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
|
#
# Licensed to the Apache Software Foundation (ASF) under one or more
# contributor license agreements. See the NOTICE file distributed with
# this work for additional information regarding copyright ownership.
# The ASF licenses this file to You under the Apache License, Version 2.0
# (the "License"); you may not use this file except in compliance with
# the License. You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
#
import sys
import array as pyarray
import warnings
if sys.version > '3':
xrange = range
basestring = str
from math import exp, log
from numpy import array, random, tile
from collections import namedtuple
from pyspark import SparkContext, since
from pyspark.rdd import RDD, ignore_unicode_prefix
from pyspark.mllib.common import JavaModelWrapper, callMLlibFunc, callJavaFunc, _py2java, _java2py
from pyspark.mllib.linalg import SparseVector, _convert_to_vector, DenseVector
from pyspark.mllib.regression import LabeledPoint
from pyspark.mllib.stat.distribution import MultivariateGaussian
from pyspark.mllib.util import Saveable, Loader, inherit_doc, JavaLoader, JavaSaveable
from pyspark.streaming import DStream
__all__ = ['BisectingKMeansModel', 'BisectingKMeans', 'KMeansModel', 'KMeans',
'GaussianMixtureModel', 'GaussianMixture', 'PowerIterationClusteringModel',
'PowerIterationClustering', 'StreamingKMeans', 'StreamingKMeansModel',
'LDA', 'LDAModel']
@inherit_doc
class BisectingKMeansModel(JavaModelWrapper):
"""
.. note:: Experimental
A clustering model derived from the bisecting k-means method.
>>> data = array([0.0,0.0, 1.0,1.0, 9.0,8.0, 8.0,9.0]).reshape(4, 2)
>>> bskm = BisectingKMeans()
>>> model = bskm.train(sc.parallelize(data, 2), k=4)
>>> p = array([0.0, 0.0])
>>> model.predict(p)
0
>>> model.k
4
>>> model.computeCost(p)
0.0
.. versionadded:: 2.0.0
"""
def __init__(self, java_model):
super(BisectingKMeansModel, self).__init__(java_model)
self.centers = [c.toArray() for c in self.call("clusterCenters")]
@property
@since('2.0.0')
def clusterCenters(self):
"""Get the cluster centers, represented as a list of NumPy
arrays."""
return self.centers
@property
@since('2.0.0')
def k(self):
"""Get the number of clusters"""
return self.call("k")
@since('2.0.0')
def predict(self, x):
"""
Find the cluster that each of the points belongs to in this
model.
:param x: the point (or RDD of points) to determine
compute the clusters for.
"""
if isinstance(x, RDD):
vecs = x.map(_convert_to_vector)
return self.call("predict", vecs)
x = _convert_to_vector(x)
return self.call("predict", x)
@since('2.0.0')
def computeCost(self, x):
"""
Return the Bisecting K-means cost (sum of squared distances of
points to their nearest center) for this model on the given
data. If provided with an RDD of points returns the sum.
:param point: the point or RDD of points to compute the cost(s).
"""
if isinstance(x, RDD):
vecs = x.map(_convert_to_vector)
return self.call("computeCost", vecs)
return self.call("computeCost", _convert_to_vector(x))
class BisectingKMeans(object):
"""
.. note:: Experimental
A bisecting k-means algorithm based on the paper "A comparison of
document clustering techniques" by Steinbach, Karypis, and Kumar,
with modification to fit Spark.
The algorithm starts from a single cluster that contains all points.
Iteratively it finds divisible clusters on the bottom level and
bisects each of them using k-means, until there are `k` leaf
clusters in total or no leaf clusters are divisible.
The bisecting steps of clusters on the same level are grouped
together to increase parallelism. If bisecting all divisible
clusters on the bottom level would result more than `k` leaf
clusters, larger clusters get higher priority.
