R中运行predict函数遇UseMethod错误：SIS-SCAD模型预测问题

问题 #

在高维数据集上运行SIS(Sure Independence screening)-SCAD模型，目标是生成混淆矩阵并计算accuracy、sensitivity、specificity及F1 score等指标，但调用predict函数时出错：

model_prediction <- predict(model,X_test,type="response")

报错信息：

Error in UseMethod("predict") :
no applicable method for 'predict' applied to an object of class "c('integer', 'numeric')"

运行代码如下：

rm(list = ls())

#Installing necessary libraries
library(HiDimDA) #this is where the high-dimensional data is stored
library(SIS)
library(caTools)

data("AlonDS")
set.seed(123)  # Set seed for reproducibility

split <- sample.split(AlonDS$grouping, SplitRatio = 0.8)

train_data <- subset(AlonDS, split == TRUE)
test_data <- subset(AlonDS, split == FALSE)

# Extracting predictor variables and grouping variable
X_train <- train_data[, -(1:6)]  # Exclude non-predictor columns
y_train <- train_data$grouping

X_test <- test_data[, -(1:6)]  # Exclude non-predictor columns
y_test <- test_data$grouping

# Convert grouping variable to factor
y_train <- as.factor(y_train)
y_test <- as.factor(y_test)

# Check the structure of X_train
str(X_train)

# Check the structure of y_train
str(y_train)

# Set greedy to FALSE
greedy <- FALSE

# Standardize the features
X_train_matrix <- scale(X_train)

# Convert grouping variable to numeric (0 and 1)
y_train_numeric <- as.numeric(y_train) - 1

# SIS-SCAD with the full training set
model <- SIS(X_train_matrix, 
             y_train_numeric,
             family = "binomial",
             penalty = "MCP",
             concavity.parameter = 3,
             tune = "bic",
             nfolds = 10,
             type.measure = "deviance",
             gamma.ebic = 1,
             nsis = NULL,
             iter = TRUE,
             iter.max = ifelse(greedy == FALSE, 10, 
                               floor(nrow(X_train_matrix) / log(nrow(X_train_matrix)))), 
             varISIS = "vanilla", 
             perm = FALSE, 
             greedy = FALSE, 
             greedy.size = 1, 
             seed = NULL, 
             standardize = TRUE) 

# Making predictions
model_prediction <- predict(model,X_test,type="response")

# Convert predicted probabilities to class labels
predictions_class <- ifelse(predictions_prob > 0.5, "healthy", "colonc")

# Assuming you have the true labels for comparison
true_labels <- as.character(y_test)

# Convert predictions_class and true_labels to factors with the same levels
predictions_class <- factor(predictions_class, levels = c("healthy", "colonc"))
true_labels <- factor(true_labels, levels = c("healthy", "colonc"))

# Check the levels of predicted and true labels
levels(predictions_class)
levels(true_labels)

# Create a confusion matrix
conf_matrix <- confusionMatrix(predictions_class, true_labels)

# Extract values from the confusion matrix
conf_matrix_values <- as.table(conf_matrix)

# Calculate accuracy
accuracy <- sum(diag(conf_matrix_values)) / sum(conf_matrix_values)
cat("Accuracy:", accuracy, "\n")

# Calculate sensitivity (recall)
sensitivity <- conf_matrix_values[2, 2] / sum(conf_matrix_values[2, ])
cat("Sensitivity (Recall):", sensitivity, "\n")

# Calculate specificity
specificity <- conf_matrix_values[1, 1] / sum(conf_matrix_values[1, ])
cat("Specificity:", specificity, "\n")

# Calculate F1 score
precision <- conf_matrix_values[2, 2] / sum(conf_matrix_values[, 2])
recall <- sensitivity
f1_score <- 2 * (precision * recall) / (precision + recall)
cat("F1 Score:", f1_score, "\n")

