Statistical models on the GPU: a performance comparison.

modelCreationExecutionTimingsGPUvsNonGPU.R

library(gputools)
library(e1071)
library(fueleconomy)
data(vehicles)
 
vehicles$id <- NULL
vehicles$hwy <- NULL

vehicles <- as.data.frame(unclass(vehicles))
vehicles <- vehicles[complete.cases(vehicles), ]

system.time(model.lm <- lm(cty ~ ., data = vehicles))
#   user  system elapsed 
#317.727   0.093 318.022 

system.time(model.gpuLm <- gpuLm(cty ~ ., data = vehicles))
#   user  system elapsed 
#  3.231   0.676   3.909 

system.time(model.glm <- glm(cty ~ ., data = vehicles))
#    user  system elapsed 
# 573.386   0.289 574.071 

system.time(model.gpuGlm <- gpuGlm(cty ~ ., data = vehicles))
#    user  system elapsed 
#  18.063   2.653  20.724 

system.time(model.svm <- svm(cty ~ ., data = vehicles))
#    user  system elapsed 
# 150.502   0.791 151.343 

vehicles$cty <- NULL

system.time(predict(model.lm, vehicles))
#    user  system elapsed 
#   4.214   0.000   4.215 

system.time(predict(model.gpuLm, vehicles))
#    user  system elapsed 
#   0.460   0.000   0.459 

system.time(predict(model.glm, vehicles))
#    user  system elapsed 
#   4.224   0.000   4.223 

system.time(predict(model.gpuGlm, vehicles))
#    user  system elapsed 
#   4.311   0.001   4.313 

system.time(predict(model.svm, vehicles))
#    user  system elapsed 
#  39.584   0.005  39.599