Based on
U{http://glaros.dtc.umn.edu/gkhome/fetch/papers/docclusterKDDTMW00.pdf}
Steinbach, Karypis, and Kumar, A comparison of document clustering
techniques, KDD Workshop on Text Mining, 2000.
.. versionadded:: 2.0.0
"""
@since('2.0.0')
def train(self, rdd, k=4, maxIterations=20, minDivisibleClusterSize=1.0, seed=-1888008604):
"""
Runs the bisecting k-means algorithm return the model.
:param rdd: input RDD to be trained on
:param k: The desired number of leaf clusters (default: 4).
The actual number could be smaller if there are no divisible
leaf clusters.
:param maxIterations: the max number of k-means iterations to
split clusters (default: 20)
:param minDivisibleClusterSize: the minimum number of points
(if >= 1.0) or the minimum proportion of points (if < 1.0)
of a divisible cluster (default: 1)
:param seed: a random seed (default: -1888008604 from
classOf[BisectingKMeans].getName.##)
"""
java_model = callMLlibFunc(
"trainBisectingKMeans", rdd.map(_convert_to_vector),
k, maxIterations, minDivisibleClusterSize, seed)
return BisectingKMeansModel(java_model)
@inherit_doc
class KMeansModel(Saveable, Loader):
"""A clustering model derived from the k-means method.
>>> data = array([0.0,0.0, 1.0,1.0, 9.0,8.0, 8.0,9.0]).reshape(4, 2)
>>> model = KMeans.train(
... sc.parallelize(data), 2, maxIterations=10, runs=30, initializationMode="random",
... seed=50, initializationSteps=5, epsilon=1e-4)
>>> model.predict(array([0.0, 0.0])) == model.predict(array([1.0, 1.0]))
True
>>> model.predict(array([8.0, 9.0])) == model.predict(array([9.0, 8.0]))
True
>>> model.k
2
>>> model.computeCost(sc.parallelize(data))
2.0000000000000004
>>> model = KMeans.train(sc.parallelize(data), 2)
>>> sparse_data = [
... SparseVector(3, {1: 1.0}),
... SparseVector(3, {1: 1.1}),
... SparseVector(3, {2: 1.0}),
... SparseVector(3, {2: 1.1})
... ]
>>> model = KMeans.train(sc.parallelize(sparse_data), 2, initializationMode="k-means||",
... seed=50, initializationSteps=5, epsilon=1e-4)
>>> model.predict(array([0., 1., 0.])) == model.predict(array([0, 1.1, 0.]))
True
>>> model.predict(array([0., 0., 1.])) == model.predict(array([0, 0, 1.1]))
True
>>> model.predict(sparse_data[0]) == model.predict(sparse_data[1])
True
>>> model.predict(sparse_data[2]) == model.predict(sparse_data[3])
True
>>> isinstance(model.clusterCenters, list)
True
>>> import os, tempfile
>>> path = tempfile.mkdtemp()
>>> model.save(sc, path)
>>> sameModel = KMeansModel.load(sc, path)
>>> sameModel.predict(sparse_data[0]) == model.predict(sparse_data[0])
True
>>> from shutil import rmtree
>>> try:
... rmtree(path)
... except OSError:
... pass
>>> data = array([-383.1,-382.9, 28.7,31.2, 366.2,367.3]).reshape(3, 2)
>>> model = KMeans.train(sc.parallelize(data), 3, maxIterations=0,
... initialModel = KMeansModel([(-1000.0,-1000.0),(5.0,5.0),(1000.0,1000.0)]))
>>> model.clusterCenters
[array([-1000., -1000.]), array([ 5., 5.]), array([ 1000., 1000.])]
.. versionadded:: 0.9.0
"""
def __init__(self, centers):
self.centers = centers
@property
@since('1.0.0')
def clusterCenters(self):
"""Get the cluster centers, represented as a list of NumPy arrays."""
return self.centers
@property
@since('1.4.0')
def k(self):
"""Total number of clusters."""
return len(self.centers)
@since('0.9.0')
def predict(self, x):
"""
Find the cluster that each of the points belongs to in this
model.
