Comparing speed of Random Forest like algorithms

gistfile1.txt

# sandbox

library(caret)
library(tictoc)


g1 <- data.frame(mtry = c(2, 5))

tic()
m <- train(price ~ ., 
           data = tail(ggplot2::diamonds, 1e4), 
           method = 'rf', 
           trControl = trainControl(number = 3),
           tuneGrid = g1)
toc()
plot(m)

g2 <- data.frame(mtry = c(2, 5),
                 splitrule = 'variance',
                 min.node.size = 5)
tic()
m2 <- train(price ~ ., 
            data = tail(ggplot2::diamonds, 1e4), 
            method = 'ranger', 
            trControl = trainControl(number = 3),
            tuneGrid = g2)
toc()
plot(m2)

g3 <- data.frame(nrounds = 1,
                 max_depth = c(2, 5),
                 subsample = 0.8,
                 eta = 1,
                 gamma = 0,
                 colsample_bytree = 0.5,
                 min_child_weight = 1)


tic()                 
m3 <- train(price ~ ., 
            data = tail(ggplot2::diamonds, 1e4), 
            method = 'xgbTree', 
            trControl = trainControl(number = 3),
            tuneGrid = g3,
            num_parallel_tree = 500)
toc()

plot(m3)





# Larger data


# sandbox

library(caret)
library(tictoc)


g1 <- data.frame(mtry = c(2, 5))

tic()
m <- train(price ~ ., 
           data = tail(ggplot2::diamonds, 1e5), 
           method = 'rf', 
           trControl = trainControl(number = 3),
           tuneGrid = g1)
toc()
plot(m)

g2 <- data.frame(mtry = c(2, 5),
                 splitrule = 'variance',
                 min.node.size = 5)
tic()
m2 <- train(price ~ ., 
            data = tail(ggplot2::diamonds, 1e5), 
            method = 'ranger', 
            trControl = trainControl(number = 3),
            tuneGrid = g2)
toc()
plot(m2)

g3 <- data.frame(nrounds = 1,
                 max_depth = c(2, 5),
                 subsample = 0.8,
                 eta = 1,
                 gamma = 0,
                 colsample_bytree = 0.5,
                 min_child_weight = 1)


tic()                 
m3 <- train(price ~ ., 
            data = tail(ggplot2::diamonds, 1e5), 
            method = 'xgbTree', 
            trControl = trainControl(number = 3),
            tuneGrid = g3,
            num_parallel_tree = 500)
toc()

plot(m3)