of Y
if code_init is not None and dict_init is not None:
code = np.array(code_init, order='F')
# Don't copy V, it will happen below
dictionary = dict_init
else:
code, S, dictionary = linalg.svd(X, full_matrices=False)
dictionary = S[:, np.newaxis] * dictionary
r = len(dictionary)
if n_components <= r: # True even if n_components=None
code = code[:, :n_components]
dictionary = dictionary[:n_components, :]
else:
code = np.c_[code, np.zeros((len(code), n_components - r))]
dictionary = np.r_[dictionary,
np.zeros((n_components - r, dictionary.shape[1]))]
# Fortran-order dict, as we are going to access its row vectors
dictionary = np.array(dictionary, order='F')
residuals = 0
errors = []
current_cost = np.nan
if verbose == 1:
print('[dict_learning]', end=' ')
# If max_iter is 0, number of iterations returned should be zero
ii = -1
for ii in range(max_iter):
dt = (time.time() - t0)
if verbose == 1:
sys.stdout.write(".")
sys.stdout.flush()
elif verbose:
print("Iteration % 3i "
"(elapsed time: % 3is, % 4.1fmn, current cost % 7.3f)"
% (ii, dt, dt / 60, current_cost))
# Update code
code = sparse_encode(X, dictionary, algorithm=method, alpha=alpha,
init=code, n_jobs=n_jobs)
# Update dictionary
dictionary, residuals = _update_dict(dictionary.T, X.T, code.T,
verbose=verbose, return_r2=True,
random_state=random_state)
dictionary = dictionary.T
# Cost function
current_cost = 0.5 * residuals + alpha * np.sum(np.abs(code))
errors.append(current_cost)
if ii > 0:
dE = errors[-2] - errors[-1]
# assert(dE >= -tol * errors[-1])
if dE < tol * errors[-1]:
if verbose == 1:
# A line return
print("")
elif verbose:
print("--- Convergence reached after %d iterations" % ii)
break
if ii % 5 == 0 and callback is not None:
callback(locals())
if return_n_iter:
return code, dictionary, errors, ii + 1
else:
return code, dictionary, errors
def dict_learning_online(X, n_components=2, alpha=1, n_iter=100,
return_code=True, dict_init=None, callback=None,
batch_size=3, verbose=False, shuffle=True, n_jobs=1,
method='lars', iter_offset=0, random_state=None,
return_inner_stats=False, inner_stats=None,
return_n_iter=False):
"""Solves a dictionary learning matrix factorization problem online.
Finds the best dictionary and the corresponding sparse code for
approximating the data matrix X by solving::
(U^*, V^*) = argmin 0.5 || X - U V ||_2^2 + alpha * || U ||_1
(U,V)
with || V_k ||_2 = 1 for all 0 <= k < n_components
where V is the dictionary and U is the sparse code. This is
accomplished by repeatedly iterating over mini-batches by slicing
the input data.
Read more in the :ref:`User Guide <DictionaryLearning>`.
Parameters
----------
X: array of shape (n_samples, n_features)
Data matrix.
n_components : int,
Number of dictionary atoms to extract.
alpha : float,
Sparsity controlling parameter.
n_iter : int,
Number of iterations to perform.
return_code : boolean,
Whether to also return the code U or just the dictionary V.
dict_init : array of shape (n_components, n_features),
Initial value for the dictionary for warm restart scenarios.
callback :
Callable that gets invoked every five iterations.
batch_size : int,
The number of samples to take in each batch.
verbose :
Degree of output the procedure will print.
shuffle : boolean,
Whether to shuffle the data before splitting it in batches.
n_jobs : int,
Number of parallel jobs to run, or -1 to autodetect.
method : {'lars', 'cd'}
lars: uses the least angle regression method to solve the lasso problem
(linear_model.lars_path)
cd: uses the coordinate descent method to compute the
Lasso solution (linear_model.Lasso). Lars will be faster if
the estimated components are sparse.
iter_offset : int, default 0
Number of previous iterations completed on the dictionary used for
initialization.
random_state : int or RandomState
Pseudo number generator state used for random sampling.
return_inner_stats : boolean, optional
Return the inner statistics A (dictionary covariance) and B
(data approximation). Useful to restart the algorithm in an
online setting. If return_inner_stats is True, return_code is
ignored
inner_stats : tuple of (A, B) ndarrays
Inner sufficient statistics that are kept by the algorithm.
