├── .gitignore
├── README.md
└── parallel_demo.R


/.gitignore:
--------------------------------------------------------------------------------
 1 | # History files
 2 | .Rhistory
 3 | 
 4 | # Example code in package build process
 5 | *-Ex.R
 6 | 
 7 | # R data files from past sessions
 8 | .Rdata
 9 | 
10 | # RStudio files
11 | .Rproj.user/
12 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
1 | ## Example code for my parallel R talk at the Jan 20, 2015 Bay Area R Users Group Meetup
2 | 
3 | ## Other tools/links I mentioned
4 | - http://cran.r-project.org/web/views/HighPerformanceComputing.html
5 | - http://topepo.github.io/caret/index.html
6 | - http://www.dominodatalab.com/
7 | - http://www.dominodatalab.com/webinar
8 | 


--------------------------------------------------------------------------------
/parallel_demo.R:
--------------------------------------------------------------------------------
  1 | library(parallel)
  2 | numCores <- detectCores()
  3 | numCores
  4 | 
  5 | testFunction = function(i) {
  6 |   summary(rnorm(1000000))
  7 | }
  8 | 
  9 | inputs = 1:20
 10 | 
 11 | system.time({
 12 |   results = mclapply(inputs, testFunction, mc.cores = 8)
 13 | })
 14 | 
 15 | system.time({
 16 |   results = lapply(inputs, testFunction)
 17 | })
 18 | 
 19 | 
 20 | # NOTE FOR WINDOWS USERS
 21 | # cl <- makeCluster(numCores)  
 22 | # results = parLapply(cl, inputs, testFunction)  
 23 | # stopCluster(cl)  
 24 | 
 25 | 
 26 | # overhead
 27 | 
 28 | system.time({
 29 |   results = mclapply(1:10000, sqrt, mc.cores = 8)
 30 | })
 31 | 
 32 | system.time({
 33 |   results = lapply(1:10000, sqrt)
 34 | })
 35 | 
 36 | 
 37 | #############
 38 | ## FOREACH ##
 39 | #############
 40 | 
 41 | library(foreach)
 42 | 
 43 | # on windows
 44 | # library(doParallel)
 45 | # registerDoParallel(numCores)
 46 | 
 47 | # or for multiple machines
 48 | # library(snow)
 49 | # registerDoSNOW()
 50 | 
 51 | library(doMC)
 52 | registerDoMC(numCores)
 53 | 
 54 | 
 55 | x <- iris[which(iris[,5] != "setosa"), c(1,5)]
 56 | trials <- 10000
 57 | system.time({
 58 |     r <- foreach(icount(trials), .combine=cbind) %dopar% {
 59 |         ind <- sample(100, 100, replace=TRUE)
 60 |         result1 <- glm(x[ind,2]~x[ind,1], family=binomial(logit))
 61 |         coefficients(result1)
 62 |     }
 63 | })
 64 | 
 65 | system.time({
 66 |   r <- foreach(icount(trials), .combine=cbind) %do% {
 67 |     ind <- sample(100, 100, replace=TRUE)
 68 |     result1 <- glm(x[ind,2]~x[ind,1], family=binomial(logit))
 69 |     coefficients(result1)
 70 |   }
 71 | })
 72 | 
 73 | # gotcha: data copying
 74 | 
 75 | x = rep(0, times = 5)
 76 | for(i in 1:5) {
 77 |   x[i] = i * 2
 78 | }
 79 | x
 80 | 
 81 | x = rep(0, times = 5)
 82 | foreach(i = 1:5) %dopar% {
 83 |   x[i] = i * 2
 84 | }
 85 | x
 86 | 
 87 | 
 88 | # combine
 89 | 
 90 | mclapply(1:100, sqrt, mc.cores = 8)
 91 | 
 92 | foreach(x = 1:100) %dopar% {
 93 |   sqrt(x)
 94 | }
 95 | 
 96 | foreach(x = 1:100, .combine=c) %dopar% {
 97 |   sqrt(x)
 98 | }
 99 | 
100 | # preschedule -- when will this be faster?
101 | 
102 | system.time({
103 |   result <- foreach(x = 1:1000, .options.multicore=list(preschedule=FALSE)) %dopar% {
104 |     sqrt(x)
105 |   }
106 | })
107 | 
108 | system.time({
109 |   result <- foreach(x = 1:1000, .options.multicore=list(preschedule=TRUE)) %dopar% {
110 |     sqrt(x)
111 |   }
112 | })
113 | 
114 | # when might prescheduling be slower?
115 | 
116 | deciseconds = sample(1:10, 50, replace = TRUE)
117 | system.time({
118 |   result <- foreach(x = deciseconds, .options.multicore=list(preschedule=FALSE)) %dopar% {
119 |     Sys.sleep(x / 10)
120 |   }
121 | })
122 | 
123 | system.time({
124 |   result <- foreach(x = deciseconds, .options.multicore=list(preschedule=TRUE)) %dopar% {
125 |     Sys.sleep(x / 10)
126 |   }
127 | })
128 | 
129 | 
130 | 
131 | # back to slides
132 | 
133 | 
134 | # random forests
135 | wine <- read.csv( "winequality-red.csv", sep=';', header = TRUE ) 
136 | head(wine)
137 | y_dat = wine$quality
138 | x_dat <- wine[,1:11]
139 | 
140 | 
141 | library(randomForest)
142 | num_trees = 500
143 | system.time({
144 |   randomForest(y = y_dat, x = x_dat, ntree = num_trees)
145 | })
146 | 
147 | 
148 | trees_per_core = floor(num_trees / numCores)
149 | system.time({
150 |   wine_model <- foreach(trees=rep(trees_per_core, numCores), .combine=combine, .multicombine=TRUE) %dopar% {
151 |     randomForest(y = y_dat, x = x_dat, ntree = trees)
152 |   }
153 | })
154 | 
155 | 
156 | 
157 | #, .packages='randomForest'
158 | 
159 | 
160 | # caret (Classification And REgression Training)
161 | 
162 | library(caret)
163 | library(mlbench)
164 | data(Sonar)
165 | 
166 | inTrain <- createDataPartition(y = Sonar$Class, p = .75, list = FALSE)
167 | training <- Sonar[ inTrain,]
168 | testing <- Sonar[-inTrain,]
169 | 
170 | 
171 | ctrl <- trainControl(method = "repeatedcv", number = 8, repeats = 8)
172 | grid_rf <- expand.grid(.mtry = c(2, 3, 4))
173 | system.time({
174 |   rf <- train(Class ~ ., data = training,  method = "rf", trControl = ctrl, ntree=750,  tuneGrid = grid_rf)
175 | })
176 | 
177 | registerDoMC(1)
178 | system.time({
179 |   rf <- train(Class ~ ., data = training,  method = "rf", trControl = ctrl, ntree=750,  tuneGrid = grid_rf)
180 | })
181 | 
182 | 
183 | registerDoMC(numCores)


--------------------------------------------------------------------------------