pyspark.mllib.classification

1 # 2 # Licensed to the Apache Software Foundation (ASF) under one or more 3 # contributor license agreements. See the NOTICE file distributed with 4 # this work for additional information regarding copyright ownership. 5 # The ASF licenses this file to You under the Apache License, Version 2.0 6 # (the "License"); you may not use this file except in compliance with 7 # the License. You may obtain a copy of the License at 8 # 9 # http://www.apache.org/licenses/LICENSE-2.0 10 # 11 # Unless required by applicable law or agreed to in writing, software 12 # distributed under the License is distributed on an "AS IS" BASIS, 13 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. 14 # See the License for the specific language governing permissions and 15 # limitations under the License. 16 # 17 18 import numpy 19 20 from numpy import array, shape 21 from pyspark import SparkContext 22 from pyspark.mllib._common import \ 23 _dot, _get_unmangled_rdd, _get_unmangled_double_vector_rdd, \ 24 _serialize_double_matrix, _deserialize_double_matrix, \ 25 _serialize_double_vector, _deserialize_double_vector, \ 26 _get_initial_weights, _serialize_rating, _regression_train_wrapper, \ 27 _linear_predictor_typecheck, _get_unmangled_labeled_point_rdd 28 from pyspark.mllib.linalg import SparseVector 29 from pyspark.mllib.regression import LabeledPoint, LinearModel 30 from math import exp, log

31 32 33 -class LogisticRegressionModel(LinearModel):

34 """A linear binary classification model derived from logistic regression. 35 36 >>> data = [ 37 ... LabeledPoint(0.0, [0.0]), 38 ... LabeledPoint(1.0, [1.0]), 39 ... LabeledPoint(1.0, [2.0]), 40 ... LabeledPoint(1.0, [3.0]) 41 ... ] 42 >>> lrm = LogisticRegressionWithSGD.train(sc.parallelize(data)) 43 >>> lrm.predict(array([1.0])) > 0 44 True 45 >>> lrm.predict(array([0.0])) <= 0 46 True 47 >>> sparse_data = [ 48 ... LabeledPoint(0.0, SparseVector(2, {0: 0.0})), 49 ... LabeledPoint(1.0, SparseVector(2, {1: 1.0})), 50 ... LabeledPoint(0.0, SparseVector(2, {0: 0.0})), 51 ... LabeledPoint(1.0, SparseVector(2, {1: 2.0})) 52 ... ] 53 >>> lrm = LogisticRegressionWithSGD.train(sc.parallelize(sparse_data)) 54 >>> lrm.predict(array([0.0, 1.0])) > 0 55 True 56 >>> lrm.predict(array([0.0, 0.0])) <= 0 57 True 58 >>> lrm.predict(SparseVector(2, {1: 1.0})) > 0 59 True 60 >>> lrm.predict(SparseVector(2, {1: 0.0})) <= 0 61 True 62 """

63 - def predict(self, x):

64 _linear_predictor_typecheck(x, self._coeff) 65 margin = _dot(x, self._coeff) + self._intercept 66 prob = 1/(1 + exp(-margin)) 67 return 1 if prob > 0.5 else 0

68

69 70 -class LogisticRegressionWithSGD(object):

71 @classmethod

72 - def train(cls, data, iterations=100, step=1.0, miniBatchFraction=1.0, initialWeights=None):

73 """Train a logistic regression model on the given data.""" 74 sc = data.context 75 train_func = lambda d, i: sc._jvm.PythonMLLibAPI().trainLogisticRegressionModelWithSGD( 76 d._jrdd, iterations, step, miniBatchFraction, i) 77 return _regression_train_wrapper(sc, train_func, LogisticRegressionModel, data, 78 initialWeights)

79

80 81 -class SVMModel(LinearModel):

82 """A support vector machine. 83 84 >>> data = [ 85 ... LabeledPoint(0.0, [0.0]), 86 ... LabeledPoint(1.0, [1.0]), 87 ... LabeledPoint(1.0, [2.0]), 88 ... LabeledPoint(1.0, [3.0]) 89 ... ] 90 >>> svm = SVMWithSGD.train(sc.parallelize(data)) 91 >>> svm.predict(array([1.0])) > 0 92 True 93 >>> sparse_data = [ 94 ... LabeledPoint(0.0, SparseVector(2, {0: -1.0})), 95 ... LabeledPoint(1.0, SparseVector(2, {1: 1.0})), 96 ... LabeledPoint(0.0, SparseVector(2, {0: 0.0})), 97 ... LabeledPoint(1.0, SparseVector(2, {1: 2.0})) 98 ... ] 99 >>> svm = SVMWithSGD.train(sc.parallelize(sparse_data)) 100 >>> svm.predict(SparseVector(2, {1: 1.0})) > 0 101 True 102 >>> svm.predict(SparseVector(2, {0: -1.0})) <= 0 103 True 104 """