Passing them at initialization is useful in online settings, to
avoid loosing the history of the evolution.
A (n_components, n_components) is the dictionary covariance matrix.
B (n_features, n_components) is the data approximation matrix
return_n_iter : bool
Whether or not to return the number of iterations.
Returns
-------
code : array of shape (n_samples, n_components),
the sparse code (only returned if `return_code=True`)
dictionary : array of shape (n_components, n_features),
the solutions to the dictionary learning problem
n_iter : int
Number of iterations run. Returned only if `return_n_iter` is
set to `True`.
See also
--------
dict_learning
DictionaryLearning
MiniBatchDictionaryLearning
SparsePCA
MiniBatchSparsePCA
"""
if n_components is None:
n_components = X.shape[1]
if method not in ('lars', 'cd'):
raise ValueError('Coding method not supported as a fit algorithm.')
method = 'lasso_' + method
t0 = time.time()
n_samples, n_features = X.shape
# Avoid integer division problems
alpha = float(alpha)
random_state = check_random_state(random_state)
if n_jobs == -1:
n_jobs = cpu_count()
# Init V with SVD of X
if dict_init is not None:
dictionary = dict_init
else:
_, S, dictionary = randomized_svd(X, n_components,
random_state=random_state)
dictionary = S[:, np.newaxis] * dictionary
r = len(dictionary)
if n_components <= r:
dictionary = dictionary[:n_components, :]
else:
dictionary = np.r_[dictionary,
np.zeros((n_components - r, dictionary.shape[1]))]
if verbose == 1:
print('[dict_learning]', end=' ')
if shuffle:
X_train = X.copy()
random_state.shuffle(X_train)
else:
X_train = X
dictionary = check_array(dictionary.T, order='F', dtype=np.float64,
copy=False)
X_train = check_array(X_train, order='C', dtype=np.float64, copy=False)
batches = gen_batches(n_samples, batch_size)
batches = itertools.cycle(batches)
# The covariance of the dictionary
if inner_stats is None:
A = np.zeros((n_components, n_components))
# The data approximation
B = np.zeros((n_features, n_components))
else:
A = inner_stats[0].copy()
B = inner_stats[1].copy()
# If n_iter is zero, we need to return zero.
ii = iter_offset - 1
for ii, batch in zip(range(iter_offset, iter_offset + n_iter), batches):
this_X = X_train[batch]
dt = (time.time() - t0)
if verbose == 1:
sys.stdout.write(".")
sys.stdout.flush()
elif verbose:
if verbose > 10 or ii % ceil(100. / verbose) == 0:
print ("Iteration % 3i (elapsed time: % 3is, % 4.1fmn)"
% (ii, dt, dt / 60))
this_code = sparse_encode(this_X, dictionary.T, algorithm=method,
alpha=alpha, n_jobs=n_jobs).T
# Update the auxiliary variables
if ii < batch_size - 1:
theta = float((ii + 1) * batch_size)
else:
theta = float(batch_size ** 2 + ii + 1 - batch_size)
beta = (theta + 1 - batch_size) / (theta + 1)
A *= beta
A += np.dot(this_code, this_code.T)
B *= beta
B += np.dot(this_X.T, this_code.T)
# Update dictionary
dictionary = _update_dict(dictionary, B, A, verbose=verbose,
random_state=random_state)
