Scikit-learn 如何检查模型（例如 TfidfVectorizer）是否已经适合

Question

对于从文本中提取特征，如何检查向量器（例如 TfIdfVectorizer 或 CountVectorizer）是否已经适合训练数据？
特别是，我希望代码能够自动确定矢量化器是否已经适合。

from sklearn.feature_extraction.text import TfidfVectorizer

vectorizer = TfidfVectorizer()

def vectorize_data(texts):
  # if vectorizer has not been already fit
  vectorizer.fit_transform(texts)
  # else
  vectorizer.transform(texts)

Answer 1

您可以使用check_is_fitted，它基本上是为此而设计的。

在source of TfidfVectorizer.transform()你可以查看它的用法：

def transform(self, raw_documents, copy=True):

    # This is what you need.
    check_is_fitted(self, '_tfidf', 'The tfidf vector is not fitted')

    X = super(TfidfVectorizer, self).transform(raw_documents)
    return self._tfidf.transform(X, copy=False)

所以在你的情况下，你可以这样做：

from sklearn.utils.validation import check_is_fitted

def vectorize_data(texts):

    try:
        check_is_fitted(vectorizer, '_tfidf', 'The tfidf vector is not fitted')
    except NotFittedError:
        vectorizer.fit(texts)

    # In all cases vectorizer if fit here, so just call transform()
    vectorizer.transform(texts)

Answer 2

我提出了两种检查方法：

涵盖所有 scikit-learn 模型的个人代码：

import inspect

def my_inspector(model):
    return 0 < len( [k for k,v in inspect.getmembers(model) if k.endswith('_') and not k.startswith('__')] )

现在让我们测试一下这段代码：

from sklearn.feature_extraction.text import TfidfVectorizer
import inspect

vectorizer = TfidfVectorizer()

def my_inspector(model):
        return 0 < len( [k for k,v in inspect.getmembers(model) if k.endswith('_') and not k.startswith('__')] )

my_inspector(vectorizer)
# False

---

第二种方式使用check_is_fitted

from sklearn.utils.validation import check_is_fitted

check_is_fitted(vectorizer, '_tfidf', 'The tfidf vector is not fitted')