In cheapr, ‘cheap’ means fast and memory-efficient, and that’s exactly the philosophy that cheapr aims to follow.
You can install cheapr like so:
install.packages("cheapr")or you can install the development version of cheapr:
remotes::install_github("NicChr/cheapr")Some common operations that cheapr can do much faster and more efficiently include:
-
Counting, finding, removing and replacing
NAand scalar values -
Creating factors
-
Creating multiple sequences in a vectorised way
-
Sub-setting vectors and data frames efficiently
-
Safe, flexible and fast greatest common divisor and lowest common multiple
-
Lags/leads
-
Lightweight
integer64support -
In-memory Math (no copies, vectors updated by reference)
-
Summary statistics of data frame variables
-
Binning of continuous data
Let’s first load the required packages
library(cheapr)
#>
#> Attaching package: 'cheapr'
#> The following objects are masked from 'package:base':
#>
#> round, trunc
library(bench)num_na() is a useful function to efficiently return the number of NA
values and can be used in a variety of problems.
Almost all the NA handling functions in cheapr can make use of
multiple cores on your machine through openMP.
x <- rep(NA, 10^6)
# 1 core by default
mark(num_na(x), sum(is.na(x)))
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 num_na(x) 983µs 1.01ms 973. 2.41KB 0
#> 2 sum(is.na(x)) 780µs 1.84ms 543. 3.81MB 44.8
# 4 cores
options(cheapr.cores = 4)
mark(num_na(x), sum(is.na(x)))
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 num_na(x) 258µs 346.5µs 2666. 0B 0
#> 2 sum(is.na(x)) 812µs 1.93ms 493. 3.81MB 40.4
options(cheapr.cores = 1)m <- matrix(x, ncol = 10^3)
# Number of NA values by row
mark(row_na_counts(m),
rowSums(is.na(m)))
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 row_na_counts(m) 1.89ms 2.21ms 455. 9.14KB 0
#> 2 rowSums(is.na(m)) 2.63ms 3.71ms 267. 3.82MB 25.8
# Number of NA values by col
mark(col_na_counts(m),
colSums(is.na(m)))
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 col_na_counts(m) 2.49ms 2.55ms 384. 9.14KB 0
#> 2 colSums(is.na(m)) 1.77ms 2.85ms 353. 3.82MB 31.0is_na is a multi-threaded alternative to is.na
x <- rnorm(10^6)
x[sample.int(10^6, 10^5)] <- NA
mark(is.na(x), is_na(x))
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 is.na(x) 818µs 2.01ms 481. 3.81MB 61.6
#> 2 is_na(x) 1.33ms 2.6ms 391. 3.82MB 36.2
### posixlt method is much faster
hours <- as.POSIXlt(seq.int(0, length.out = 10^6, by = 3600),
tz = "UTC")
hours[sample.int(10^6, 10^5)] <- NA
mark(is.na(hours), is_na(hours))
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 is.na(hours) 1.19s 1.19s 0.843 61MB 0.843
#> 2 is_na(hours) 13.26ms 15.35ms 64.3 13.9MB 7.80It differs in 2 regards:
- List elements are regarded as
NAwhen either that element is anNAvalue or it is a list containing onlyNAvalues. - For data frames,
is_nareturns a logical vector whereTRUEdefines an empty row of onlyNAvalues.
# List example
is.na(list(NA, list(NA, NA), 10))
#> [1] TRUE FALSE FALSE
is_na(list(NA, list(NA, NA), 10))
#> [1] TRUE TRUE FALSE
# Data frame example
df <- data.frame(x = c(1, NA, 3),
y = c(NA, NA, NA))
df
#> x y
#> 1 1 NA
#> 2 NA NA
#> 3 3 NA
is_na(df)
#> [1] FALSE TRUE FALSE
is_na(df)
#> [1] FALSE TRUE FALSE
# The below identity should hold
identical(is_na(df), row_na_counts(df) == ncol(df))
#> [1] TRUEis_na and all the NA handling functions fall back on calling
is.na() if no suitable method is found. This means that custom objects
like vctrs rcrds and more are supported.
