拟合ML模型触发TypeError：特征名称类型混合问题求助

运行线性回归模型拟合时触发以下错误：

TypeError: Feature names are only supported if all input features have string names, but your input has ['int', 'str'] as feature name / column name types. If you want feature names to be stored and validated, you must convert them all to strings, by using X.columns = X.columns.astype(str) for example. Otherwise you can remove feature / column names from your input data, or convert them all to a non-string data type.

完整代码如下：

import pandas as pd, numpy as np
import csv
import warnings
from bs4 import BeautifulSoup, MarkupResemblesLocatorWarning
from sklearn.impute import SimpleImputer
from sklearn.exceptions import ConvergenceWarning
from sklearn.feature_extraction.text import TfidfVectorizer
from sklearn.preprocessing import LabelEncoder
from sklearn.linear_model import LinearRegression, LogisticRegression, Perceptron
from sklearn.tree import DecisionTreeClassifier
from sklearn.metrics import mean_squared_error, r2_score, accuracy_score, confusion_matrix, ConfusionMatrixDisplay 
import seaborn as sns
import matplotlib.pyplot as plt

## Reading the data
train_url = 'https://github.com/Rakesh9100/ML-Project-Drug-Review-Dataset/raw/main/datasets/drugsComTrain_raw.tsv'
test_url = 'https://github.com/Rakesh9100/ML-Project-Drug-Review-Dataset/raw/main/datasets/drugsComTest_raw.tsv'

dtypes = { 'Unnamed: 0': 'int32', 'drugName': 'category', 'condition': 'category', 'review': 'category', 'rating': 'float16', 'date': 'string', 'usefulCount': 'int16' }
train_df = pd.read_csv(train_url, sep='\t', quoting=2, dtype=dtypes, parse_dates=['date'])

train_df = train_df.sample(frac=0.8, random_state=42)
test_df = pd.read_csv(test_url, sep='\t', quoting=2, dtype=dtypes, parse_dates=['date'])

## Extracting day, month, and year into separate columns
for df in [train_df, test_df]:
    df['day'] = df['date'].dt.day.astype('int8')
    df['month'] = df['date'].dt.month.astype('int8')
    df['year'] = df['date'].dt.year.astype('int16')

## Suppressing MarkupResemblesLocatorWarning, FutureWarning and ConvergenceWarning
warnings.filterwarnings('ignore', category=MarkupResemblesLocatorWarning)
warnings.simplefilter(action='ignore', category=FutureWarning)
warnings.filterwarnings("ignore", category=ConvergenceWarning)

## Defining function to decode HTML-encoded characters
def decode_html(text):
    decoded_text = BeautifulSoup(text, 'html.parser').get_text()
    return decoded_text

## Applying the function to the review column
train_df['review'], test_df['review'] = train_df['review'].apply(decode_html), test_df['review'].apply(decode_html)

## Dropped the original date column and removed the useless column
train_df, test_df = [df.drop('date', axis=1).drop(df.columns[0], axis=1) for df in (train_df, test_df)]

## Handling the missing values
train_imp, test_imp = [pd.DataFrame(SimpleImputer(strategy='most_frequent').fit_transform(df)) for df in (train_df, test_df)]

## Assigning old column names
train_imp.columns = ['drugName', 'condition', 'review', 'rating', 'usefulCount', 'day', 'month', 'year']
test_imp.columns = ['drugName', 'condition', 'review', 'rating', 'usefulCount', 'day', 'month', 'year']

## Converting the text in the review column to numerical data
vectorizer = TfidfVectorizer(stop_words='english', max_features=3000)
train_reviews = vectorizer.fit_transform(train_imp['review'])
test_reviews = vectorizer.transform(test_imp['review'])

## Replacing the review column with the numerical data
train_imp.drop('review', axis=1, inplace=True)
test_imp.drop('review', axis=1, inplace=True)
train_imp = pd.concat([train_imp, pd.DataFrame(train_reviews.toarray())], axis=1)
test_imp = pd.concat([test_imp, pd.DataFrame(test_reviews.toarray())], axis=1)

