使用Scikit-learn ColumnTransformer转换数据时遇AttributeError求助

在使用Scikit-learn的ColumnTransformer转换数据时触发如下错误：

AttributeError: 'ColumnTransformer' object has no attribute '_feature_names_in'

训练阶段代码 #

feature_columns_names = [
   'transaction_id', 'created_at', 'amount', 'device_model','device_mode',
   'transaction_sum', 'daily_amt_ratio', 'monthly_amt_ratio'
]  
label_column = "is_fraud"
non_scaled_cols = ['created_at','device_model','device_mode','transaction_id','is_fraud']
numeric_features = [col for col in list(feature_columns_names) if col not in non_scaled_cols]
categorical_features = ['device_model','device_mode']

numeric_transformer = make_pipeline(
    SimpleImputer(strategy="constant", fill_value=0), 
    StandardScaler())

categorical_transformer = make_pipeline(
    SimpleImputer(strategy="constant", fill_value="unknown"),
    OneHotEncoder(handle_unknown="ignore"),
)

preprocessor = ColumnTransformer(
    transformers=[
        ('num', numeric_transformer, numeric_features),            
        ('cat', categorical_transformer, categorical_features)],
        remainder="drop")

preprocessor.fit(data)

joblib.dump(preprocessor, os.path.join(args.model_dir, "model.joblib"))

推理阶段代码 #

feature_columns_dtype = {
'transaction_id'                     :'object',
'created_at'                         :'object' ,
'amount'                             :'float64',
'device_model'                       :'object' ,
'device_mode'             :           'object' ,
'transaction_sum'          :          'float64',
'daily_amt_ratio'              :      'float64',
'monthly_amt_ratio'              :    'float64',
}

label_column_dtype = {"is_fraud": "int64"}  


def merge_two_dicts(x, y):
    z = x.copy()  # start with x's keys and values
    z.update(y)  # modifies z with y's keys and values & returns None
    return z

df = pd.read_csv('s3://data/dataset_sample.csv',
                 header=None, 
                 names=feature_columns_names + [label_column],
                 dtype=merge_two_dicts(feature_columns_dtype, label_column_dtype))

if len(df.columns) == len(feature_columns_names) + 1:
    # This is a labelled example, includes the ring label
    df.columns = feature_columns_names + [label_column]
elif len(df.columns) == len(feature_columns_names):
    # This is an unlabelled example.
    df.columns = feature_columns_names

model = joblib.load(os.path.join(model_dir, "model.joblib"))

model.transform(df)

已确认训练与推理的Scikit-learn版本一致，特征名称匹配，数据加载正常，调用model.transform(df)时触发错误，求原因及解决办法。

核心原因 #

这个错误本质是训练时ColumnTransformer拟合的不是带列名的Pandas DataFrame，而是numpy数组或无列名的结构，导致模型实例未保存_feature_names_in_属性（错误信息中少写了末尾下划线，但本质是模型未记录特征列名）。

当推理时传入带列名的DataFrame，ColumnTransformer会尝试校验输入列名与训练时是否一致，但因训练时无列名记录，就会触发属性不存在的错误。

验证方法 #

检查训练阶段的data变量：如果是numpy数组、未设置列名的DataFrame，或者是用df.values这类方式传入的无列名数据，就会导致该问题。

解决办法 #

1. 确保训练时传入带列名的DataFrame
   训练阶段preprocessor.fit(data)中的data必须是带有正确列名的Pandas DataFrame，这样ColumnTransformer才会自动记录_feature_names_in_属性。

2. 推理时只传入模型需要的特征列
   直接传入ColumnTransformer配置中用到的特征列，跳过列名校验步骤：

   # 只保留训练时用到的特征列
   input_cols = numeric_features + categorical_features
   model.transform(df[input_cols])
   

   因为原配置中remainder="drop"会丢弃其他列，所以直接传入目标列即可避免触发列名校验错误。

3. 补充训练时的列名记录（可选）
   如果无法修改训练代码，可以手动给加载后的模型添加_feature_names_in_属性：

   model = joblib.load(os.path.join(model_dir, "model.joblib"))
   model._feature_names_in_ = np.array(feature_columns_names)
   model.transform(df)
   

内容的提问来源于stack exchange，提问作者MSS