解决Scikit-learn管道转换DataFrame时的列数不匹配问题

# ... (前面的数据准备和管道定义) ...

# 原始列名列表，长度为17
names = X_train.columns.tolist()

# ... (ColumnTransformer和Pipeline定义) ...

# 在FunctionTransformer中，尝试用原始列名列表创建DataFrame
pipe_preprocessor = Pipeline(steps = [
    ("preprocessor", preprocessor),
    ("pandarizer", FunctionTransformer(lambda x: pd.DataFrame(x, columns = names))) # 错误发生在这里
]).fit(X_train)

from sklearn.preprocessing import FunctionTransformer, OneHotEncoder, LabelEncoder
from sklearn.impute import SimpleImputer
from sklearn.pipeline import Pipeline
from sklearn.compose import ColumnTransformer
import pandas as pd
from sklearn.model_selection import train_test_split

print("step1: import lib")
print("step2: loading raw data")
df = pd.read_csv("online_shoppers_intention.csv")

print("step3: data preparition")
X = df.drop(['Revenue'], axis = 1)
y = df['Revenue']

print("step4: data splitting")
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size = .3, random_state = 0)
# 原始列名列表 (不再直接用于pd.DataFrame的columns参数)
original_names = X_train.columns.tolist()

numeric_transformer = SimpleImputer(strategy = 'constant')
categorical_transformer = OneHotEncoder(handle_unknown = 'ignore')

numerical_cols = X.select_dtypes(exclude = "object").columns.values.tolist()
categorical_cols = X.select_dtypes(exclude = ['int', 'float64', 'bool']).columns.values.tolist()

preprocessor = ColumnTransformer(
    transformers=[
    ('num', numeric_transformer, numerical_cols)
    ,('cat', categorical_transformer, categorical_cols)
    ],
    remainder = 'passthrough',
    verbose_feature_names_out=False # 避免在旧版本sklearn中出现警告，新版本默认为True
)

# 拟合管道以获取转换后的特征名
pipe_preprocessor_fitted = Pipeline(steps = [("preprocessor", preprocessor)]).fit(X_train)

# 获取转换后的所有特征名称
# 注意：在sklearn 1.0+版本中，ColumnTransformer的get_feature_names_out()方法可以直接使用
# 对于更早的版本，可能需要从fit_transform的结果中推断，或者升级sklearn
transformed_feature_names = pipe_preprocessor_fitted.named_steps['preprocessor'].get_feature_names_out()

# 定义一个FunctionTransformer，使用动态获取的列名
pandarizer_transformer = FunctionTransformer(lambda x: pd.DataFrame(x, columns = transformed_feature_names))

# 构建最终的管道
final_pipeline = Pipeline(steps = [
    ("preprocessor", preprocessor),
    ("pandarizer", pandarizer_transformer)
])

# 拟合并转换数据
X_train_pipe = final_pipeline.fit_transform(X_train)
X_test_pipe = final_pipeline.transform(X_test)

print(f"转换后训练集形状: {X_train_pipe.shape}")
print(f"转换后训练集列名数量: {len(X_train_pipe.columns)}")
print(X_train_pipe.head())

from sklearn.preprocessing import LabelEncoder

# ... (前面的数据准备) ...

# 将OneHotEncoder替换为LabelEncoder
# 注意：LabelEncoder通常一次处理一个特征，或者需要对每个分类列单独应用
# 如果在ColumnTransformer中使用，需要将其包装在一个自定义转换器中，或者对每个列应用
# 更直接的方式是：
# categorical_transformer = Pipeline(steps=[
#     ('imputer', SimpleImputer(strategy='most_frequent')), # 如果分类特征有缺失值
#     ('label_encoder', FunctionTransformer(lambda x: LabelEncoder().fit_transform(x.ravel()).reshape(-1, 1), validate=False))
# ])
# 实际操作中，ColumnTransformer与LabelEncoder的组合会更复杂，通常会选择OneHotEncoder并处理列名。
# 这里仅强调LabelEncoder不增加列数这一特性。