简单的贝叶斯分类器的python实现

  1 # -*- coding: utf-8 -*-
  2 '''
  3 >>> c = Classy()
  4 >>> c.train(['cpu', 'RAM', 'ALU', 'io', 'bridge', 'disk'], 'architecture')
  5 True
  6 >>> c.train(['monitor', 'mouse', 'keyboard', 'microphone', 'headphones'], 'input_devices')
  7 True
  8 >>> c.train(['desk', 'chair', 'cabinet', 'lamp'], 'office furniture')
  9 True
 10 >>> my_office = ['cpu', 'monitor', 'mouse', 'chair']
 11 >>> c.classify(my_office)
 12 ('input_devices', -1.0986122886681098)
 13 ...
 14 >>> c = Classy()
 15 >>> c.train(['cpu', 'RAM', 'ALU', 'io', 'bridge', 'disk'], 'architecture')
 16 True
 17 >>> c.train(['monitor', 'mouse', 'keyboard', 'microphone', 'headphones'], 'input_devices')
 18 True
 19 >>> c.train(['desk', 'chair', 'cabinet', 'lamp'], 'office furniture')
 20 True
 21 >>> my_office = ['cpu', 'monitor', 'mouse', 'chair']
 22 >>> c.classify(my_office)
 23 ('input_devices', -1.0986122886681098)
 24 ...
 25 '''
 26
 27 from collections import Counter
 28 import math
 29
 30 class ClassifierNotTrainedException(Exception):
 31
 32     def __str__(self):
 33         return "Classifier is not trained."
 34
 35 class Classy(object):
 36
 37     def __init__(self):
 38         self.term_count_store = {}
 39         self.data = {
 40             'class_term_count': {},
 41             'beta_priors': {},
 42             'class_doc_count': {},
 43         }
 44         self.total_term_count = 0
 45         self.total_doc_count = 0
 46
 47     def train(self, document_source, class_id):
 48
 49         '''
 50         Trains the classifier.
 51
 52         '''
 53         count = Counter(document_source)
 54         try:
 55             self.term_count_store[class_id]
 56         except KeyError:
 57             self.term_count_store[class_id] = {}
 58         for term in count:
 59             try:
 60                 self.term_count_store[class_id][term] += count[term]
 61             except KeyError:
 62                 self.term_count_store[class_id][term] = count[term]
 63         try:
 64             self.data['class_term_count'][class_id] += document_source.__len__()
 65         except KeyError:
 66             self.data['class_term_count'][class_id] = document_source.__len__()
 67         try:
 68             self.data['class_doc_count'][class_id] += 1
 69         except KeyError:
 70             self.data['class_doc_count'][class_id] = 1
 71         self.total_term_count += document_source.__len__()
 72         self.total_doc_count += 1
 73         self.compute_beta_priors()
 74         return True
 75
 76     def classify(self, document_input):
 77         if not self.total_doc_count: raise ClassifierNotTrainedException()
 78
 79         term_freq_matrix = Counter(document_input)
 80         arg_max_matrix = []
 81         for class_id in self.data['class_doc_count']:
 82             summation = 0
 83             for term in document_input:
 84                 try:
 85                     conditional_probability = (self.term_count_store[class_id][term] + 1)
 86                     conditional_probability = conditional_probability / (self.data['class_term_count'][class_id] + self.total_doc_count)
 87                     summation += term_freq_matrix[term] * math.log(conditional_probability)
 88                 except KeyError:
 89                     break
 90             arg_max = summation + self.data['beta_priors'][class_id]
 91             arg_max_matrix.insert(0, (class_id, arg_max))
 92         arg_max_matrix.sort(key=lambda x:x[1])
 93         return (arg_max_matrix[-1][0], arg_max_matrix[-1][1])
 94
 95     def compute_beta_priors(self):
 96         if not self.total_doc_count: raise ClassifierNotTrainedException()
 97
 98         for class_id in self.data['class_doc_count']:
 99             tmp = self.data['class_doc_count'][class_id] / self.total_doc_count
100             self.data['beta_priors'][class_id] = math.log(tmp)