Inspired by the excellent skimr package, overview() is a cheaper
alternative designed for larger data.
set.seed(42)
df <- data.frame(
x = sample.int(100, 10^7, TRUE),
y = factor_(sample(LETTERS, 10^7, TRUE)),
z = rnorm(10^7)
)
overview(df)
#> obs: 10000000
#> cols: 3
#>
#> ----- Numeric -----
#> col class n_mssn p_cmpl n_uniq mean p0 p25 p50 p75 p100 iqr sd
#> 1 x intgr 0 1 100 50.51 1 26 51 76 100 50 28.86
#> 2 z numrc 0 1 10000000 0 -5.16 -0.68 0 0.67 5.26 1.35 1
#> hist
#> 1 ▇▇▇▇▇
#> 2 ▁▂▇▂▁
#>
#> ----- Categorical -----
#> col class n_mssn p_cmpl n_uniq n_lvls min max
#> 1 y factr 0 1 26 26 A Z
mark(overview(df, hist = FALSE))
#> # A tibble: 1 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 overview(df, hist = FALSE) 1.33s 1.33s 0.754 2.09KB 0sset(iris, 1:5)
#> Sepal.Length Sepal.Width Petal.Length Petal.Width Species
#> 1 5.1 3.5 1.4 0.2 setosa
#> 2 4.9 3.0 1.4 0.2 setosa
#> 3 4.7 3.2 1.3 0.2 setosa
#> 4 4.6 3.1 1.5 0.2 setosa
#> 5 5.0 3.6 1.4 0.2 setosa
sset(iris, 1:5, j = "Species")
#> Species
#> 1 setosa
#> 2 setosa
#> 3 setosa
#> 4 setosa
#> 5 setosa
# sset always returns a data frame when input is a data frame
sset(iris, 1, 1) # data frame
#> Sepal.Length
#> 1 5.1
iris[1, 1] # not a data frame
#> [1] 5.1
x <- sample.int(10^6, 10^4, TRUE)
y <- sample.int(10^6, 10^4, TRUE)
mark(sset(x, x %in_% y), sset(x, x %in% y), x[x %in% y])
#> # A tibble: 3 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 sset(x, x %in_% y) 81.9µs 119µs 8606. 83.2KB 6.54
#> 2 sset(x, x %in% y) 156.4µs 238µs 4096. 285.4KB 8.90
#> 3 x[x %in% y] 131.2µs 210µs 4802. 324.5KB 13.8sset uses an internal range-based subset when i is an ALTREP integer
sequence of the form m:n.
mark(sset(df, 0:10^5), df[0:10^5, , drop = FALSE])
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 sset(df, 0:10^5) 137.5µs 527.1µs 1899. 1.53MB 27.5
#> 2 df[0:10^5, , drop = FALSE] 6.01ms 7.14ms 141. 4.83MB 7.43It also accepts negative indexes
mark(sset(df, -10^4:0),
df[-10^4:0, , drop = FALSE],
check = FALSE) # The only difference is the row names
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 sset(df, -10^4:0) 50.8ms 65.1ms 13.1 152MB 9.34
#> 2 df[-10^4:0, , drop = FALSE] 796.2ms 796.2ms 1.26 776MB 5.02The biggest difference between sset and [ is the way logical vectors
are handled. The two main differences when i is a logical vector are:
NAvalues are ignored, only the locations ofTRUEvalues are used.imust be the same length asxand is not recycled.
# Examples with NAs
x <- c(1, 5, NA, NA, -5)
x[x > 0]
#> [1] 1 5 NA NA
sset(x, x > 0)
#> [1] 1 5
# Example with length(i) < length(x)
sset(x, TRUE)
#> Error in check_length(i, length(x)): i must have length 5
# This is equivalent
x[TRUE]
#> [1] 1 5 NA NA -5
# to..
sset(x)
#> [1] 1 5 NA NA -5set.seed(37)
lag_(1:10, 3) # Lag(3)
#> [1] NA NA NA 1 2 3 4 5 6 7
lag_(1:10, -3) # Lead(3)
#> [1] 4 5 6 7 8 9 10 NA NA NA
# Using an example from data.table
library(data.table)
#> Warning: package 'data.table' was built under R version 4.4.1
dt <- data.table(year=2010:2014, v1=runif(5), v2=1:5, v3=letters[1:5])
# Similar to data.table::shift()
lag_(dt, 1) # Lag
#> year v1 v2 v3
#> <int> <num> <int> <char>
#> 1: NA NA NA <NA>
#> 2: 2010 0.54964085 1 a
#> 3: 2011 0.07883715 2 b
#> 4: 2012 0.64879698 3 c
#> 5: 2013 0.49685336 4 d
lag_(dt, -1) # Lead
#> year v1 v2 v3
#> <int> <num> <int> <char>
#> 1: 2011 0.07883715 2 b
#> 2: 2012 0.64879698 3 c
#> 3: 2013 0.49685336 4 d
#> 4: 2014 0.71878731 5 e
#> 5: NA NA NA <NA>With lag_ we can update variables by reference, including entire data
frames
# At the moment, shift() cannot do this
lag_(dt, set = TRUE)
#> year v1 v2 v3
#> <int> <num> <int> <char>
#> 1: NA NA NA <NA>
#> 2: 2010 0.54964085 1 a
#> 3: 2011 0.07883715 2 b
#> 4: 2012 0.64879698 3 c
#> 5: 2013 0.49685336 4 d
dt # Was updated by reference
#> year v1 v2 v3
#> <int> <num> <int> <char>
#> 1: NA NA NA <NA>
#> 2: 2010 0.54964085 1 a
#> 3: 2011 0.07883715 2 b
#> 4: 2012 0.64879698 3 c
#> 5: 2013 0.49685336 4 dlag2_ is a more generalised variant that supports vectors of lags,
custom ordering and run lengths.
