如何解决“无法将字符串‘lemonde’转换为浮点数”错误

问题场景 #

我有一个名为Source的列，先去除其中的“-”字符（比如把“le-monde”处理成“lemonde”），但执行astype(float)转换时出现错误：could not convert string to float: 'lemonde'，用该列训练MultinomialNB模型也因此受阻，相关代码如下：

from sklearn.naive_bayes import MultinomialNB
from datetime import datetime

##change date 
TrainDF['Date'] = pd.to_numeric(pd.to_datetime(TrainDF['Date']))
TestDF['Date'] = pd.to_numeric(pd.to_datetime(TestDF['Date']))

##get rid of '-' in column
TrainDF['Source'] = TrainDF['Source'].str.replace('-', '')
TestDF['Source'] = TestDF['Source'].str.replace('-', '')

##convert source to float
TrainDF['Source'] = TrainDF['Source'].astype(float)
TestDF['Source'] = TestDF['Source'].astype(float)  

## Instantiate
MyModelNB= MultinomialNB()

## FIT
MyNB=MyModelNB.fit(TrainDF, TrainLabels)
#print(MyNB.classes_)
#print(MyNB.class_count_)
#print(MyNB.feature_log_prob_)


Prediction = MyModelNB.predict(TestDF)
print(np.round(MyModelNB.predict_proba(TestDF),2))

## COnfusion Matrix Accuracies
cnf_matrix = confusion_matrix(TestLabels, Prediction)
print("\nThe confusion matrix is:")
print(cnf_matrix)

数据集示例：

错误原因 #

1. 纯字母组成的字符串（如"lemonde"）无法直接转换为浮点数，astype(float)仅支持解析可转为数字的字符串，对非数字字符串无效。
2. MultinomialNB（多项式朴素贝叶斯）要求输入特征为非负数值型（如计数、编码后的整数），不能直接传入原始字符串特征。

解决方案 #

方案1：标签编码（LabelEncoder）

将每个唯一字符串映射为整数，适合类别间存在隐含顺序，或希望保持特征维度简洁的场景。注意必须用训练集拟合编码器，再转换测试集，避免数据泄露。

from sklearn.preprocessing import LabelEncoder
from sklearn.naive_bayes import MultinomialNB
import pandas as pd
import numpy as np
from sklearn.metrics import confusion_matrix

# 处理日期列
TrainDF['Date'] = pd.to_numeric(pd.to_datetime(TrainDF['Date']))
TestDF['Date'] = pd.to_numeric(pd.to_datetime(TestDF['Date']))

# 去除Source列的"-"
TrainDF['Source'] = TrainDF['Source'].str.replace('-', '')
TestDF['Source'] = TestDF['Source'].str.replace('-', '')

# 初始化并拟合标签编码器
le = LabelEncoder()
TrainDF['Source'] = le.fit_transform(TrainDF['Source'])
# 用训练集的编码器转换测试集
TestDF['Source'] = le.transform(TestDF['Source'])

# 训练模型
MyModelNB = MultinomialNB()
MyNB = MyModelNB.fit(TrainDF, TrainLabels)

# 预测与评估
Prediction = MyModelNB.predict(TestDF)
print(np.round(MyModelNB.predict_proba(TestDF), 2))

cnf_matrix = confusion_matrix(TestLabels, Prediction)
print("\n混淆矩阵：")
print(cnf_matrix)

方案2：独热编码（OneHotEncoder）

将每个类别转为独立的二进制特征列，适合类别间无顺序关系的场景。通过Pipeline整合预处理与模型训练，避免手动处理数据泄露问题。

from sklearn.preprocessing import OneHotEncoder
from sklearn.compose import ColumnTransformer
from sklearn.pipeline import Pipeline
from sklearn.naive_bayes import MultinomialNB
import pandas as pd
import numpy as np
from sklearn.metrics import confusion_matrix

# 处理日期列
TrainDF['Date'] = pd.to_numeric(pd.to_datetime(TrainDF['Date']))
TestDF['Date'] = pd.to_numeric(pd.to_datetime(TestDF['Date']))

# 去除Source列的"-"
TrainDF['Source'] = TrainDF['Source'].str.replace('-', '')
TestDF['Source'] = TestDF['Source'].str.replace('-', '')

# 定义预处理流程：对Source列独热编码，保留Date列
preprocessor = ColumnTransformer(
    transformers=[
        ('source_onehot', OneHotEncoder(sparse_output=False), ['Source'])
    ],
    remainder='passthrough'
)

# 创建训练管道
pipe = Pipeline(steps=[
    ('preprocessor', preprocessor),
    ('classifier', MultinomialNB())
])

# 拟合模型并预测
pipe.fit(TrainDF, TrainLabels)
Prediction = pipe.predict(TestDF)
print(np.round(pipe.predict_proba(TestDF), 2))

# 输出混淆矩阵
cnf_matrix = confusion_matrix(TestLabels, Prediction)
print("\n混淆矩阵：")
print(cnf_matrix)

内容的提问来源于stack exchange，提问作者user18291032