:param x: the point (or RDD of points) to determine
compute the clusters for.
"""
best = 0
best_distance = float("inf")
if isinstance(x, RDD):
return x.map(self.predict)
x = _convert_to_vector(x)
for i in xrange(len(self.centers)):
distance = x.squared_distance(self.centers[i])
if distance < best_distance:
best = i
best_distance = distance
return best
@since('1.4.0')
def computeCost(self, rdd):
"""
Return the K-means cost (sum of squared distances of points to
their nearest center) for this model on the given
data.
:param point: the RDD of points to compute the cost on.
"""
cost = callMLlibFunc("computeCostKmeansModel", rdd.map(_convert_to_vector),
[_convert_to_vector(c) for c in self.centers])
return cost
@since('1.4.0')
def save(self, sc, path):
"""
Save this model to the given path.
"""
java_centers = _py2java(sc, [_convert_to_vector(c) for c in self.centers])
java_model = sc._jvm.org.apache.spark.mllib.clustering.KMeansModel(java_centers)
java_model.save(sc._jsc.sc(), path)
@classmethod
@since('1.4.0')
def load(cls, sc, path):
"""
Load a model from the given path.
"""
java_model = sc._jvm.org.apache.spark.mllib.clustering.KMeansModel.load(sc._jsc.sc(), path)
return KMeansModel(_java2py(sc, java_model.clusterCenters()))
class KMeans(object):
"""
.. versionadded:: 0.9.0
"""
@classmethod
@since('0.9.0')
def train(cls, rdd, k, maxIterations=100, runs=1, initializationMode="k-means||",
seed=None, initializationSteps=5, epsilon=1e-4, initialModel=None):
"""Train a k-means clustering model."""
if runs != 1:
warnings.warn(
"Support for runs is deprecated in 1.6.0. This param will have no effect in 2.0.0.")
clusterInitialModel = []
if initialModel is not None:
if not isinstance(initialModel, KMeansModel):
raise Exception("initialModel is of "+str(type(initialModel))+". It needs "
"to be of <type 'KMeansModel'>")
clusterInitialModel = [_convert_to_vector(c) for c in initialModel.clusterCenters]
model = callMLlibFunc("trainKMeansModel", rdd.map(_convert_to_vector), k, maxIterations,
runs, initializationMode, seed, initializationSteps, epsilon,
clusterInitialModel)
centers = callJavaFunc(rdd.context, model.clusterCenters)
return KMeansModel([c.toArray() for c in centers])
@inherit_doc
class GaussianMixtureModel(JavaModelWrapper, JavaSaveable, JavaLoader):
"""
.. note:: Experimental
A clustering model derived from the Gaussian Mixture Model method.
>>> from pyspark.mllib.linalg import Vectors, DenseMatrix
>>> from numpy.testing import assert_equal
>>> from shutil import rmtree
>>> import os, tempfile
>>> clusterdata_1 = sc.parallelize(array([-0.1,-0.05,-0.01,-0.1,
... 0.9,0.8,0.75,0.935,
... -0.83,-0.68,-0.91,-0.76 ]).reshape(6, 2), 2)
>>> model = GaussianMixture.train(clusterdata_1, 3, convergenceTol=0.0001,
... maxIterations=50, seed=10)
>>> labels = model.predict(clusterdata_1).collect()
>>> labels[0]==labels[1]
False
>>> labels[1]==labels[2]
False
>>> labels[4]==labels[5]
True
>>> model.predict([-0.1,-0.05])
0
>>> softPredicted = model.predictSoft([-0.1,-0.05])
>>> abs(softPredicted[0] - 1.0) < 0.001
True
>>> abs(softPredicted[1] - 0.0) < 0.001
