In WinVector/replyr: Patches to Use 'dplyr' on Remote Data Sources

Standard deviation with SQLite is zero when there is one data item, not the expected NA. Nocie the sd() calculation agrees with R's local calculation when n>1 so this isn't just a sample variance versus population variance issue.

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = " # "
)
options(width =100)

library('dplyr')
library('RSQLite')
packageVersion('dplyr')
packageVersion('RSQLite')

my_db <- dplyr::src_sqlite(":memory:", create = TRUE)

# confirm sqlite can represent NA
d <- data.frame(x = c(1,NA,3))
dbData <- dplyr::copy_to(my_db, d, name='d', 
                           create=TRUE, overwrite=TRUE)
print(dbData)

for(n in 1:3) {
  print("***********")
  print(paste('n',n))
  dplyr::db_drop_table(my_db$con, 'd')
  d <- data.frame(x= seq_len(n))
  print("local")
  print(dplyr::summarise_all(d, dplyr::funs(sd)))
  dbData <- dplyr::copy_to(my_db, d, name='d', 
                           create=TRUE, overwrite=TRUE)
  print("RSQLite")
  print(dplyr::summarise_all(dbData, dplyr::funs(sd)))
  print("***********")
}