lag2_(dt, order = 5:1) # Reverse order lag (same as lead)
#> year v1 v2 v3
#> <int> <num> <int> <char>
#> 1: 2010 0.54964085 1 a
#> 2: 2011 0.07883715 2 b
#> 3: 2012 0.64879698 3 c
#> 4: 2013 0.49685336 4 d
#> 5: NA NA NA <NA>
lag2_(dt, -1) # Same as above
#> year v1 v2 v3
#> <int> <num> <int> <char>
#> 1: 2010 0.54964085 1 a
#> 2: 2011 0.07883715 2 b
#> 3: 2012 0.64879698 3 c
#> 4: 2013 0.49685336 4 d
#> 5: NA NA NA <NA>
lag2_(dt, c(1, -1)) # Alternating lead/lag
#> year v1 v2 v3
#> <int> <num> <int> <char>
#> 1: NA NA NA <NA>
#> 2: 2011 0.07883715 2 b
#> 3: 2010 0.54964085 1 a
#> 4: 2013 0.49685336 4 d
#> 5: 2012 0.64879698 3 c
lag2_(dt, c(-1, 0, 0, 0, 0)) # Lead e.g. only first row
#> year v1 v2 v3
#> <int> <num> <int> <char>
#> 1: 2010 0.54964085 1 a
#> 2: 2010 0.54964085 1 a
#> 3: 2011 0.07883715 2 b
#> 4: 2012 0.64879698 3 c
#> 5: 2013 0.49685336 4 dgcd2(5, 25)
#> [1] 5
scm2(5, 6)
#> [1] 30
gcd(seq(5, 25, by = 5))
#> [1] 5
scm(seq(5, 25, by = 5))
#> [1] 300
x <- seq(1L, 1000000L, 1L)
mark(gcd(x))
#> # A tibble: 1 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 gcd(x) 1.3µs 1.4µs 658788. 0B 0
x <- seq(0, 10^6, 0.5)
mark(gcd(x))
#> # A tibble: 1 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 gcd(x) 51.6ms 51.9ms 19.2 0B 0As an example, to create 3 sequences with different increments,
the usual approach might be to use lapply to loop through the increment
values together with seq()
# Base R
increments <- c(1, 0.5, 0.1)
start <- 1
end <- 5
unlist(lapply(increments, \(x) seq(start, end, x)))
#> [1] 1.0 2.0 3.0 4.0 5.0 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 1.0 1.1 1.2 1.3 1.4
#> [20] 1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3.0 3.1 3.2 3.3
#> [39] 3.4 3.5 3.6 3.7 3.8 3.9 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5.0In cheapr you can use seq_() which accepts vector arguments.
seq_(start, end, increments)
#> [1] 1.0 2.0 3.0 4.0 5.0 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 1.0 1.1 1.2 1.3 1.4
#> [20] 1.5 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3.0 3.1 3.2 3.3
#> [39] 3.4 3.5 3.6 3.7 3.8 3.9 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5.0Use add_id = TRUE to label the individual sequences.
