我在使用SVM进行预测时遇到了以下错误

AttributeError: 'numpy.ndarray' object has no attribute 'lower'

在执行代码行 text_clf.fit(X_train,y_train) 时出现此错误。如何解决这个问题，并使用SVM获取预测正确的概率？

我正在根据输入文件中其余列的值预测第一列（gold）的值。我的输入文件 dataExtended.txt 的格式如下：

   gold,T-x-T,T-x-N,T-x-U,T-x-NT,T-x-UT,T-x-UN,T-x-UNT,N-x-T,N-x-N,N-x-U,N-x-NT,N-x-UT,N-x-UN,N-x-UNT,U-x-T,U-x-N,U-x-U,U-x-NT,U-x-UT,U-x-UN,U-x-UNT,NT-x-T,NT-x-N,NT-x-U,NT-x-NT,NT-x-UT,NT-x-UN,NT-x-UNT,UT-x-T,UT-x-N,UT-x-U,UT-x-NT,UT-x-UT,UT-x-UN,UT-x-UNT,UN-x-T,UN-x-N,UN-x-U,UN-x-NT,UN-x-UT,UN-x-UN,UN-x-UNT,UNT-x-T,UNT-x-N,UNT-x-U,UNT-x-NT,UNT-x-UT,UNT-x-UN,UNT-x-UNT,T-T-x,T-N-x,T-U-x,T-NT-x,T-UT-x,T-UN-x,T-UNT-x,N-T-x,N-N-x,N-U-x,N-NT-x,N-UT-x,N-UN-x,N-UNT-x,U-T-x,U-N-x,U-U-x,U-NT-x,U-UT-x,U-UN-x,U-UNT-x,NT-T-x,NT-N-x,NT-U-x,NT-NT-x,NT-UT-x,NT-UN-x,NT-UNT-x,UT-T-x,UT-N-x,UT-U-x,UT-NT-x,UT-UT-x,UT-UN-x,UT-UNT-x,UN-T-x,UN-N-x,UN-U-x,UN-NT-x,UN-UT-x,UN-UN-x,UN-UNT-x,UNT-T-x,UNT-N-x,UNT-U-x,UNT-NT-x,UNT-UT-x,UNT-UN-x,UNT-UNT-x,x-T-T,x-T-N,x-T-U,x-T-NT,x-T-UT,x-T-UN,x-T-UNT,x-N-T,x-N-N,x-N-U,x-N-NT,x-N-UT,x-N-UN,x-N-UNT,x-U-T,x-U-N,x-U-U,x-U-NT,x-U-UT,x-U-UN,x-U-UNT,x-NT-T,x-NT-N,x-NT-U,x-NT-NT,x-NT-UT,x-NT-UN,x-NT-UNT,x-UT-T,x-UT-N,x-UT-U,x-UT-NT,x-UT-UT,x-UT-UN,x-UT-UNT,x-UN-T,x-UN-N,x-UN-U,x-UN-NT,x-UN-UT,x-UN-UN,x-UN-UNT,x-UNT-T,x-UNT-N,x-UNT-U,x-UNT-NT,x-UNT-UT,x-UNT-UN,x-UNT-UNT,callersAtLeast1T,CalleesAtLeast1T,callersAllT,calleesAllT,CallersAtLeast1N,CalleesAtLeast1N,CallersAllN,CalleesAllN,childrenAtLeast1T,parentsAtLeast1T,childrenAtLeast1N,parentsAtLeast1N,childrenAllT,parentsAllT,childrenAllN,ParentsAllN,ParametersatLeast1T,FieldMethodsAtLeast1T,ReturnTypeAtLeast1T,ParametersAtLeast1N,FieldMethodsAtLeast1N,ReturnTypeN,ParametersAllT,FieldMethodsAllT,ParametersAllN,FieldMethodsAllN,ClassGoldN,ClassGoldT,Inner,Leaf,Root,Isolated,EmptyCallers,EmptyCallees,EmptyCallersCallers,EmptyCalleesCallees,Program,Requirement,MethodIDT,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,chess,1,1N,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,1,0,0,0,0,0,0,0,chess,2,1N,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,1,0,0,0,0,0,0,0,chess,3,1N,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,chess,4,1N,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,1,0,0,0,0,0,0,0,chess,5,1N,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,chess,6,1N,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,chess,7,1N,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,1,0,0,0,0,0,0,0,chess,8,1N,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,chess,1,3N,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,1,0,0,0,0,0,0,0,chess,2,3N,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,1,0,0,0,0,0,0,0,chess,3,3N,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,0,0,0,0,0,0,0,chess,4,3N,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,1,1,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,1,0,1,0,0,0,0,0,0,0,chess,5,3