105 - def predict(self, x):

106 _linear_predictor_typecheck(x, self._coeff) 107 margin = _dot(x, self._coeff) + self._intercept 108 return 1 if margin >= 0 else 0

109

110 111 -class SVMWithSGD(object):

112 @classmethod

113 - def train(cls, data, iterations=100, step=1.0, regParam=1.0, 114 miniBatchFraction=1.0, initialWeights=None):

115 """Train a support vector machine on the given data.""" 116 sc = data.context 117 train_func = lambda d, i: sc._jvm.PythonMLLibAPI().trainSVMModelWithSGD( 118 d._jrdd, iterations, step, regParam, miniBatchFraction, i) 119 return _regression_train_wrapper(sc, train_func, SVMModel, data, initialWeights)

120

121 122 -class NaiveBayesModel(object):

123 """ 124 Model for Naive Bayes classifiers. 125 126 Contains two parameters: 127 - pi: vector of logs of class priors (dimension C) 128 - theta: matrix of logs of class conditional probabilities (CxD) 129 130 >>> data = [ 131 ... LabeledPoint(0.0, [0.0, 0.0]), 132 ... LabeledPoint(0.0, [0.0, 1.0]), 133 ... LabeledPoint(1.0, [1.0, 0.0]), 134 ... ] 135 >>> model = NaiveBayes.train(sc.parallelize(data)) 136 >>> model.predict(array([0.0, 1.0])) 137 0.0 138 >>> model.predict(array([1.0, 0.0])) 139 1.0 140 >>> sparse_data = [ 141 ... LabeledPoint(0.0, SparseVector(2, {1: 0.0})), 142 ... LabeledPoint(0.0, SparseVector(2, {1: 1.0})), 143 ... LabeledPoint(1.0, SparseVector(2, {0: 1.0})) 144 ... ] 145 >>> model = NaiveBayes.train(sc.parallelize(sparse_data)) 146 >>> model.predict(SparseVector(2, {1: 1.0})) 147 0.0 148 >>> model.predict(SparseVector(2, {0: 1.0})) 149 1.0 150 """ 151

152 - def __init__(self, labels, pi, theta):

153 self.labels = labels 154 self.pi = pi 155 self.theta = theta

156

157 - def predict(self, x):

158 """Return the most likely class for a data vector x""" 159 return self.labels[numpy.argmax(self.pi + _dot(x, self.theta.transpose()))]

160

161 162 -class NaiveBayes(object):

163 @classmethod

164 - def train(cls, data, lambda_=1.0):

165 """ 166 Train a Naive Bayes model given an RDD of (label, features) vectors. 167 168 This is the Multinomial NB (U{http://tinyurl.com/lsdw6p}) which can 169 handle all kinds of discrete data. For example, by converting 170 documents into TF-IDF vectors, it can be used for document 171 classification. By making every vector a 0-1 vector, it can also be 172 used as Bernoulli NB (U{http://tinyurl.com/p7c96j6}). 173 174 @param data: RDD of NumPy vectors, one per element, where the first 175 coordinate is the label and the rest is the feature vector 176 (e.g. a count vector). 177 @param lambda_: The smoothing parameter 178 """ 179 sc = data.context 180 dataBytes = _get_unmangled_labeled_point_rdd(data) 181 ans = sc._jvm.PythonMLLibAPI().trainNaiveBayes(dataBytes._jrdd, lambda_) 182 return NaiveBayesModel( 183 _deserialize_double_vector(ans[0]), 184 _deserialize_double_vector(ans[1]), 185 _deserialize_double_matrix(ans[2]))

186

187 188 -def _test():

189 import doctest 190 globs = globals().copy() 191 globs['sc'] = SparkContext('local[4]', 'PythonTest', batchSize=2) 192 (failure_count, test_count) = doctest.testmod(globs=globs, optionflags=doctest.ELLIPSIS) 193 globs['sc'].stop() 194 if failure_count: 195 exit(-1)

196 197 if __name__ == "__main__": 198 _test() 199

Source Code for Module pyspark.mllib.classification