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 NA and
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 integer64 support
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)
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) 120µs 123.1µs 7900. 2.41KB 0
#> 2 sum(is.na(x)) 829µs 1.95ms 521. 3.81MB 41.6
# 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) 56.9µs 72.6µs 12840. 0B 0
#> 2 sum(is.na(x)) 893.7µs 2ms 494. 3.81MB 39.9
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) 1.99ms 2.04ms 489. 9.14KB 0
#> 2 rowSums(is.na(m)) 2.83ms 3.86ms 262. 3.82MB 23.1
# 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) 1.96ms 2ms 499. 9.14KB 0
#> 2 colSums(is.na(m)) 1.88ms 2.92ms 342. 3.82MB 33.3is_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) 812µs 2.02ms 503. 3.81MB 97.8
#> 2 is_na(x) 625µs 1.84ms 559. 3.82MB 52.6
### 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) 1.27s 1.27s 0.790 61MB 0.790
#> 2 is_na(hours) 4.05ms 6.37ms 149. 13.9MB 17.9It differs in 2 regards:
NA when either that
element is an NA value or it is a list containing only
NA values.is_na returns a logical vector where
TRUE defines an empty row of only NA
values.# 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.
overviewInspired by the excellent skimr package, overview() is a
cheaper alternative designed for larger data.
df <- data.frame(
x = sample.int(100, 10^7, TRUE),
y = factor_(sample(LETTERS, 10^7, TRUE)),
z = rnorm(10^7)
)
overview(df, hist = TRUE)
#> obs: 10000000
#> cols: 3
#>
#> ----- Numeric -----
#> col class n_missing p_complete n_unique mean p0 p25 p50 p75 p100 iqr
#> 1 x integer 0 1 100 50.5 1 26 51 75 100 49
#> 2 z numeric 0 1 10000000 0 -5.5 -0.68 0 0.67 5.16 1.35
#> sd hist
#> 1 28.86 ▇▇▇▇▇
#> 2 1 ▁▁▇▂▁
#>
#> ----- Categorical -----
#> col class n_missing p_complete n_unique n_levels min max
#> 1 y factor 0 1 26 26 A Z
mark(overview(df))
#> # 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) 1.04s 1.04s 0.966 2.09KB 0ssetsset(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) 87µs 129µs 8010. 88.2KB 6.48
#> 2 sset(x, x %in% y) 164µs 256µs 3882. 285.4KB 8.79
#> 3 x[x %in% y] 136µs 226µs 4516. 324.5KB 11.2sset 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) 190µs 615µs 1654. 1.53MB 23.9
#> 2 df[0:10^5, , drop = FALSE] 6.69ms 7.88ms 129. 4.83MB 6.55It 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) 54.1ms 72ms 12.7 152MB 9.06
#> 2 df[-10^4:0, , drop = FALSE] 840.4ms 840ms 1.19 776MB 3.57The biggest difference between sset and [
is the way logical vectors are handled. The two main differences when
i is a logical vector are:
NA values are ignored, only the locations of
TRUE values are used.i must be the same length as x and 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 -5lag_()set.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) 1.4µs 1.5µs 618682. 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) 54.5ms 55.9ms 17.8 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 2.77ms 4.24ms 230. 3.81MB 2.07
#> 2 base_which 865.8µs 2.98ms 346. 7.63MB 9.17
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 823µs 833µs 1192. 0B 0
#> 2 base_which 487µs 499µs 1896. 3.81MB 24.6
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 2.24ms 2.92ms 337. 1.91MB 2.07
#> 2 base_which 923.1µs 2.08ms 485. 7.63MB 11.1
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 3.83ms 5.2ms 192. 3.81MB 2.06
#> 2 base_which 1.08ms 3.69ms 275. 9.54MB 6.66
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 2.71ms 3.18ms 313. 1.89MB 2.07
#> 2 base_which 3.42ms 4.41ms 227. 5.7MB 4.24x <- 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 9.83ms 10.2ms 97.5 4.59MB 0
#> 2 base_factor 598.01ms 598ms 1.67 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 4.51ms 5.17ms 187. 1.53MB 0
#> 2 base_factor 975.8ms 975.8ms 1.02 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 231.34ms 247.59ms 4.09 5.23MB 0
#> 2 base_factor 3.87s 3.87s 0.258 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 5.46ms 6.72ms 144. 3.49MB 0
#> 2 base_factor 58.82ms 65.16ms 15.4 39.89MB 2.56x <- 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 2.76ms 3.02ms 326. 1.18MB 0
#> 2 base_intersect 5.12ms 5.42ms 171. 5.16MB 2.16
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 3.03ms 3.29ms 291. 1.76MB 2.14
#> 2 base_setdiff 4.56ms 5.49ms 181. 5.71MB 2.18%in_% and %!in_%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 1.77ms 1.89ms 522. 781.34KB 0
#> 2 base 2.96ms 3.12ms 305. 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 1.72ms 1.82ms 540. 787.84KB 0
#> 2 base 2.81ms 3.34ms 296. 2.91MB 2.16as_discreteas_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 148ms 149ms 6.42 38.2MB 1.60
#> 2 base_cut 601ms 601ms 1.66 267.1MB 1.66