True
>>> abs(softPredicted[2] - 0.0) < 0.001
True
>>> path = tempfile.mkdtemp()
>>> model.save(sc, path)
>>> sameModel = GaussianMixtureModel.load(sc, path)
>>> assert_equal(model.weights, sameModel.weights)
>>> mus, sigmas = list(
... zip(*[(g.mu, g.sigma) for g in model.gaussians]))
>>> sameMus, sameSigmas = list(
... zip(*[(g.mu, g.sigma) for g in sameModel.gaussians]))
>>> mus == sameMus
True
>>> sigmas == sameSigmas
True
>>> from shutil import rmtree
>>> try:
... rmtree(path)
... except OSError:
... pass
>>> data = array([-5.1971, -2.5359, -3.8220,
... -5.2211, -5.0602, 4.7118,
... 6.8989, 3.4592, 4.6322,
... 5.7048, 4.6567, 5.5026,
... 4.5605, 5.2043, 6.2734])
>>> clusterdata_2 = sc.parallelize(data.reshape(5,3))
>>> model = GaussianMixture.train(clusterdata_2, 2, convergenceTol=0.0001,
... maxIterations=150, seed=10)
>>> labels = model.predict(clusterdata_2).collect()
>>> labels[0]==labels[1]
True
>>> labels[2]==labels[3]==labels[4]
True
.. versionadded:: 1.3.0
"""
@property
@since('1.4.0')
def weights(self):
"""
Weights for each Gaussian distribution in the mixture, where weights[i] is
the weight for Gaussian i, and weights.sum == 1.
"""
return array(self.call("weights"))
@property
@since('1.4.0')
def gaussians(self):
"""
Array of MultivariateGaussian where gaussians[i] represents
the Multivariate Gaussian (Normal) Distribution for Gaussian i.
"""
return [
MultivariateGaussian(gaussian[0], gaussian[1])
for gaussian in self.call("gaussians")]
@property
@since('1.4.0')
def k(self):
"""Number of gaussians in mixture."""
return len(self.weights)
@since('1.3.0')
def predict(self, x):
"""
Find the cluster to which the point 'x' or each point in RDD 'x'
has maximum membership in this model.
:param x: vector or RDD of vector represents data points.
:return: cluster label or RDD of cluster labels.
"""
if isinstance(x, RDD):
cluster_labels = self.predictSoft(x).map(lambda z: z.index(max(z)))
return cluster_labels
else:
z = self.predictSoft(x)
return z.argmax()
@since('1.3.0')
def predictSoft(self, x):
"""
Find the membership of point 'x' or each point in RDD 'x' to all mixture components.
:param x: vector or RDD of vector represents data points.
:return: the membership value to all mixture components for vector 'x'
or each vector in RDD 'x'.
"""
if isinstance(x, RDD):
means, sigmas = zip(*[(g.mu, g.sigma) for g in self.gaussians])
membership_matrix = callMLlibFunc("predictSoftGMM", x.map(_convert_to_vector),
_convert_to_vector(self.weights), means, sigmas)
return membership_matrix.map(lambda x: pyarray.array('d', x))
else:
return self.call("predictSoft", _convert_to_vector(x)).toArray()
@classmethod
@since('1.5.0')
def load(cls, sc, path):
"""Load the GaussianMixtureModel from disk.
:param sc: SparkContext
:param path: str, path to where the model is stored.
"""
model = cls._load_java(sc, path)
wrapper = sc._jvm.GaussianMixtureModelWrapper(model)
return cls(wrapper)
class GaussianMixture(object):
"""
.. note:: Experimental
Learning algorithm for Gaussian Mixtures using the expectation-maximization algorithm.
