使用Sklearn Pipeline做分类任务时Shap树解释器绘图报错求助

使用Sklearn Pipeline完成泰坦尼克号分类任务后，调用SHAP可视化时出现多种错误，代码及报错如下：

完整代码 #

from sklearn.datasets import fetch_openml
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestClassifier
from sklearn.impute import SimpleImputer
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import OrdinalEncoder

import shap


# -----------------------------------------------------------------------------
# Data
# -----------------------------------------------------------------------------

X, y = fetch_openml("titanic", version=1, as_frame=True, return_X_y=True)

categorical_columns = ["pclass", "sex", "embarked"]
numerical_columns = ["age", "sibsp", "parch", "fare"]

X = X[categorical_columns + numerical_columns]   # [1309, 7] , there is Nan values.
X_train, X_test, y_train, y_test = train_test_split(X, y, stratify=y, random_state=42)


# -----------------------------------------------------------------------------
# Data preprocessing
# -----------------------------------------------------------------------------


categorical_encoder = OrdinalEncoder(
    handle_unknown="use_encoded_value", unknown_value=-1, encoded_missing_value=-1
)

numerical_imputer = SimpleImputer(strategy="mean")


preprocessing = ColumnTransformer(
    [
        ("cat", categorical_encoder, categorical_columns),   
        ("num", numerical_imputer, numerical_columns),
    ],
    verbose_feature_names_out=False,   
)


# -----------------------------------------------------------------------------
# Pipeline
# -----------------------------------------------------------------------------

rf = Pipeline(
    [
        ("preprocess", preprocessing),
        ("classifier", RandomForestClassifier(random_state=42)),
    ]
)


rf.fit(X_train, y_train)


print(f"RF train accuracy: {rf.score(X_train, y_train):.3f}")
print(f"RF test accuracy: {rf.score(X_test, y_test):.3f}")


# -----------------------------------------------------------------------------
# Shap
# -----------------------------------------------------------------------------

explainer = shap.Explainer(rf["classifier"], feature_names=rf["preprocess"].get_feature_names_out())

X_test_processed = rf['preprocess'].transform(X_test)

shap_values = explainer(X_test_processed)

报错情况 #

• 执行shap.summary_plot(shap_values, X_test_processed)：TypeError: 仅整数标量数组可转换为标量索引
• 执行shap.summary_plot(shap_values, X_test_processed, plot_type="bar")：TypeError: 仅整数标量数组可转换为标量索引
• 执行shap.plots.beeswarm(shap_values)：ValueError: 蜂群图不支持绘制具有多维实例的解释结果！
• 执行shap.plots.bar(shap_values)：IndexError: 列表索引超出范围

核心问题原因 #

RandomForestClassifier即使是二分类任务，默认也会输出每个类别的概率，属于多输出模型。因此explainer(X_test_processed)返回的SHAP值是三维数组（样本数 × 特征数 × 类别数），而多数SHAP可视化函数仅支持处理单输出（二维数组）的情况，这是所有报错的根源。

具体修复步骤 #

1. 提取单类别SHAP值
   二分类任务中通常关注正类（比如泰坦尼克号任务的“存活”类）的SHAP值，只需从三维数组中提取对应类别的维度：

   # 先查看类别顺序，确认正类索引
   print(rf["classifier"].classes_)
   # 假设正类是索引1的类别，提取对应SHAP值
   shap_values = shap_values[..., 1]
   

2. 修改SHAP代码块
   更新后的SHAP部分代码如下：

   # -----------------------------------------------------------------------------
   # Shap
   # -----------------------------------------------------------------------------
   
   explainer = shap.Explainer(rf["classifier"], feature_names=rf["preprocess"].get_feature_names_out())
   
   X_test_processed = rf['preprocess'].transform(X_test)
   
   shap_values = explainer(X_test_processed)
   # 提取正类SHAP值，适配可视化要求
   shap_values = shap_values[..., 1]
   

3. 重新运行可视化函数
   现在以下代码均可正常执行：

   • 摘要散点图：

     shap.summary_plot(shap_values, X_test, feature_names=rf["preprocess"].get_feature_names_out())
     

   • 特征重要性条形图：

     shap.summary_plot(shap_values, X_test, plot_type="bar", feature_names=rf["preprocess"].get_feature_names_out())
     

   • 蜂群图：

     shap.plots.beeswarm(shap_values)
     

   • 特征重要性条形图（另一种接口）：

     shap.plots.bar(shap_values)
     

额外优化建议 #

• 传入原始测试集X_test而非预处理后的X_test_processed，可视化时会显示原始特征取值（如性别显示male/female，而非编码后的0/1），可读性更强。

内容的提问来源于stack exchange，提问作者Mohammad