这是我的完整可复现代码：

# 使用朴素贝叶斯对Iris数据集进行预测from sklearn.cross_validation import train_test_split from sklearn import metricsimport pandas as pd import numpy as npimport seaborn as sns; sns.set()from sklearn.metrics import confusion_matrix from sklearn.metrics import accuracy_score from sklearn.metrics import classification_report import seaborn as snsfrom sklearn import svmfrom sklearn.svm import LinearSVCfrom sklearn.feature_extraction.text import TfidfVectorizerfrom sklearn.pipeline import Pipelinedata = pd.read_csv( 'dataExtended.txt', sep= ',') row_count, column_count = data.shape    # 打印数据集形状 print ("数据集长度: ", len(data)) print ("数据集形状: ", data.shape) print("列数 ", column_count)    # 打印数据集观察值 print ("数据集: ",data.head()) data['gold'] = data['gold'].astype('category').cat.codesdata['Program'] = data['Program'].astype('category').cat.codes    # 构建阶段 分离目标变量 X = data.values[:, 1:column_count] Y = data.values[:, 0]     # 将数据集分割成训练集和测试集 X_train, X_test, y_train, y_test = train_test_split( X, Y, test_size = 0.3, random_state = 100)     #创建一个SVM分类器svclassifier = svm.LinearSVC()print('拟合前')svclassifier.fit(X_train, y_train)predicted = svclassifier.predict(X_test)text_clf = Pipeline([('tfidf',TfidfVectorizer()),('clf',LinearSVC())])text_clf.fit(X_train,y_train)

导致错误的跟踪信息：

Traceback (most recent call last):  File "<ipython-input-9-8e85a0a9f81c>", line 1, in <module>    runfile('C:/Users/mouna/ownCloud/Mouna Hammoudi/dumps/Python/Paper4SVM.py', wdir='C:/Users/mouna/ownCloud/Mouna Hammoudi/dumps/Python')  File "C:\Users\mouna\Anaconda3\lib\site-packages\spyder_kernels\customize\spydercustomize.py", line 668, in runfile    execfile(filename, namespace)  File "C:\Users\mouna\Anaconda3\lib\site-packages\spyder_kernels\customize\spydercustomize.py", line 108, in execfile    exec(compile(f.read(), filename, 'exec'), namespace)  File "C:/Users/mouna/ownCloud/Mouna Hammoudi/dumps/Python/Paper4SVM.py", line 53, in <module>    text_clf.fit(X_train,y_train)  File "C:\Users\mouna\Anaconda3\lib\site-packages\sklearn\pipeline.py", line 248, in fit    Xt, fit_params = self._fit(X, y, **fit_params)  File "C:\Users\mouna\Anaconda3\lib\site-packages\sklearn\pipeline.py", line 213, in _fit    **fit_params_steps[name])  File "C:\Users\mouna\Anaconda3\lib\site-packages\sklearn\externals\joblib\memory.py", line 362, in __call__    return self.func(*args, **kwargs)  File "C:\Users\mouna\Anaconda3\lib\site-packages\sklearn\pipeline.py", line 581, in _fit_transform_one    res = transformer.fit_transform(X, y, **fit_params)  File "C:\Users\mouna\Anaconda3\lib\site-packages\sklearn\feature_extraction\text.py", line 1381, in fit_transform    X = super(TfidfVectorizer, self).fit_transform(raw_documents)  File "C:\Users\mouna\Anaconda3\lib\site-packages\sklearn\feature_extraction\text.py", line 869, in fit_transform    self.fixed_vocabulary_)  File "C:\Users\mouna\Anaconda3\lib\site-packages\sklearn\feature_extraction\text.py", line 792, in _count_vocab    for feature in analyze(doc):  File "C:\Users\mouna\Anaconda3\lib\site-packages\sklearn\feature_extraction\text.py", line 266, in <lambda>    tokenize(preprocess(self.decode(doc))), stop_words)  File "C:\Users\mouna\Anaconda3\lib\site-packages\sklearn\feature_extraction\text.py", line 232, in <lambda>    return lambda x: strip_accents(x.lower())

回答：

你不能对数值数据使用TF-IDF相关的方法；该方法专用于文本数据，因此它使用了诸如 .tolower() 的方法，这些方法默认适用于字符串，因此会引发错误。从文档中已经可以看出这一点：

fit(self, raw_documents, y=None)

从训练集中学习词汇和idf。

参数

raw_documents: iterable

一个可迭代对象，它会生成str、unicode或文件对象。

我担心你解释的理由，如评论中所述：

我只是想得到每个预测正确的概率，使用SVM时似乎只有TF-IDF能做到这一点

非常薄弱。首先，没有所谓的“每个预测正确的概率” – 我认为你指的是概率预测，与硬类预测相反（参见 预测类别还是类概率？）

回到你实际需求的重点：与你在这里使用的 LinearSVC 不同，SVC 确实提供了 predict_proba 方法，应该能完成这项工作（参见文档及其中的说明）。请注意，LinearSVC实际上不是 SVM – 有关详细信息，请参阅在什么参数下scikit-learn中的SVC和LinearSVC是等价的？的回答。

简而言之，忘记TF-IDF，转而使用SVC而不是LinearSVC。