:param data: RDD of data points
:param k: Number of components
:param convergenceTol: Threshold value to check the convergence criteria. Defaults to 1e-3
:param maxIterations: Number of iterations. Default to 100
:param seed: Random Seed
:param initialModel: GaussianMixtureModel for initializing learning
.. versionadded:: 1.3.0
"""
@classmethod
@since('1.3.0')
def train(cls, rdd, k, convergenceTol=1e-3, maxIterations=100, seed=None, initialModel=None):
"""Train a Gaussian Mixture clustering model."""
initialModelWeights = None
initialModelMu = None
initialModelSigma = None
if initialModel is not None:
if initialModel.k != k:
raise Exception("Mismatched cluster count, initialModel.k = %s, however k = %s"
% (initialModel.k, k))
initialModelWeights = list(initialModel.weights)
initialModelMu = [initialModel.gaussians[i].mu for i in range(initialModel.k)]
initialModelSigma = [initialModel.gaussians[i].sigma for i in range(initialModel.k)]
java_model = callMLlibFunc("trainGaussianMixtureModel", rdd.map(_convert_to_vector),
k, convergenceTol, maxIterations, seed,
initialModelWeights, initialModelMu, initialModelSigma)
return GaussianMixtureModel(java_model)
class PowerIterationClusteringModel(JavaModelWrapper, JavaSaveable, JavaLoader):
"""
.. note:: Experimental
Model produced by [[PowerIterationClustering]].
>>> data = [(0, 1, 1.0), (0, 2, 1.0), (0, 3, 1.0), (1, 2, 1.0), (1, 3, 1.0),
... (2, 3, 1.0), (3, 4, 0.1), (4, 5, 1.0), (4, 15, 1.0), (5, 6, 1.0),
... (6, 7, 1.0), (7, 8, 1.0), (8, 9, 1.0), (9, 10, 1.0), (10, 11, 1.0),
... (11, 12, 1.0), (12, 13, 1.0), (13, 14, 1.0), (14, 15, 1.0)]
>>> rdd = sc.parallelize(data, 2)
>>> model = PowerIterationClustering.train(rdd, 2, 100)
>>> model.k
2
>>> result = sorted(model.assignments().collect(), key=lambda x: x.id)
>>> result[0].cluster == result[1].cluster == result[2].cluster == result[3].cluster
True
>>> result[4].cluster == result[5].cluster == result[6].cluster == result[7].cluster
True
>>> import os, tempfile
>>> path = tempfile.mkdtemp()
>>> model.save(sc, path)
>>> sameModel = PowerIterationClusteringModel.load(sc, path)
>>> sameModel.k
2
>>> result = sorted(model.assignments().collect(), key=lambda x: x.id)
>>> result[0].cluster == result[1].cluster == result[2].cluster == result[3].cluster
True
>>> result[4].cluster == result[5].cluster == result[6].cluster == result[7].cluster
True
>>> from shutil import rmtree
>>> try:
... rmtree(path)
... except OSError:
... pass
.. versionadded:: 1.5.0
"""
@property
@since('1.5.0')
def k(self):
"""
Returns the number of clusters.
"""
return self.call("k")
@since('1.5.0')
def assignments(self):
"""
Returns the cluster assignments of this model.
"""
return self.call("getAssignments").map(
lambda x: (PowerIterationClustering.Assignment(*x)))
@classmethod
@since('1.5.0')
def load(cls, sc, path):
"""
Load a model from the given path.
"""
model = cls._load_java(sc, path)
wrapper = sc._jvm.PowerIterationClusteringModelWrapper(model)
return PowerIterationClusteringModel(wrapper)
class PowerIterationClustering(object):
"""
.. note:: Experimental
Power Iteration Clustering (PIC), a scalable graph clustering algorithm
developed by [[http://www.icml2010.org/papers/387.pdf Lin and Cohen]].
From the abstract: PIC finds a very low-dimensional embedding of a
dataset using truncated power iteration on a normalized pair-wise
similarity matrix of the data.
.. versionadded:: 1.5.0
"""
@classmethod
@since('1.5.0')
def train(cls, rdd, k, maxIterations=100, initMode="random"):
"""
:param rdd: an RDD of (i, j, s,,ij,,) tuples representing the
affinity matrix, which is the matrix A in the PIC paper.
The similarity s,,ij,, must be nonnegative.
This is a symmetric matrix and hence s,,ij,, = s,,ji,,.
For any (i, j) with nonzero similarity, there should be
either (i, j, s,,ij,,) or (j, i, s,,ji,,) in the input.
Tuples with i = j are ignored, because we assume
s,,ij,, = 0.0.
:param k: Number of clusters.
:param maxIterations: Maximum number of iterations of the
PIC algorithm.
:param initMode: Initialization mode.
"""
model = callMLlibFunc("trainPowerIterationClusteringModel",
rdd.map(_convert_to_vector), int(k), int(maxIterations), initMode)
return PowerIterationClusteringModel(model)
class Assignment(namedtuple("Assignment", ["id", "cluster"])):
"""
Represents an (id, cluster) tuple.
.. versionadded:: 1.5.0
"""
class StreamingKMeansModel(KMeansModel):
"""
.. note:: Experimental
Clustering model which can perform an online update of the centroids.
The update formula for each centroid is given by
* c_t+1 = ((c_t * n_t * a) + (x_t * m_t)) / (n_t + m_t)
* n_t+1 = n_t * a + m_t
where
* c_t: Centroid at the n_th iteration.
* n_t: Number of samples (or) weights associated with the centroid
at the n_th iteration.
* x_t: Centroid of the new data closest to c_t.
* m_t: Number of samples (or) weights of the new data closest to c_t
* c_t+1: New centroid.
* n_t+1: New number of weights.
* a: Decay Factor, which gives the forgetfulness.
Note that if a is set to 1, it is the weighted mean of the previous
and new data. If it set to zero, the old centroids are completely
forgotten.
:param clusterCenters: Initial cluster centers.
:param clusterWeights: List of weights assigned to each cluster.
>>> initCenters = [[0.0, 0.0], [1.0, 1.0]]
>>> initWeights = [1.0, 1.0]
>>> stkm = StreamingKMeansModel(initCenters, initWeights)
>>> data = sc.parallelize([[-0.1, -0.1], [0.1, 0.1],
... [0.9, 0.9], [1.1, 1.1]])
>>> stkm = stkm.update(data, 1.0, u"batches")
>>> stkm.centers
array([[ 0., 0.],
[ 1., 1.]])
>>> stkm.predict([-0.1, -0.1])
0
>>> stkm.predict([0.9, 0.9])
1
>>> stkm.clusterWeights
[3.0, 3.0]
>>> decayFactor = 0.0
>>> data = sc.parallelize([DenseVector([1.5, 1.5]), DenseVector([0.2, 0.2])])
>>> stkm = stkm.update(data, 0.0, u"batches")
>>> stkm.centers
array([[ 0.2, 0.2],
[ 1.5, 1.5]])
>>> stkm.clusterWeights
[1.0, 1.0]
>>> stkm.predict([0.2, 0.2])
0
>>> stkm.predict([1.5, 1.5])
1
.. versionadded:: 1.5.0
"""
def __init__(self, clusterCenters, clusterWeights):
super(StreamingKMeansModel, self).__init__(centers=clusterCenters)
self._clusterWeights = list(clusterWeights)
@property
@since('1.5.0')
def clusterWeights(self):
"""Return the cluster weights."""
return self._clusterWeights
@ignore_unicode_prefix
@since('1.5.0')
def update(self, data, decayFactor, timeUnit):
"""Update the centroids, according to data
:param data: Should be a RDD that represents the new data.
:param decayFactor: forgetfulness of the previous centroids.
:param timeUnit: Can be "batches" or "points". If points, then the
decay factor is raised to the power of number of new
points and if batches, it is used as it is.
"""
if not isinstance(data, RDD):
raise TypeError("Data should be of an RDD, got %s." % type(data))
data = data.map(_convert_to_vector)
decayFactor = float(decayFactor)
if timeUnit not in ["batches", "points"]:
raise ValueError(
"timeUnit should be 'batches' or 'points', got %s." % timeUnit)
vectorCenters = [_convert_to_vector(center) for center in self.centers]
updatedModel = callMLlibFunc(
"updateStreamingKMeansModel", vectorCenters, self._clusterWeights,
data, decayFactor, timeUnit)
self.centers = array(updatedModel[0])
self._clusterWeights = list(updatedModel[1])
return self
class StreamingKMeans(object):
"""
.. note:: Experimental
Provides methods to set k, decayFactor, timeUnit to configure the
KMeans algorithm for fitting and predicting on incoming dstreams.
