python - 使用 scikit LinearSVC 的预测置信度

Question

我正在使用 LinerSVC 技术对文本进行分类，但我想获得每个预测都附带的预测置信度。

这是我现在拥有的:

    train_set = self.read_training_files()
    count_vect = CountVectorizer()
    X_train_counts = count_vect.fit_transform([e[0] for e in train_set])
    tfidf_transformer = TfidfTransformer()
    X_train_tfidf = tfidf_transformer.fit_transform(X_train_counts)
    clf = LinearSVC(C=1).fit(X_train_tfidf, [e[1] for e in train_set])
    _ = text_clf.fit([e[0] for e in train_set], [e[1] for e in train_set])
    foods = list(self.get_foods())
    lenfoods = len(foods)
    i = 0
    for food in foods:
        fd = self.get_modified_food(food)
        food_desc = fd['fields']['title'].replace(',', '').lower()
        X_new_counts = count_vect.transform([food_desc])
        X_new_tfidf = tfidf_transformer.transform(X_new_counts)
        predicted = clf.predict(X_new_tfidf)

变量“predicted”将包含预测的类别编号，不包括置信度。我一直在阅读源代码 here但我没有找到合适的属性来执行此操作。

Answer 1

我认为您找错地方了 :)。你看过吗:

相关decision function ？

---

就我个人而言，sklearn 中的文档非常有帮助；有时比代码更重要:)