CODEWITHSUNDEEP
rrandom-forest
user30985
user30985

Reputation: 683

Run randomForest with many mtry values

I am a new user of the randomForest R package. I would like to run randomForest classifications for iterations of mtry =1,5,7.

For example, I would like to run mtry =1 100 times and mtry=2 100 times. The output should show the out of bag errors for each run (100 results for mtry =1 and 100 results for mtry =2).

I only can write the code for 1 run, I do not know how to run the iteration of code with different values for mtry.

rf  <-  randomForest(class_name  ~  .,  data=tr,ntree=1000, 
                     importance=TRUE, proximity=TRUE, mtry=2)

Upvotes: 1

Views: 1814

Answers (1)

eipi10
eipi10

Reputation: 93761

Here's a quick and dirty method using the built-in iris data frame. The code below gives the final OOB error rate for each value of mtry. 

mtry = 1:4
oob = data.frame()

# Loop over each value of mtry and store result in a data frame
for (i in mtry) {

  rf1 <- randomForest(Species ~ ., data=iris, ntree=100, mtry=i) 

  result = data.frame(mtry=i, 
                      OOB=rf1[["err.rate"]][nrow(rf1[["err.rate"]]),"OOB"])
  oob = rbind(oob, result)
} 

oob
     mtry        OOB
OOB     1 0.04666667
OOB1    2 0.04000000
OOB2    3 0.04000000
OOB3    4 0.04000000

To keep all of the ntree OOB errors from each value of mtry, just change this:

OOB=rf1[["err.rate"]][nrow(rf1[["err.rate"]]),"OOB"])

To this:

OOB=rf1[["err.rate"]][ ,"OOB"])

Upvotes: 4

Related Questions