## Encoding the categorical columns
for i in ["drugName", "condition"]:
    train_imp[i] = LabelEncoder().fit_transform(train_imp[i])
    test_imp[i] = LabelEncoder().fit_transform(test_imp[i])

## Converting the data types of columns to reduce the memory usage
train_imp, test_imp = train_imp.astype('float16'), test_imp.astype('float16')
train_imp[['drugName', 'condition', 'usefulCount', 'year']] = train_imp[['drugName', 'condition', 'usefulCount', 'year']].astype('int16')
test_imp[['drugName', 'condition', 'usefulCount', 'year']] = test_imp[['drugName', 'condition', 'usefulCount', 'year']].astype('int16')
train_imp[['rating']] = train_imp[['rating']].astype('float16')
test_imp[['rating']] = test_imp[['rating']].astype('float16')
train_imp[['day', 'month']] = train_imp[['day', 'month']].astype('int8')
test_imp[['day', 'month']] = test_imp[['day', 'month']].astype('int8')

#print(train_imp.iloc[:,:15].dtypes)
#print(test_imp.iloc[:,:15].dtypes)

## Splitting the train and test datasets into feature variables
X_train, Y_train = train_imp.drop('rating', axis=1), train_imp['rating']
X_test, Y_test = test_imp.drop('rating', axis=1), test_imp['rating']

##### LinearRegression regression algorithm #####

linear=LinearRegression()
linear.fit(X_train, Y_train)
line_train=linear.predict(X_train)
line_test=linear.predict(X_test)

报错堆栈：

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-23-e787292883ec> in <cell line: 89>()
     87 
     88 linear=LinearRegression()
---> 89 linear.fit(X_train, Y_train)
     90 line_train=linear.predict(X_train)
     91 line_test=linear.predict(X_test)

3 frames
/usr/local/lib/python3.10/dist-packages/sklearn/utils/validation.py in _get_feature_names(X)
   1901     # mixed type of string and non-string is not supported
   1902     if len(types) > 1 and "str" in types:
-> 1903         raise TypeError(
   1904             "Feature names are only supported if all input features have string names, "
   1905             f"but your input has {types} as feature name / column name types. "

TypeError: Feature names are only supported if all input features have string names, but your input has ['int', 'str'] as feature name / column name types. If you want feature names to be stored and validated, you must convert them all to strings, by using X.columns = X.columns.astype(str) for example. Otherwise you can remove feature / column names from your input data, or convert them all to a non-string data type.

错误源于特征列名类型不统一：

• 原数据集的列名（如drugName、condition）是字符串类型
• 将TF-IDF矩阵转为DataFrame时，默认生成的列名是整数（0、1、2...）
• 拼接后，特征集的列名同时存在字符串和整数两种类型，触发Scikit-learn的特征名校验机制。

提供两种简单可行的修复方式：

方式1：统一将所有列名转为字符串 #

在分割特征与标签后，添加代码将列名全部转为字符串类型：

## Splitting the train and test datasets into feature variables
X_train, Y_train = train_imp.drop('rating', axis=1), train_imp['rating']
X_test, Y_test = test_imp.drop('rating', axis=1), test_imp['rating']

# 新增：统一列名为字符串
X_train.columns = X_train.columns.astype(str)
X_test.columns = X_test.columns.astype(str)

##### LinearRegression regression algorithm #####
linear=LinearRegression()
linear.fit(X_train, Y_train)

方式2：移除列名（直接使用数组输入） #

Scikit-learn模型支持无列名的数组输入，只需在拟合和预测时将DataFrame转为数组：

##### LinearRegression regression algorithm #####
linear=LinearRegression()
# 使用.values或.to_numpy()跳过列名校验
linear.fit(X_train.values, Y_train.values)
line_train=linear.predict(X_train.values)
line_test=linear.predict(X_test.values)

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