# XXX: Can the residuals be of any use?
# Maybe we need a stopping criteria based on the amount of
# modification in the dictionary
if callback is not None:
callback(locals())
if return_inner_stats:
if return_n_iter:
return dictionary.T, (A, B), ii - iter_offset + 1
else:
return dictionary.T, (A, B)
if return_code:
if verbose > 1:
print('Learning code...', end=' ')
elif verbose == 1:
print('|', end=' ')
code = sparse_encode(X, dictionary.T, algorithm=method, alpha=alpha,
n_jobs=n_jobs, check_input=False)
if verbose > 1:
dt = (time.time() - t0)
print('done (total time: % 3is, % 4.1fmn)' % (dt, dt / 60))
if return_n_iter:
return code, dictionary.T, ii - iter_offset + 1
else:
return code, dictionary.T
if return_n_iter:
return dictionary.T, ii - iter_offset + 1
else:
return dictionary.T
class SparseCodingMixin(TransformerMixin):
"""Sparse coding mixin"""
def _set_sparse_coding_params(self, n_components,
transform_algorithm='omp',
transform_n_nonzero_coefs=None,
transform_alpha=None, split_sign=False,
n_jobs=1):
self.n_components = n_components
self.transform_algorithm = transform_algorithm
self.transform_n_nonzero_coefs = transform_n_nonzero_coefs
self.transform_alpha = transform_alpha
self.split_sign = split_sign
self.n_jobs = n_jobs
def transform(self, X, y=None):
"""Encode the data as a sparse combination of the dictionary atoms.
Coding method is determined by the object parameter
`transform_algorithm`.
Parameters
----------
X : array of shape (n_samples, n_features)
Test data to be transformed, must have the same number of
features as the data used to train the model.
Returns
-------
X_new : array, shape (n_samples, n_components)
Transformed data
"""
check_is_fitted(self, 'components_')
# XXX : kwargs is not documented
X = check_array(X)
n_samples, n_features = X.shape
code = sparse_encode(
X, self.components_, algorithm=self.transform_algorithm,
n_nonzero_coefs=self.transform_n_nonzero_coefs,
alpha=self.transform_alpha, n_jobs=self.n_jobs)
if self.split_sign:
# feature vector is split into a positive and negative side
n_samples, n_features = code.shape
split_code = np.empty((n_samples, 2 * n_features))
split_code[:, :n_features] = np.maximum(code, 0)
split_code[:, n_features:] = -np.minimum(code, 0)
code = split_code
return code
class SparseCoder(BaseEstimator, SparseCodingMixin):
"""Sparse coding
Finds a sparse representation of data against a fixed, precomputed
dictionary.
Each row of the result is the solution to a sparse coding problem.
The goal is to find a sparse array `code` such that::
X ~= code * dictionary
Read more in the :ref:`User Guide <SparseCoder>`.
Parameters
----------
dictionary : array, [n_components, n_features]
The dictionary atoms used for sparse coding. Lines are assumed to be
normalized to unit norm.
transform_algorithm : {'lasso_lars', 'lasso_cd', 'lars', 'omp', \
'threshold'}
Algorithm used to transform the data:
lars: uses the least angle regression method (linear_model.lars_path)
lasso_lars: uses Lars to compute the Lasso solution
lasso_cd: uses the coordinate descent method to compute the
Lasso solution (linear_model.Lasso). lasso_lars will be faster if
the estimated components are sparse.
omp: uses orthogonal matching pursuit to estimate the sparse solution
threshold: squashes to zero all coefficients less than alpha from
the projection ``dictionary * X'``
transform_n_nonzero_coefs : int, ``0.1 * n_features`` by default
Number of nonzero coefficients to target in each column of the
solution. This is only used by `algorithm='lars'` and `algorithm='omp'`
and is overridden by `alpha` in the `omp` case.
transform_alpha : float, 1. by default
If `algorithm='lasso_lars'` or `algorithm='lasso_cd'`, `alpha` is the
penalty applied to the L1 norm.
If `algorithm='threshold'`, `alpha` is the absolute value of the
threshold below which coefficients will be squashed to zero.
