The main purpose of this benchmark is to show that polars and tidypolars are
close and to give an idea of the performance. For more thorough, representative
benchmarks about polars, take a look at DuckDB benchmarks instead.

library(collapse, warn.conflicts = FALSE)
library(dplyr, warn.conflicts = FALSE)
library(dtplyr)
library(polars)
library(tidypolars)

large_iris <- data.table::rbindlist(rep(list(iris), 100000))
large_iris_pl <- as_polars_lf(large_iris)
large_iris_dt <- lazy_dt(large_iris)

format(nrow(large_iris), big.mark = ",")

bench::mark(
  polars = {
    large_iris_pl$
      select(c("Sepal.Length", "Sepal.Width", "Petal.Length", "Petal.Width"))$
      with_columns(
        pl$when(
          (pl$col("Petal.Length") / pl$col("Petal.Width") > 3)
        )$then(pl$lit("long"))$
          otherwise(pl$lit("large"))$
          alias("petal_type")
      )$
      filter(pl$col("Sepal.Length")$is_between(4.5, 5.5))$
      collect()
  },
  tidypolars = {
    large_iris_pl |>
      select(starts_with(c("Sep", "Pet"))) |>
      mutate(
        petal_type = ifelse((Petal.Length / Petal.Width) > 3, "long", "large")
      ) |> 
      filter(between(Sepal.Length, 4.5, 5.5)) |> 
      compute()
  },
  dplyr = {
    large_iris |>
      select(starts_with(c("Sep", "Pet"))) |>
      mutate(
        petal_type = ifelse((Petal.Length / Petal.Width) > 3, "long", "large")
      ) |>
      filter(between(Sepal.Length, 4.5, 5.5))
  },
  dtplyr = {
    large_iris_dt |>
      select(starts_with(c("Sep", "Pet"))) |>
      mutate(
        petal_type = ifelse((Petal.Length / Petal.Width) > 3, "long", "large")
      ) |>
      filter(between(Sepal.Length, 4.5, 5.5)) |> 
      as.data.frame()
  },
  collapse = {
    large_iris |>
      fselect(c("Sepal.Length", "Sepal.Width", "Petal.Length", "Petal.Width")) |>
      fmutate(
        petal_type = data.table::fifelse((Petal.Length / Petal.Width) > 3, "long", "large")
      ) |>
      fsubset(Sepal.Length >= 4.5 & Sepal.Length <= 5.5)
  },
  check = FALSE,
  iterations = 40
)

# NOTE: do NOT take the "mem_alloc" results into account.
# `bench::mark()` doesn't report the accurate memory usage for packages calling
# Rust code.