问题分析与解决方案 #

错误原因 #

• SIS()返回值非模型对象：SIS包的SIS()函数默认返回筛选出的变量索引（整数向量），而非可用于predict()的模型对象，因此触发“无对应predict方法”的错误。
• 参数缺失：未设置return.model = TRUE，导致函数不返回训练好的模型。
• 惩罚项不符：代码中用penalty = "MCP"，但目标是SCAD模型，参数设置错误。
• 测试集未标准化：训练集做了标准化处理，测试集未同步用相同参数标准化，会导致预测偏差。
• 变量名错误：后续代码中用predictions_prob，但实际赋值的变量是model_prediction，会引发未定义错误。
• 依赖包缺失：confusionMatrix函数来自caret包，原代码未加载。

修正步骤 #

1. 在SIS()中添加return.model = TRUE，让函数返回可用于预测的模型对象。
2. 将penalty = "MCP"改为penalty = "SCAD"，匹配目标模型。
3. 用训练集的标准化参数（均值、标准差）处理测试集，保证数据分布一致。
4. 修正变量名错误，统一使用model_prediction。
5. 加载caret包以使用confusionMatrix函数。

修正后的完整代码 #

rm(list = ls())

# 加载必要库
library(HiDimDA)
library(SIS)
library(caTools)
library(caret)  # 用于生成混淆矩阵

data("AlonDS")
set.seed(123)  # 保证结果可复现

split <- sample.split(AlonDS$grouping, SplitRatio = 0.8)
train_data <- subset(AlonDS, split == TRUE)
test_data <- subset(AlonDS, split == FALSE)

# 提取特征与标签
X_train <- train_data[, -(1:6)]
y_train <- train_data$grouping
X_test <- test_data[, -(1:6)]
y_test <- test_data$grouping

# 转换标签为因子
y_train <- as.factor(y_train)
y_test <- as.factor(y_test)

# 标准化训练集并保存参数
scale_result <- scale(X_train)
X_train_matrix <- scale_result
center_params <- attr(scale_result, "scaled:center")
scale_params <- attr(scale_result, "scaled:scale")

# 转换标签为0-1数值型
y_train_numeric <- as.numeric(y_train) - 1

# 训练SIS-SCAD模型，设置return.model=TRUE返回模型对象
model <- SIS(X_train_matrix, 
             y_train_numeric,
             family = "binomial",
             penalty = "SCAD",  # 改为SCAD惩罚项
             concavity.parameter = 3,
             tune = "bic",
             nfolds = 10,
             type.measure = "deviance",
             gamma.ebic = 1,
             nsis = NULL,
             iter = TRUE,
             iter.max = ifelse(FALSE, 10, 
                               floor(nrow(X_train_matrix) / log(nrow(X_train_matrix)))), 
             varISIS = "vanilla", 
             perm = FALSE, 
             greedy = FALSE, 
             greedy.size = 1, 
             seed = NULL, 
             standardize = TRUE,
             return.model = TRUE)  # 添加该参数返回训练好的模型

# 用训练集的标准化参数处理测试集
X_test_matrix <- scale(X_test, center = center_params, scale = scale_params)

# 生成预测概率
model_prediction <- predict(model, X_test_matrix, type = "response")

# 转换为类别标签
predictions_class <- ifelse(model_prediction > 0.5, "healthy", "colonc")

# 统一标签的因子水平
true_labels <- as.character(y_test)
predictions_class <- factor(predictions_class, levels = c("healthy", "colonc"))
true_labels <- factor(true_labels, levels = c("healthy", "colonc"))

# 生成混淆矩阵并输出
conf_matrix <- confusionMatrix(predictions_class, true_labels)
print(conf_matrix)

# 提取并打印各项指标
accuracy <- conf_matrix$overall["Accuracy"]
sensitivity <- conf_matrix$byClass["Sensitivity"]
specificity <- conf_matrix$byClass["Specificity"]
f1_score <- conf_matrix$byClass["F1"]

cat("\nAccuracy:", accuracy, "\n")
cat("Sensitivity (Recall):", sensitivity, "\n")
cat("Specificity:", specificity, "\n")
cat("F1 Score:", f1_score, "\n")

内容的提问来源于stack exchange，提问作者Ronit