If `algorithm='omp'`, `alpha` is the tolerance parameter: the value of
the reconstruction error targeted. In this case, it overrides
`n_nonzero_coefs`.
split_sign : bool, False by default
Whether to split the sparse feature vector into the concatenation of
its negative part and its positive part. This can improve the
performance of downstream classifiers.
n_jobs : int,
number of parallel jobs to run
Attributes
----------
components_ : array, [n_components, n_features]
The unchanged dictionary atoms
See also
--------
DictionaryLearning
MiniBatchDictionaryLearning
SparsePCA
MiniBatchSparsePCA
sparse_encode
"""
def __init__(self, dictionary, transform_algorithm='omp',
transform_n_nonzero_coefs=None, transform_alpha=None,
split_sign=False, n_jobs=1):
self._set_sparse_coding_params(dictionary.shape[0],
transform_algorithm,
transform_n_nonzero_coefs,
transform_alpha, split_sign, n_jobs)
self.components_ = dictionary
def fit(self, X, y=None):
"""Do nothing and return the estimator unchanged
This method is just there to implement the usual API and hence
work in pipelines.
"""
return self
class DictionaryLearning(BaseEstimator, SparseCodingMixin):
"""Dictionary learning
Finds a dictionary (a set of atoms) that can best be used to represent data
using a sparse code.
Solves the optimization problem::
(U^*,V^*) = argmin 0.5 || Y - U V ||_2^2 + alpha * || U ||_1
(U,V)
with || V_k ||_2 = 1 for all 0 <= k < n_components
Read more in the :ref:`User Guide <DictionaryLearning>`.
Parameters
----------
n_components : int,
number of dictionary elements to extract
alpha : float,
sparsity controlling parameter
max_iter : int,
maximum number of iterations to perform
tol : float,
tolerance for numerical error
fit_algorithm : {'lars', 'cd'}
lars: uses the least angle regression method to solve the lasso problem
(linear_model.lars_path)
cd: uses the coordinate descent method to compute the
Lasso solution (linear_model.Lasso). Lars will be faster if
the estimated components are sparse.
.. versionadded:: 0.17
*cd* coordinate descent method to improve speed.
transform_algorithm : {'lasso_lars', 'lasso_cd', 'lars', 'omp', \
'threshold'}
Algorithm used to transform the data
lars: uses the least angle regression method (linear_model.lars_path)
lasso_lars: uses Lars to compute the Lasso solution
lasso_cd: uses the coordinate descent method to compute the
Lasso solution (linear_model.Lasso). lasso_lars will be faster if
the estimated components are sparse.
omp: uses orthogonal matching pursuit to estimate the sparse solution
threshold: squashes to zero all coefficients less than alpha from
the projection ``dictionary * X'``
.. versionadded:: 0.17
*lasso_cd* coordinate descent method to improve speed.
transform_n_nonzero_coefs : int, ``0.1 * n_features`` by default
Number of nonzero coefficients to target in each column of the
solution. This is only used by `algorithm='lars'` and `algorithm='omp'`
and is overridden by `alpha` in the `omp` case.
transform_alpha : float, 1. by default
If `algorithm='lasso_lars'` or `algorithm='lasso_cd'`, `alpha` is the
penalty applied to the L1 norm.
If `algorithm='threshold'`, `alpha` is the absolute value of the
threshold below which coefficients will be squashed to zero.
If `algorithm='omp'`, `alpha` is the tolerance parameter: the value of
the reconstruction error targeted. In this case, it overrides
`n_nonzero_coefs`.
split_sign : bool, False by default
Whether to split the sparse feature vector into the concatenation of
its negative part and its positive part. This can improve the
performance of downstream classifiers.
n_jobs : int,
number of parallel jobs to run
code_init : array of shape (n_samples, n_components),
initial value for the code, for warm restart
dict_init : array of shape (n_components, n_features),
initial values for the dictionary, for warm restart
verbose :
degree of verbosity of the printed output
random_state : int or RandomState
Pseudo number generator state used for random sampling.
Attributes
----------
components_ : array, [n_components, n_features]
dictionary atoms extracted from the data
error_ : array
vector of errors at each iteration
n_iter_ : int
Number of iterations run.
