在 R 中使用神经网络方法进行超参数调整

Question

I'm using Wisconsin Breast Cancer Dataset for classification problem: diagnosis of malignant or benign tumors. The dataset has 33 features, all features are numerical besides diagnosis (factor). I'm trying to build DNN model and tune hyperparameters of neuralnet method in train function , but when I ran the model I get an error: "Error: wrong model type for classification" . What should I do in this case?

set.seed(1)
library(neuralnet)

grid <- expand.grid(layer1=c(1:20), layer2=c(1:20), layer3=c(1:20))
DNN <- train(diagnosis ~., data = train.df, method = "neuralnet", linear.output = FALSE,
         tuneGrid = grid, metric = "Kappa", trControl=train_control, 
         maxit = 500, allowParallel = TRUE )

pred <- predict(DNN, newdata = valid.df)
confusionMatrix(pred, valid.df$diagnosis)

Answer 1

The neuralnet method used in caret, can only be used for regression modelling, not classification. You need to select a different model. You can check which models can be used for classification with caret here .

For neuralnets and classification you can use mxnet and nnet methods.