Sklearn决策树训练报错：无法将字符串'male'转为浮点数求解决

问题原因 #

训练数据X_train中存在字符串类型的分类特征（如示例中的'male'），而Scikit-learn的DecisionTreeClassifier仅支持数值型输入，因此触发ValueError: could not convert string to float错误。

修复方案 #

需要先对字符串特征做编码处理，将文本转换为模型可识别的数值。以下是两种常用方法及完整修改代码：

方法1：独热编码（适合无序分类特征）

独热编码会为每个分类值生成新的二进制列，避免给分类特征强加顺序关系，是处理无序分类的常用方式。

%%time
from sklearn.tree import DecisionTreeClassifier
from sklearn.model_selection import GridSearchCV
from sklearn.preprocessing import OneHotEncoder
from sklearn.compose import ColumnTransformer
import matplotlib.pyplot as plt

print(cross_val_scores['DecisionTreeClassifier']['best_params'])

# 1. 预处理：对字符串特征做独热编码
# 筛选所有字符串类型的特征列
categorical_cols = X_train.select_dtypes(include=['object']).columns.tolist()
# 创建预处理管道：分类列编码，数值列保持不变
preprocessor = ColumnTransformer(
    transformers=[
        ('cat_encoder', OneHotEncoder(sparse_output=False, drop='first'), categorical_cols)
    ],
    remainder='passthrough'
)
# 转换训练集（仅在训练集上fit，避免数据泄露）
X_train_processed = preprocessor.fit_transform(X_train)

# 2. 初始化并训练模型
decision_tree = DecisionTreeClassifier(
    random_state=RAND_STATE,
    class_weight='balanced',
    max_depth=3
)
decision_tree.fit(X_train_processed, y_train)

# 3. 获取处理后的特征名称，用于绘制重要性
cat_feature_names = preprocessor.named_transformers_['cat_encoder'].get_feature_names_out(categorical_cols)
numeric_feature_names = X_train.select_dtypes(exclude=['object']).columns.tolist()
all_feature_names = list(cat_feature_names) + numeric_feature_names

# 4. 绘制特征重要性
feature_imp = decision_tree.feature_importances_
plt.barh(range(len(feature_imp)), feature_imp)
plt.title('DecisionTreeClassifier Feature Importance')
plt.yticks(range(len(all_feature_names)), all_feature_names)
plt.show()

方法2：标签编码（适合有序分类特征）

如果你的字符串特征是有序的（如'low'/'medium'/'high'），可以用标签编码将其映射为连续数值：

%%time
from sklearn.tree import DecisionTreeClassifier
from sklearn.model_selection import GridSearchCV
from sklearn.preprocessing import LabelEncoder
import matplotlib.pyplot as plt

print(cross_val_scores['DecisionTreeClassifier']['best_params'])

# 1. 预处理：对每个字符串列做标签编码
X_train_processed = X_train.copy()
categorical_cols = X_train_processed.select_dtypes(include=['object']).columns.tolist()
for col in categorical_cols:
    le = LabelEncoder()
    X_train_processed[col] = le.fit_transform(X_train_processed[col])

# 2. 初始化并训练模型
decision_tree = DecisionTreeClassifier(
    random_state=RAND_STATE,
    class_weight='balanced',
    max_depth=3
)
decision_tree.fit(X_train_processed, y_train)

# 3. 绘制特征重要性
feature_imp = decision_tree.feature_importances_
labels = list(X_train_processed.columns)
plt.barh(range(len(feature_imp)), feature_imp)
plt.title('DecisionTreeClassifier Feature Importance')
plt.yticks(range(len(labels)), labels)
plt.show()

注意事项 #

• 预处理时仅在训练集上调用fit，测试集使用transform，防止数据泄露。
• 独热编码会增加特征维度，若分类类别过多需注意维度爆炸问题。

内容的提问来源于stack exchange，提问作者Cycloo