Notes
-----
**References:**
J. Mairal, F. Bach, J. Ponce, G. Sapiro, 2009: Online dictionary learning
for sparse coding (http://www.di.ens.fr/sierra/pdfs/icml09.pdf)
See also
--------
SparseCoder
MiniBatchDictionaryLearning
SparsePCA
MiniBatchSparsePCA
"""
def __init__(self, n_components=None, alpha=1, max_iter=1000, tol=1e-8,
fit_algorithm='lars', transform_algorithm='omp',
transform_n_nonzero_coefs=None, transform_alpha=None,
n_jobs=1, code_init=None, dict_init=None, verbose=False,
split_sign=False, random_state=None):
self._set_sparse_coding_params(n_components, transform_algorithm,
transform_n_nonzero_coefs,
transform_alpha, split_sign, n_jobs)
self.alpha = alpha
self.max_iter = max_iter
self.tol = tol
self.fit_algorithm = fit_algorithm
self.code_init = code_init
self.dict_init = dict_init
self.verbose = verbose
self.random_state = random_state
def fit(self, X, y=None):
"""Fit the model from data in X.
Parameters
----------
X: array-like, shape (n_samples, n_features)
Training vector, where n_samples in the number of samples
and n_features is the number of features.
Returns
-------
self: object
Returns the object itself
"""
random_state = check_random_state(self.random_state)
X = check_array(X)
if self.n_components is None:
n_components = X.shape[1]
else:
n_components = self.n_components
V, U, E, self.n_iter_ = dict_learning(
X, n_components, self.alpha,
tol=self.tol, max_iter=self.max_iter,
method=self.fit_algorithm,
n_jobs=self.n_jobs,
code_init=self.code_init,
dict_init=self.dict_init,
verbose=self.verbose,
random_state=random_state,
return_n_iter=True)
self.components_ = U
self.error_ = E
return self
class MiniBatchDictionaryLearning(BaseEstimator, SparseCodingMixin):
"""Mini-batch dictionary learning
Finds a dictionary (a set of atoms) that can best be used to represent data
using a sparse code.
Solves the optimization problem::
(U^*,V^*) = argmin 0.5 || Y - U V ||_2^2 + alpha * || U ||_1
(U,V)
with || V_k ||_2 = 1 for all 0 <= k < n_components
Read more in the :ref:`User Guide <DictionaryLearning>`.
Parameters
----------
n_components : int,
number of dictionary elements to extract
alpha : float,
sparsity controlling parameter
n_iter : int,
total number of iterations to perform
fit_algorithm : {'lars', 'cd'}
lars: uses the least angle regression method to solve the lasso problem
(linear_model.lars_path)
cd: uses the coordinate descent method to compute the
Lasso solution (linear_model.Lasso). Lars will be faster if
the estimated components are sparse.
transform_algorithm : {'lasso_lars', 'lasso_cd', 'lars', 'omp', \
'threshold'}
Algorithm used to transform the data.
lars: uses the least angle regression method (linear_model.lars_path)
lasso_lars: uses Lars to compute the Lasso solution
lasso_cd: uses the coordinate descent method to compute the
Lasso solution (linear_model.Lasso). lasso_lars will be faster if
the estimated components are sparse.
omp: uses orthogonal matching pursuit to estimate the sparse solution
threshold: squashes to zero all coefficients less than alpha from
the projection dictionary * X'
transform_n_nonzero_coefs : int, ``0.1 * n_features`` by default
Number of nonzero coefficients to target in each column of the
solution. This is only used by `algorithm='lars'` and `algorithm='omp'`
and is overridden by `alpha` in the `omp` case.
transform_alpha : float, 1. by default
If `algorithm='lasso_lars'` or `algorithm='lasso_cd'`, `alpha` is the
penalty applied to the L1 norm.
If `algorithm='threshold'`, `alpha` is the absolute value of the
threshold below which coefficients will be squashed to zero.
If `algorithm='omp'`, `alpha` is the tolerance parameter: the value of
the reconstruction error targeted. In this case, it overrides
`n_nonzero_coefs`.
split_sign : bool, False by default
Whether to split the sparse feature vector into the concatenation of
its negative part and its positive part. This can improve the
performance of downstream classifiers.
n_jobs : int,
number of parallel jobs to run
dict_init : array of shape (n_components, n_features),
initial value of the dictionary for warm restart scenarios
verbose :
degree of verbosity of the printed output
batch_size : int,
number of samples in each mini-batch
shuffle : bool,
whether to shuffle the samples before forming batches
random_state : int or RandomState
Pseudo number generator state used for random sampling.
