difference of filtering operation (dt vs df)

library(data.table)
library(profmem)
library(nycflights13)

dt <- rbindlist(replicate(100L, flights, FALSE))
setkey(dt, month, hour, arr_delay)
object.size(dt) # 4041516584 bytes ~ 3.76 GB
df <- copy(dt)
dt2 <- copy(dt)
setDF(df)

# data.table
profmem({
  dt <- dt[month < 5]
  dt <- dt[hour > 6]
  dt <- dt[arr_delay > 0]
})
# total 3546226792 bytes ~ 3.3 GB

# data.frame
profmem({
  df <- df[df$month < 5, ]
  df <- df[df$hour > 6, ]
  df <- df[df$arr_delay > 0, ]
})
# total 10471614192 bytes ~ 9.75 GB

# best case
profmem({
  dt2 <- dt2[eval("month < 5 & hour > 6 & arr_delay > 0")]
})
# total 32944 ~ 32.17 KB