如何解决SVR训练时出现的Input X contains NaN错误？

问题场景 #

使用SVR进行回归预测时，执行以下代码触发ValueError：

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from sklearn.svm import SVR
from sklearn.model_selection import GridSearchCV
from sklearn.model_selection import KFold
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error
from sklearn.metrics import mean_absolute_percentage_error
from sklearn.metrics import mean_absolute_error
import csv
from math import sqrt
from scipy import stats
from sklearn.preprocessing import StandardScaler
from sklearn.metrics import r2_score

filepath = '/content/drive/MyDrive/TESIS/DATA/'

pilihankolom = 35

X = pd.read_csv(filepath+'Data_Radiomic.csv').to_numpy()
Y = pd.read_csv(filepath+'Data_Dosiomic.csv', usecols=[pilihankolom])
print(X.shape)
Y_label = str(Y.columns)
Y = Y.to_numpy()

model=SVR(kernel='rbf', C=10, epsilon=0.01)
kf = KFold(n_splits=4)

X_training, X_testing, Y_training, Y_testing = train_test_split(X, Y, test_size=0.2, random_state=0)
print(X_training.shape, X_testing.shape)

prediction = []
mse_set = []
for train_index, test_index in kf.split(X_training):
  print(train_index, test_index)
  X_train, X_test = X_training[train_index], X_training[test_index]
  Y_train, Y_test = Y_training[train_index], Y_training[test_index]
  model.fit(X_train, Y_train)  # 此处触发错误
  Y_pred = model.predict(X_test)
  mse_set.append(mean_squared_error(Y_test, Y_pred))
  prediction.extend(Y_pred)

错误信息 #

ValueError: 输入X包含NaN。SVR原生不接受以NaN编码的缺失值。对于监督学习，可考虑使用sklearn.ensemble.HistGradientBoostingClassifier和Regressor，它们原生支持NaN缺失值；也可对数据预处理，如用imputer转换器或删除含缺失值样本。

解决方案 #

SVR属于不支持缺失值的模型，必须对数据进行预处理，或更换模型，以下是三种可行方案：

1. 删除含缺失值的样本或特征 #

如果缺失值占比极低，可直接删除对应样本或特征：

# 删除含NaN的行（样本）
X = pd.read_csv(filepath+'Data_Radiomic.csv').dropna(axis=0).to_numpy()
# 或删除含NaN的列（特征）
X = pd.read_csv(filepath+'Data_Radiomic.csv').dropna(axis=1).to_numpy()

注意：删除样本会减少数据集规模，删除特征可能丢失重要信息，需根据数据情况判断。

2. 填充缺失值（推荐） #

使用sklearn的填充器对NaN进行插值，常用策略包括均值、中位数、众数，或基于邻近样本的KNN填充：

示例：均值填充+标准化

from sklearn.impute import SimpleImputer

# 读取数据保留DataFrame格式，方便预处理
X_df = pd.read_csv(filepath+'Data_Radiomic.csv')
Y_df = pd.read_csv(filepath+'Data_Dosiomic.csv', usecols=[pilihankolom])

# 用均值填充NaN
imputer = SimpleImputer(strategy='mean')
X_imputed = imputer.fit_transform(X_df)

# 分割数据集
X_training, X_testing, Y_training, Y_testing = train_test_split(X_imputed, Y_df.to_numpy(), test_size=0.2, random_state=0)

# 标准化（需在填充之后执行）
scaler = StandardScaler()
X_training_scaled = scaler.fit_transform(X_training)
X_testing_scaled = scaler.transform(X_testing)

# 训练模型（注意SVR要求y为一维数组，用ravel()转换）
model=SVR(kernel='rbf', C=10, epsilon=0.01)
kf = KFold(n_splits=4)
prediction = []
mse_set = []
for train_index, test_index in kf.split(X_training_scaled):
  X_train, X_test = X_training_scaled[train_index], X_training_scaled[test_index]
  Y_train, Y_test = Y_training[train_index], Y_training[test_index]
  model.fit(X_train, Y_train.ravel())
  Y_pred = model.predict(X_test)
  mse_set.append(mean_squared_error(Y_test, Y_pred))
  prediction.extend(Y_pred)

可选填充策略：

• strategy='median'：中位数填充，适合含异常值的数据
• strategy='most_frequent'：众数填充，适合分类特征
• KNNImputer(n_neighbors=5)：基于K近邻样本的均值填充，适合缺失值分布有规律的场景

3. 更换支持原生NaN的模型 #

如果不想预处理数据，可使用HistGradientBoostingRegressor，它原生支持含NaN的输入：

from sklearn.ensemble import HistGradientBoostingRegressor

# 替换SVR为HistGradientBoostingRegressor
model = HistGradientBoostingRegressor(random_state=0)

# 后续训练流程无需修改（直接用含NaN的X）
X_training, X_testing, Y_training, Y_testing = train_test_split(X, Y, test_size=0.2, random_state=0)
kf = KFold(n_splits=4)
prediction = []
mse_set = []
for train_index, test_index in kf.split(X_training):
  X_train, X_test = X_training[train_index], X_training[test_index]
  Y_train, Y_test = Y_training[train_index], Y_training[test_index]
  model.fit(X_train, Y_train.ravel())
  Y_pred = model.predict(X_test)
  mse_set.append(mean_squared_error(Y_test, Y_pred))
  prediction.extend(Y_pred)

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