一次热编码丢失特征名

Question

使用 onehotencoding 构建管道，当拟合和转换为 training/test 集并转换为数据框时，会导致特征没有名称。有什么方法可以获取每个编码特征的名称吗？

# Numerical column transformer
num_transformer = Pipeline(steps=[
    ('imputer', SimpleImputer(strategy='mean')),
    ('scaler', StandardScaler())
])

# Categorical column transformer
cat_transformer = Pipeline(steps=[
    ('imputer', SimpleImputer(strategy='most_frequent')),
    ('onehot', OneHotEncoder(handle_unknown='ignore'))
])

# Preprocessing pipeline
preprocessor = ColumnTransformer(
    transformers=[
        ('num', num_transformer, numerical_cols),
        ('cat', cat_transformer, categorical_cols)
    ])


# Fitting the data and transforming the training & test set
X_train_preprocessed = preprocessor.fit_transform(X_train)
test_preprocessed = preprocessor.fit_transform(test)

Answer 1

您可以使用 Pipeline 的 named_transformers_ of ColumnTransformer. You have 2 transformers named 'num' and 'cat', so preprocessor.named_transformers_['cat'] gives you access to your cat_transformer. Then using named_steps 属性访问转换器您可以访问名为 'onehot' 的 OneHotEncoder 及其 categories_ 属性：

X = [['Male', 1], ['Female', 3], ['Female', 2]]

preprocessor.fit_transform(X)
Out[6]: 
array([[-1.22474487,  0.        ,  1.        ],
       [ 1.22474487,  1.        ,  0.        ],
       [ 0.        ,  1.        ,  0.        ]])

preprocessor.named_transformers_['cat'].named_steps['onehot'].categories_
Out[7]: [array(['Female', 'Male'], dtype=object)]

一次热编码丢失特征名

Loss of feature names when onehotencoding

python

pipeline

one-hot-encoding