Question

以下是用于提取二进制分类信息最多的特征的代码对于每个课程：

def most_informative_feature_for_binary_classification(vectorizer, classifier, n=100):
    """
    Identify most important features if given a vectorizer and binary classifier.
    Set n to the number of weighted features you would like to show.
    """
    def read_counter():
        return loads(open("counter.json", "r").read()) + 1 if path.exists("counter.json") else 0

    def write_counter():
        with open("counter.json", "w") as f:
            f.write(dumps(counter))

    counter = 1

    counter = read_counter()
    atexit.register(write_counter)
    if counter >= 7:
        counter = 0

    # additional stopwords to be remove
    # Open a file and read it into memory
    file = open('..\stopwords.txt')
    additional_stopwords = file.read()
    additional_stopwords = additional_stopwords.split()

    class_labels = classifier.classes_
    feature_names = vectorizer.get_feature_names()
    feature_names = [word for word in feature_names if word not in additional_stopwords]

    topn_class1 = sorted(zip(classifier.coef_[0], feature_names))[:n]
    topn_class2 = sorted(zip(classifier.coef_[0], feature_names))[-n:]

    # class_labels = category
    # coef = co-effecient
    # feat = most informative feature
    if 1 <= counter <= 6:
        for coef, feat in topn_class1:
            print(class_labels[counter - 1], coef, feat)
            # print(class_labels)
            # -> output: [2 3 4 5 6 7 8] index of this 0 1 2 3 4 5 6

        print()

        for coef, feat in reversed(topn_class2):
            print(class_labels[counter], coef, feat)

    else:
        print("=== PLEASE RUN PROGRAM AGAIN TO VIEW THE CO-EFFICIENT FOR THE CHOSEN MODEL ===")

以下是most_informative_feature_for_binary_classification方法的输出：

程序首次运行时（它将显示2和3类信息功能）：

2 -8.322094697329087 aaa
2 -8.322094697329087 aaa cm
2 -8.322094697329087 aaa cm underwent
2 -8.322094697329087 aaa free
2 -8.322094697329087 aaa free ivc

3 -8.010764835561018 assymetry imp giddiness
3 -8.144858449457846 admitted feb year
3 -8.164330364141858 agreeable dre brown
3 -8.172447581146958 aerobic anaerobic labeled
3 -8.180391164585233 actually body

程序第二次运行时（它将显示第3类和第4类的信息功能）：

3 -8.322580751462969 aaa
3 -8.322580751462969 aaa cm
3 -8.322580751462969 aaa cm underwent
3 -8.322580751462969 aaa free
3 -8.322580751462969 aaa free ivc

4 -8.0112508896949 assymetry moving
4 -8.145344503591728 admitted feb year
4 -8.16481641827574 agreeable dre brown
4 -8.17293363528084 aerobic anaerobic labeled
4 -8.180877218719115 actually body

程序第三次运行时（它将显示第4类和第5类信息功能）：

4 -8.322337753927105 aaa
4 -8.322337753927105 aaa cm
4 -8.322337753927105 aaa cm underwent
4 -8.322337753927105 aaa free
4 -8.322337753927105 aaa free ivc

5 -8.011007892159036 assymetry imp
5 -8.145101506055864 admitted frequent falls
5 -8.164573420739876 agreeable early review
5 -8.172690637744976 af anticoagulation
5 -8.18063422118325 actually body

如果你们意识到以上结果，它们都是一样的。请帮助我检查most_informative_feature_binary_classification方法代码的代码。谢谢：（（

counter.json的示例：

以下是利用most_informative_feature_binary_classification方法的分类器：

朴素贝叶斯分类器：

def NB_func():

    X_train, X_test, y_train, y_test = train_test_split(df['content'], df['cat_id'], random_state=0)
    count_vect = CountVectorizer()
    X_train_counts = count_vect.fit_transform(X_train)
    tfidf_transformer = TfidfTransformer()
    X_train_tfidf = tfidf_transformer.fit_transform(X_train_counts)
    clf_NB = MultinomialNB().fit(X_train_tfidf, y_train)

    # save the model to disk
    filename = '../dataanalysis/models/Naive_Bayes.sav'
    pickle.dump(clf_NB, open(filename, 'wb'))

    print()
    '''Print the prediction of the category from the unknown document'''
    #For now its not accurate due to insufficient sample data
    #print("NAIVE BAYES CLASSIFIER: ", clf_NB.predict(count_vect.transform([""])))
    print ()

    print("===============================================")
    print("================= NAIVE BAYES =================")
    print("===============================================")
    most_informative_feature_for_binary_classification(tfidf, clf_NB, n=5)

Logistic回归分类器：

def LR_func():

    X_train, X_test, y_train, y_test = train_test_split(df['content'], df['cat_id'], random_state=0)
    count_vect = CountVectorizer()
    X_train_counts = count_vect.fit_transform(X_train)
    tfidf_transformer = TfidfTransformer()
    X_train_tfidf = tfidf_transformer.fit_transform(X_train_counts)
    clf_LR = LogisticRegression().fit(X_train_tfidf, y_train)

    print()
    '''Print the prediction of the category from the unknown document'''
    #For now its not accurate due to insufficient sample data
    #print("LOGISTICS REGRESSION CLASSIFIER: ", clf_LR.predict(count_vect.transform([""])))
    print ()

    # save the model to disk
    filename = '../dataanalysis/models/Logistics_Regression.sav'
    pickle.dump(clf_LR, open(filename, 'wb'))

    print()

    print("===============================================")
    print("============ LOGISTICS REGRESSION =============")
    print("===============================================")
    most_informative_feature_for_binary_classification(tfidf, clf_LR, n=5)

对于二进制分类中最有用的功能，为什么我的结果相同？

0 个答案: