我想使用 GLMNET 来拟合二项式逻辑回归模型。我可以直接使用插入符号或 glmnet-package。让我们以数据(BinomialExample)为例来执行以下我已经实现的代码:
#rm(list = ls(all.names = TRUE))
library(glmnet)
library(caret)
data(BinomialExample)
y[y==0] = "low"
y[y==1] = "high"
y <- as.factor(y)
#split data in training & validation set
set.seed(1)
splitSample <- createDataPartition(y, p = 0.8, list = FALSE)
training_expression <- x[splitSample,]
training_phenotype <- y[splitSample]
validation_expression <- x[-splitSample,]
validation_phenotype <- y[-splitSample]
#####################
##GLMNET with CARET##
#####################
eGrid <- expand.grid(.alpha=seq(0.1,0.9, by=0.1),.lambda=seq(0,1,by=0.01))
Control <- trainControl(verboseIter=TRUE, classProbs=TRUE, summaryFunction=twoClassSummary, method="cv")
set.seed(1)
netFit <- train(x = training_expression, y = training_phenotype,method = "glmnet", metric = "ROC", tuneGrid=eGrid,trControl = Control)
netFitPerf <- getTrainPerf(netFit)
trainROC <- netFitPerf[,1]
trainSens <- netFitPerf[,2]
trainSpec <- netFitPerf[,3]
trainAlpha <- netFit$bestTune[,1]
trainLambda <- netFit$bestTune[,2]
print(sprintf("ROC: %s Sens: %s Spec: %s Alpha: %s Lambda: %s", round(trainROC,2), round(trainSens,2), round(trainSpec,2), round(trainAlpha,2),round(trainLambda,2)))
predict_validation <- predict(netFit, newdata = validation_expression)
confusionMatrix(predict_validation,validation_phenotype)
######################
#GLMNET without CARET#
######################
set.seed(1)
elasticnet <- cv.glmnet(training_expression, training_phenotype, family = "binomial", type.measure = "class", nfolds=10, alpha=0.5, nlambda = 100)
plot(elasticnet)
predict_validation <- predict(elasticnet, newx = validation_expression, s = c(elasticnet$lambda.min), type = "class")
confusionMatrix(predict_validation,validation_phenotype)
如您所见,如果我使用插入符号包,我可以轻松打印模型的 ROC、灵敏度和特异性。但是,如果我在没有 CARET 的情况下直接使用 glmnet,我无法找到类似的方法来打印 ROC、Sens、Spec - 是否有类似的方法来获取这些指标?
谢谢你的帮助!