本文实例讲述了Python实现的朴素贝叶斯分类器。分享给大家供大家参考,具体如下:
因工作中需要,自己写了一个朴素贝叶斯分类器。
对于未出现的属性,采取了拉普拉斯平滑,避免未出现的属性的概率为零导致整个条件概率都为零的情况出现。
朴素贝叶斯的基本原理网上很容易查到,这里不再叙述,直接附上代码
因工作中需要,自己写了一个朴素贝叶斯分类器。对于未出现的属性,采取了拉普拉斯平滑,避免未出现的属性的概率为零导致整个条件概率都为零的情况出现。
class NBClassify(object): def __init__(self, fillNa = 1): self.fillNa = 1 pass def train(self, trainSet): # 计算每种类别的概率 # 保存所有tag的所有种类,及它们出现的频次 dictTag = {} for subTuple in trainSet: dictTag[str(subTuple[1])] = 1 if str(subTuple[1]) not in dictTag.keys() else dictTag[str(subTuple[1])] + 1 # 保存每个tag本身的概率 tagProbablity = {} totalFreq = sum([value for value in dictTag.values()]) for key, value in dictTag.items(): tagProbablity[key] = value / totalFreq # print(tagProbablity) self.tagProbablity = tagProbablity ############################################################################## # 计算特征的条件概率 # 保存特征属性基本信息{特征1:{值1:出现5次, 值2:出现1次}, 特征2:{值1:出现1次, 值2:出现5次}} dictFeaturesBase = {} for subTuple in trainSet: for key, value in subTuple[0].items(): if key not in dictFeaturesBase.keys(): dictFeaturesBase[key] = {value:1} else: if value not in dictFeaturesBase[key].keys(): dictFeaturesBase[key][value] = 1 else: dictFeaturesBase[key][value] += 1 # dictFeaturesBase = { # '职业': {'农夫': 1, '教师': 2, '建筑工人': 2, '护士': 1}, # '症状': {'打喷嚏': 3, '头痛': 3} # } dictFeatures = {}.fromkeys([key for key in dictTag]) for key in dictFeatures.keys(): dictFeatures[key] = {}.fromkeys([key for key in dictFeaturesBase]) for key, value in dictFeatures.items(): for subkey in value.keys(): value[subkey] = {}.fromkeys([x for x in dictFeaturesBase[subkey].keys()]) # dictFeatures = { # '感冒 ': {'症状': {'打喷嚏': None, '头痛': None}, '职业': {'护士': None, '农夫': None, '建筑工人': None, '教师': None}}, # '脑震荡': {'症状': {'打喷嚏': None, '头痛': None}, '职业': {'护士': None, '农夫': None, '建筑工人': None, '教师': None}}, # '过敏 ': {'症状': {'打喷嚏': None, '头痛': None}, '职业': {'护士': None, '农夫': None, '建筑工人': None, '教师': None}} # } # initialise dictFeatures for subTuple in trainSet: for key, value in subTuple[0].items(): dictFeatures[subTuple[1]][key][value] = 1 if dictFeatures[subTuple[1]][key][value] == None else dictFeatures[subTuple[1]][key][value] + 1 # print(dictFeatures) # 将驯良样本中没有的项目,由None改为一个非常小的数值,表示其概率极小而并非是零 for tag, featuresDict in dictFeatures.items(): for featureName, fetureValueDict in featuresDict.items(): for featureKey, featureValues in fetureValueDict.items(): if featureValues == None: fetureValueDict[featureKey] = 1 # 由特征频率计算特征的条件概率P(feature|tag) for tag, featuresDict in dictFeatures.items(): for featureName, fetureValueDict in featuresDict.items(): totalCount = sum([x for x in fetureValueDict.values() if x != None]) for featureKey, featureValues in fetureValueDict.items(): fetureValueDict[featureKey] = featureValues/totalCount if featureValues != None else None self.featuresProbablity = dictFeatures ############################################################################## def classify(self, featureDict): resultDict = {} # 计算每个tag的条件概率 for key, value in self.tagProbablity.items(): iNumList = [] for f, v in featureDict.items(): if self.featuresProbablity[key][f][v]: iNumList.append(self.featuresProbablity[key][f][v]) conditionPr = 1 for iNum in iNumList: conditionPr *= iNum resultDict[key] = value * conditionPr # 对比每个tag的条件概率的大小 resultList = sorted(resultDict.items(), key=lambda x:x[1], reverse=True) return resultList[0][0] if __name__ == '__main__': trainSet = [ ({"症状":"打喷嚏", "职业":"护士"}, "感冒 "), ({"症状":"打喷嚏", "职业":"农夫"}, "过敏 "), ({"症状":"头痛", "职业":"建筑工人"}, "脑震荡"), ({"症状":"头痛", "职业":"建筑工人"}, "感冒 "), ({"症状":"打喷嚏", "职业":"教师"}, "感冒 "), ({"症状":"头痛", "职业":"教师"}, "脑震荡"), ] monitor = NBClassify() # trainSet is something like that [(featureDict, tag), ] monitor.train(trainSet) # 打喷嚏的建筑工人 # 请问他患上感冒的概率有多大? result = monitor.classify({"症状":"打喷嚏", "职业":"建筑工人"}) print(result)
另:关于朴素贝叶斯算法详细说明还可参看本站前面一篇//www.jb51.net/article/129903.htm。
更多关于Python相关内容感兴趣的读者可查看本站专题:《Python数据结构与算法教程》、《Python加密解密算法与技巧总结》、《Python编码操作技巧总结》、《Python函数使用技巧总结》、《Python字符串操作技巧汇总》及《Python入门与进阶经典教程》
希望本文所述对大家Python程序设计有所帮助。