More details on how the centroids are updated are provided under the
docs of StreamingKMeansModel.
:param k: int, number of clusters
:param decayFactor: float, forgetfulness of the previous centroids.
:param timeUnit: can be "batches" or "points". If points, then the
decayfactor is raised to the power of no. of new points.
.. versionadded:: 1.5.0
"""
def __init__(self, k=2, decayFactor=1.0, timeUnit="batches"):
self._k = k
self._decayFactor = decayFactor
if timeUnit not in ["batches", "points"]:
raise ValueError(
"timeUnit should be 'batches' or 'points', got %s." % timeUnit)
self._timeUnit = timeUnit
self._model = None
@since('1.5.0')
def latestModel(self):
"""Return the latest model"""
return self._model
def _validate(self, dstream):
if self._model is None:
raise ValueError(
"Initial centers should be set either by setInitialCenters "
"or setRandomCenters.")
if not isinstance(dstream, DStream):
raise TypeError(
"Expected dstream to be of type DStream, "
"got type %s" % type(dstream))
@since('1.5.0')
def setK(self, k):
"""Set number of clusters."""
self._k = k
return self
@since('1.5.0')
def setDecayFactor(self, decayFactor):
"""Set decay factor."""
self._decayFactor = decayFactor
return self
@since('1.5.0')
def setHalfLife(self, halfLife, timeUnit):
"""
Set number of batches after which the centroids of that
particular batch has half the weightage.
"""
self._timeUnit = timeUnit
self._decayFactor = exp(log(0.5) / halfLife)
return self
@since('1.5.0')
def setInitialCenters(self, centers, weights):
"""
Set initial centers. Should be set before calling trainOn.
"""
self._model = StreamingKMeansModel(centers, weights)
return self
@since('1.5.0')
def setRandomCenters(self, dim, weight, seed):
"""
Set the initial centres to be random samples from
a gaussian population with constant weights.
"""
rng = random.RandomState(seed)
clusterCenters = rng.randn(self._k, dim)
clusterWeights = tile(weight, self._k)
self._model = StreamingKMeansModel(clusterCenters, clusterWeights)
return self
@since('1.5.0')
def trainOn(self, dstream):
"""Train the model on the incoming dstream."""
self._validate(dstream)
def update(rdd):
self._model.update(rdd, self._decayFactor, self._timeUnit)
dstream.foreachRDD(update)
@since('1.5.0')
def predictOn(self, dstream):
"""
Make predictions on a dstream.
Returns a transformed dstream object
"""
self._validate(dstream)
return dstream.map(lambda x: self._model.predict(x))
@since('1.5.0')
def predictOnValues(self, dstream):
"""
Make predictions on a keyed dstream.
Returns a transformed dstream object.
"""
self._validate(dstream)
return dstream.mapValues(lambda x: self._model.predict(x))
class LDAModel(JavaModelWrapper, JavaSaveable, Loader):
""" A clustering model derived from the LDA method.
Latent Dirichlet Allocation (LDA), a topic model designed for text documents.
Terminology
- "word" = "term": an element of the vocabulary
- "token": instance of a term appearing in a document
- "topic": multinomial distribution over words representing some concept
References:
- Original LDA paper (journal version):
Blei, Ng, and Jordan. "Latent Dirichlet Allocation." JMLR, 2003.