seq_(start, end, increments, add_id = TRUE)
#> 1 1 1 1 1 2 2 2 2 2 2 2 2 2 3 3 3 3 3 3
#> 1.0 2.0 3.0 4.0 5.0 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5 5.0 1.0 1.1 1.2 1.3 1.4 1.5
#> 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
#> 1.6 1.7 1.8 1.9 2.0 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8 2.9 3.0 3.1 3.2 3.3 3.4 3.5
#> 3 3 3 3 3 3 3 3 3 3 3 3 3 3 3
#> 3.6 3.7 3.8 3.9 4.0 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8 4.9 5.0If you know the sizes of your sequences beforehand, use sequence_()
seq_sizes <- c(3, 5, 10)
sequence_(seq_sizes, from = 0, by = 1/3, add_id = TRUE) |>
enframe_()
#> # A tibble: 18 × 2
#> name value
#> <chr> <dbl>
#> 1 1 0
#> 2 1 0.333
#> 3 1 0.667
#> 4 2 0
#> 5 2 0.333
#> 6 2 0.667
#> 7 2 1
#> 8 2 1.33
#> 9 3 0
#> 10 3 0.333
#> 11 3 0.667
#> 12 3 1
#> 13 3 1.33
#> 14 3 1.67
#> 15 3 2
#> 16 3 2.33
#> 17 3 2.67
#> 18 3 3You can also calculate the sequence sizes using seq_size()
seq_size(start, end, increments)
#> [1] 5 9 41x <- rep(TRUE, 10^6)
mark(cheapr_which = which_(x),
base_which = which(x))
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 cheapr_which 2.52ms 3.92ms 252. 3.81MB 4.31
#> 2 base_which 654.2µs 2.68ms 379. 7.63MB 12.0
x <- rep(FALSE, 10^6)
mark(cheapr_which = which_(x),
base_which = which(x))
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 cheapr_which 741µs 753µs 1311. 0B 0
#> 2 base_which 453µs 464µs 2081. 3.81MB 28.7
x <- c(rep(TRUE, 5e05), rep(FALSE, 1e06))
mark(cheapr_which = which_(x),
base_which = which(x))
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 cheapr_which 2.02ms 2.71ms 364. 1.91MB 2.05
#> 2 base_which 776.9µs 1.77ms 558. 7.63MB 17.3
x <- c(rep(FALSE, 5e05), rep(TRUE, 1e06))
mark(cheapr_which = which_(x),
base_which = which(x))
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 cheapr_which 3.63ms 4.93ms 203. 3.81MB 2.07
#> 2 base_which 913.1µs 2.96ms 332. 9.54MB 11.8
x <- sample(c(TRUE, FALSE), 10^6, TRUE)
x[sample.int(10^6, 10^4)] <- NA
mark(cheapr_which = which_(x),
base_which = which(x))
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 cheapr_which 2.42ms 3.01ms 331. 1.89MB 2.08
#> 2 base_which 3.16ms 4.09ms 245. 5.7MB 6.68x <- sample(seq(-10^3, 10^3, 0.01))
y <- do.call(paste0, expand.grid(letters, letters, letters, letters))
mark(cheapr_factor = factor_(x),
base_factor = factor(x))
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 cheapr_factor 9.19ms 9.91ms 97.6 4.59MB 0
#> 2 base_factor 529.85ms 529.85ms 1.89 27.84MB 0
mark(cheapr_factor = factor_(x, order = FALSE),
base_factor = factor(x, levels = unique(x)))
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 cheapr_factor 4.34ms 4.93ms 202. 1.53MB 2.02
#> 2 base_factor 839.96ms 839.96ms 1.19 22.79MB 0
mark(cheapr_factor = factor_(y),
base_factor = factor(y))
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 cheapr_factor 221.21ms 224.83ms 4.44 5.23MB 0
#> 2 base_factor 3.09s 3.09s 0.324 54.35MB 0
mark(cheapr_factor = factor_(y, order = FALSE),
base_factor = factor(y, levels = unique(y)))
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 cheapr_factor 4.88ms 6.5ms 152. 3.49MB 2.21
#> 2 base_factor 54.53ms 60.2ms 16.7 39.89MB 0x <- sample.int(10^6, 10^5, TRUE)
y <- sample.int(10^6, 10^5, TRUE)
mark(cheapr_intersect = intersect_(x, y, dups = FALSE),
base_intersect = intersect(x, y))
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 cheapr_intersect 2.71ms 3.37ms 292. 1.18MB 2.17
#> 2 base_intersect 4.17ms 5.01ms 196. 5.16MB 2.18
mark(cheapr_setdiff = setdiff_(x, y, dups = FALSE),
base_setdiff = setdiff(x, y))
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 cheapr_setdiff 2.99ms 3.12ms 312. 1.76MB 0
#> 2 base_setdiff 4.45ms 5.43ms 185. 5.71MB 2.17mark(cheapr = x %in_% y,
base = x %in% y)
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 cheapr 1.75ms 1.81ms 535. 781.34KB 2.17
#> 2 base 2.5ms 2.96ms 336. 2.53MB 2.18
mark(cheapr = x %!in_% y,
base = !x %in% y)
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 cheapr 1.75ms 1.83ms 533. 787.84KB 0
#> 2 base 2.67ms 3.14ms 313. 2.91MB 2.17as_discrete is a cheaper alternative to cut
x <- rnorm(10^7)
b <- seq(0, max(x), 0.2)
mark(cheapr_cut = as_discrete(x, b, left = FALSE),
base_cut = cut(x, b))
#> Warning: Some expressions had a GC in every iteration; so filtering is
#> disabled.
#> # A tibble: 2 × 6
#> expression min median `itr/sec` mem_alloc `gc/sec`
#> <bch:expr> <bch:tm> <bch:tm> <dbl> <bch:byt> <dbl>
#> 1 cheapr_cut 210ms 211ms 4.73 38.2MB 0
#> 2 base_cut 487ms 509ms 1.96 267.1MB 2.95