Futures

Parallel Programming in R

Nabeel Imam

Data Scientist

The demanding boss

A suited man talks into a loudspeaker.

Immediate execution

report <- "<important report text>"

task <- {
  print("Hi, here is the report.") # Email text
  report
}

"Hi, please find attached the report."

task

"<important report text>"

Delayed execution

library(future)


task_future <- future({
  print("Hi, here is the report.")
  report
})

Delayed execution

print(task_future)

Expression: {
print("Hi, here is the report.")
report
}

Environment: R_GlobalEnv

Resolved: TRUE

Value: 136 bytes of class "character"
...

The code or "how-to" of the task

The environment or the context of the task
Is it done? Seems so!
Description of the outcome

Delayed execution

value(task_future)

[1] "Hi, here is the report."
[1] "<important report text>"

A team of workers

A team of workers receives tasks from their manager.

¹ Designed by Freepik

A team of workers

task1 <- future({
  print("Hi, here is the report.")
  report
})

task2 <- future({
  run_analysis()
  print("Analysis done")
})

task3 <- future({
  book_meeting_rooms()
  print("Meeting rooms booked!")
})

value(task1) # 10:00 AM Office Administrator

[1] "Hi, here is the report."
[1] "<important report text>"

value(task3) # 11:00 AM Office Administrator

"Meeting rooms booked!"

value(task2) # 03:00 PM Analyst

"Analysis done"

Moving average with futures

print(amazon_prices)

 [1] 312.58 307.01 302.24
 [4] 297.50 300.32 301.48
 [7] 297.56 297.48 291.93
[10] 294.00 286.28 292.59
...

ma <- future({moving_average(amazon_prices)})

moving_average <- function (prices) {

  N <- length(prices) - 2
  moving_avgs <- rep(NA, N)

  for (i in 1:N) {
    moving_avgs[i] <- mean(prices[i: (i + 2)])
  }

  return(moving_avgs)
}

Moving average with futures

ggplot() +
  geom_line(aes(x = 1:(length(amazon_prices) - 2),
                y = value(ma))) + # Value of futures is required here
  labs(y = "Moving price average (USD)", x = "Days")

A line plot of the moving average price of Amazon stock during one month.

Parallel futures

print(amazon_list)

$`2015-01`
 [1] 312.58 307.01 302.24
 [4] 297.50 300.32 301.48
 ...
$`2015-02`
 [1] 350.05 360.29 358.38
 [4] 366.00 374.87 371.00
 ...
$`2015-03`
 [1] 380.85 383.95 385.71
 [4] 385.61 385.52 378.40
 ...
...

moving_average <- function (prices) {

  N <- length(prices) - 2
  moving_avgs <- rep(NA, N)

  for (i in 1:N) {
    moving_avgs[i] <- mean(prices[i: (i + 2)])
  }

  return(moving_avgs)
}

Parallel futures

plan(multisession, workers = 4)


ma <- lapply(amazon_list,
            function (x) {
              future(moving_average(x))
              })

Plan to employ four worker cores in a multisession
Create one future task for every element of amazon_list
- Futures created under multisession plan will be executed in parallel

Parallel futures

value(ma)


plan(sequential)

$`2015-01`
 [1] 307.2767 302.2500 300.0200 299.7667 299.7867 298.8400 295.6567 ...
$`2015-02`
 [1] 356.2400 361.5567 366.4167 370.6233 372.3533 371.1400 372.5067 ...
$`2015-03`
 [1] 383.5033 385.0900 385.6133 383.1767 380.4567 375.4867 372.2933 ...
...

Why future?

Sequential

plan(sequential)

ma <- lapply(amazon_list,
             function (x) {
               future(moving_average(x))
               })

value(ls)

Parallel

plan(multisession, workers = 4)

ls <- lapply(amazon_list,
             function (x) {
               future(moving_average(x))
               })

value(ls)

plan(sequential)

Let's practice!

Parallel Programming in R