test of dtplyr speed

dtplyr_test.R

library(dplyr)
library(data.table)
library(dtplyr)

df <- data.frame(customer = sample(100000, size = 5000000, replace = TRUE),
                 value = runif(5000000))

# dplyr
t0 <- Sys.time()
res_dp <- df %>%
  group_by(customer) %>%
  summarise(value = sum(value))
Sys.time() - t0

# data.table
t0 <- Sys.time()
dt <- as.data.table(df)
res_dt <- dt[, .(value = sum(value)), keyby = .(customer)]
Sys.time() - t0

# dtplyr
t0 <- Sys.time()
res_dtp <- df %>%
  lazy_dt() %>%
  group_by(customer) %>%
  summarise(value = sum(value)) %>%
  as_tibble()
Sys.time() - t0

# dtplyr is /lazy/
# dtplyr shows how R can manipulate language
#
# why use dtplyr if you know DT?
#   - consistency of style
#   - verbosity is good for larger code bases
#   - immutability is good for production code
#   - cooperating with people who don't use DT
#
# what if you must use DT snippets but don't want to import it all?
# withr::with_package()