def localWordsTest(city0Rss, city1Rss):
"""
测试根据输入的text分类城市的准确率
:param city0Rss:
:param city1Rss:
过滤垃圾邮件
:return:
"""
initialDocList, fullText, cityTypes = loadRSSText(city0Rss, city1Rss)
voclist = bayes.createVocabularyList(initialDocList)
print "未删除高频词汇的词汇表长度：", len(voclist)
# 出现频率最高的词汇，例如：I and 等辅助词
deletedVoc = calcFrequentWords(voclist, fullText)
# 去除词汇列表的高频词汇
for word in deletedVoc:
if word[0] in voclist:
voclist.remove(word[0])
print "删除后的词汇表长度：", len(voclist)
# 从initialDocList中随机创建10个待测试的文档
testDocList = []
# 待测试邮件的类型
testDocClassList = []
"""
注意此处随机选择10个数据，添加到测试集合，同时将原有的数据集删除，
这种随机选择数据的一部分作为训练集合，而剩余部分作为测试集合的过程称为
留存交叉验证：hold-out cross validation
"""
for i in range(10):
randomIndex = int(random.uniform(0, len(initialDocList)))
testDocClassList.append(cityTypes[randomIndex])
testDocList.append(initialDocList[randomIndex])
del (initialDocList[randomIndex])
del (cityTypes[randomIndex])
errorCount = 0
for j in range(len(testDocList)):
classType = bayes.classifyNavieBayesian2(voclist, initialDocList, cityTypes, testDocList[j])
if classType != testDocClassList[j]: # 预测的结果和实际的结果进行比较
print "分类错误的信息：", testDocList[j], "\n属于", testDocClassList[j], "错误分类成了：", classType
errorCount += 1
# 计算分类的误差
errorRate = float(errorCount) / len(testDocList)
print "the error rate is :", errorRate
return errorRate
