Using the classification error rate for each number of trees, find the optimal number of trees to use for random forest classifier
find.best.number.of.trees.RdFind a plateau that corresponds with the minimum error. Uses a sliding window approach where the window has a width of 3 trees.
find.best.number.of.trees(error.oob)
Arguments
| error.oob | A vector of numbers. Should be the $err.rate from a randomForest::randomForest object. |
|---|
Value
A numerical value specifying the optimal number of trees to use in random forest.
Details
Select windows with lowest mean. From these windows, I select the windows with lowest standard deviation (indicates plateau). If multiple plateaus exist, select the one with the fewest number of trees. Then select the tree corresponding to the center of the window as the optimal number of trees.
See also
Other Classification functions:
CVPredictionsRandomForest(),
CVRandomForestClassificationMatrixForPheatmap(),
GenerateExampleDataMachinelearnr(),
LOOCVPredictionsRandomForestAutomaticMtryAndNtree(),
LOOCVRandomForestClassificationMatrixForPheatmap(),
RandomForestAutomaticMtryAndNtree(),
RandomForestClassificationGiniMatrixForPheatmap(),
RandomForestClassificationPercentileMatrixForPheatmap(),
eval.classification.results()
Examples
id = c("1a", "1b", "1c", "1d", "1e", "1f", "1g", "2a", "2b", "2c", "2d", "2e", "2f", "3a", "3b", "3c", "3d", "3e", "3f", "3g", "3h", "3i") x = c(18, 21, 22, 24, 26, 26, 27, 30, 31, 35, 39, 35, 30, 40, 41, 42, 44, 46, 47, 48, 49, 54) y = c(10, 11, 22, 15, 12, 13, 14, 33, 39, 37, 44, 40, 45, 27, 29, 20, 28, 21, 30, 31, 23, 24) a = c(1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) b = c(1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1) actual = as.factor(c("1", "1", "1", "1", "1", "1", "1", "2", "2", "2", "2", "2", "2", "3", "3", "3", "3", "3", "3", "3", "3", "3")) example.data <- data.frame(id, x, y, a, b, actual) set.seed(1) rf.result <- randomForest::randomForest(x=example.data[,c("x", "y", "a", "b")], y=example.data[,"actual"], proximity=TRUE, ntree=50) error.oob <- rf.result[[4]][,1] best.tree <- find.best.number.of.trees(error.oob) trees <- 1:length(error.oob) plot(trees, error.oob, type = "l")