Attributes
----------
components_ : array, [n_components, n_features]
components extracted from the data
inner_stats_ : tuple of (A, B) ndarrays
Internal sufficient statistics that are kept by the algorithm.
Keeping them is useful in online settings, to avoid loosing the
history of the evolution, but they shouldn't have any use for the
end user.
A (n_components, n_components) is the dictionary covariance matrix.
B (n_features, n_components) is the data approximation matrix
n_iter_ : int
Number of iterations run.
Notes
-----
**References:**
J. Mairal, F. Bach, J. Ponce, G. Sapiro, 2009: Online dictionary learning
for sparse coding (http://www.di.ens.fr/sierra/pdfs/icml09.pdf)
See also
--------
SparseCoder
DictionaryLearning
SparsePCA
MiniBatchSparsePCA
"""
def __init__(self, n_components=None, alpha=1, n_iter=1000,
fit_algorithm='lars', n_jobs=1, batch_size=3,
shuffle=True, dict_init=None, transform_algorithm='omp',
transform_n_nonzero_coefs=None, transform_alpha=None,
verbose=False, split_sign=False, random_state=None):
self._set_sparse_coding_params(n_components, transform_algorithm,
transform_n_nonzero_coefs,
transform_alpha, split_sign, n_jobs)
self.alpha = alpha
self.n_iter = n_iter
self.fit_algorithm = fit_algorithm
self.dict_init = dict_init
self.verbose = verbose
self.shuffle = shuffle
self.batch_size = batch_size
self.split_sign = split_sign
self.random_state = random_state
def fit(self, X, y=None):
"""Fit the model from data in X.
Parameters
----------
X: array-like, shape (n_samples, n_features)
Training vector, where n_samples in the number of samples
and n_features is the number of features.
Returns
-------
self : object
Returns the instance itself.
"""
random_state = check_random_state(self.random_state)
X = check_array(X)
U, (A, B), self.n_iter_ = dict_learning_online(
X, self.n_components, self.alpha,
n_iter=self.n_iter, return_code=False,
method=self.fit_algorithm,
n_jobs=self.n_jobs, dict_init=self.dict_init,
batch_size=self.batch_size, shuffle=self.shuffle,
verbose=self.verbose, random_state=random_state,
return_inner_stats=True,
return_n_iter=True)
self.components_ = U
# Keep track of the state of the algorithm to be able to do
# some online fitting (partial_fit)
self.inner_stats_ = (A, B)
self.iter_offset_ = self.n_iter
return self
def partial_fit(self, X, y=None, iter_offset=None):
"""Updates the model using the data in X as a mini-batch.
Parameters
----------
X: array-like, shape (n_samples, n_features)
Training vector, where n_samples in the number of samples
and n_features is the number of features.
iter_offset: integer, optional
The number of iteration on data batches that has been
performed before this call to partial_fit. This is optional:
if no number is passed, the memory of the object is
used.
Returns
-------
self : object
Returns the instance itself.
"""
if not hasattr(self, 'random_state_'):
self.random_state_ = check_random_state(self.random_state)
X = check_array(X)
if hasattr(self, 'components_'):
dict_init = self.components_
else:
dict_init = self.dict_init
inner_stats = getattr(self, 'inner_stats_', None)
if iter_offset is None:
iter_offset = getattr(self, 'iter_offset_', 0)
U, (A, B) = dict_learning_online(
X, self.n_components, self.alpha,
n_iter=self.n_iter, method=self.fit_algorithm,
n_jobs=self.n_jobs, dict_init=dict_init,
batch_size=len(X), shuffle=False,
verbose=self.verbose, return_code=False,
iter_offset=iter_offset, random_state=self.random_state_,
return_inner_stats=True, inner_stats=inner_stats)
self.components_ = U
# Keep track of the state of the algorithm to be able to do
# some online fitting (partial_fit)
self.inner_stats_ = (A, B)
self.iter_offset_ = iter_offset + self.n_iter
return self
|