>>> from pyspark.mllib.linalg import Vectors
>>> from numpy.testing import assert_almost_equal, assert_equal
>>> data = [
... [1, Vectors.dense([0.0, 1.0])],
... [2, SparseVector(2, {0: 1.0})],
... ]
>>> rdd = sc.parallelize(data)
>>> model = LDA.train(rdd, k=2, seed=1)
>>> model.vocabSize()
2
>>> model.describeTopics()
[([1, 0], [0.5..., 0.49...]), ([0, 1], [0.5..., 0.49...])]
>>> model.describeTopics(1)
[([1], [0.5...]), ([0], [0.5...])]
>>> topics = model.topicsMatrix()
>>> topics_expect = array([[0.5, 0.5], [0.5, 0.5]])
>>> assert_almost_equal(topics, topics_expect, 1)
>>> import os, tempfile
>>> from shutil import rmtree
>>> path = tempfile.mkdtemp()
>>> model.save(sc, path)
>>> sameModel = LDAModel.load(sc, path)
>>> assert_equal(sameModel.topicsMatrix(), model.topicsMatrix())
>>> sameModel.vocabSize() == model.vocabSize()
True
>>> try:
... rmtree(path)
... except OSError:
... pass
.. versionadded:: 1.5.0
"""
@since('1.5.0')
def topicsMatrix(self):
"""Inferred topics, where each topic is represented by a distribution over terms."""
return self.call("topicsMatrix").toArray()
@since('1.5.0')
def vocabSize(self):
"""Vocabulary size (number of terms or terms in the vocabulary)"""
return self.call("vocabSize")
@since('1.6.0')
def describeTopics(self, maxTermsPerTopic=None):
"""Return the topics described by weighted terms.
WARNING: If vocabSize and k are large, this can return a large object!
:param maxTermsPerTopic: Maximum number of terms to collect for each topic.
(default: vocabulary size)
:return: Array over topics. Each topic is represented as a pair of matching arrays:
(term indices, term weights in topic).
Each topic's terms are sorted in order of decreasing weight.
"""
if maxTermsPerTopic is None:
topics = self.call("describeTopics")
else:
topics = self.call("describeTopics", maxTermsPerTopic)
return topics
@classmethod
@since('1.5.0')
def load(cls, sc, path):
"""Load the LDAModel from disk.
:param sc: SparkContext
:param path: str, path to where the model is stored.
"""
if not isinstance(sc, SparkContext):
raise TypeError("sc should be a SparkContext, got type %s" % type(sc))
if not isinstance(path, basestring):
raise TypeError("path should be a basestring, got type %s" % type(path))
model = callMLlibFunc("loadLDAModel", sc, path)
return LDAModel(model)
class LDA(object):
"""
.. versionadded:: 1.5.0
"""
@classmethod
@since('1.5.0')
def train(cls, rdd, k=10, maxIterations=20, docConcentration=-1.0,
topicConcentration=-1.0, seed=None, checkpointInterval=10, optimizer="em"):
"""Train a LDA model.
:param rdd: RDD of data points
:param k: Number of clusters you want
:param maxIterations: Number of iterations. Default to 20
:param docConcentration: Concentration parameter (commonly named "alpha")
for the prior placed on documents' distributions over topics ("theta").
:param topicConcentration: Concentration parameter (commonly named "beta" or "eta")
for the prior placed on topics' distributions over terms.
:param seed: Random Seed
:param checkpointInterval: Period (in iterations) between checkpoints.
:param optimizer: LDAOptimizer used to perform the actual calculation.
Currently "em", "online" are supported. Default to "em".
"""
model = callMLlibFunc("trainLDAModel", rdd, k, maxIterations,
docConcentration, topicConcentration, seed,
checkpointInterval, optimizer)
return LDAModel(model)
def _test():
import doctest
import pyspark.mllib.clustering
globs = pyspark.mllib.clustering.__dict__.copy()
globs['sc'] = SparkContext('local[4]', 'PythonTest', batchSize=2)
(failure_count, test_count) = doctest.testmod(globs=globs, optionflags=doctest.ELLIPSIS)
globs['sc'].stop()
if failure_count:
exit(-1)
if __name__ == "__main__":
_test()
|