In brodieG/vetr: Trust, but Verify

knitr::opts_chunk$set(error=TRUE)
library(vetr)

Tweets

Tweet #2

Fast #rstats bounds checks with vetr::all_bw https://github.com/brodieG/vetr; 10x improvement over primitives (1/3):

~30% of speedup is avoiding 3x1e6 temp logical vecs, rest is dedicated loops for input permutations (num/int/char, nas, infs, etc.) (2/3).

Peeling back abstraction layer is eye-opening. Massive complexity to implement something seemingly so simple in optimized manner (3/3).

~1/3 of speedup is avoiding 3 logical(1e6) intermediate vectors, rest is stripping inner loops to bare minimum:

Tweet #1

Declarative structural guarantees for #rstats S3 objects via templates with new pkg #vetr https://github.com/brodieG/vetr (1/4)

Like vapply

Declarative checks via templates, much like vapply:

vet(numeric(1L), 1:3)
vet(numeric(1L), "hello")
vet(numeric(1L), 42)

Handles Complex Objects

vet(matrix(integer(), ncol=3), matrix(1:12, 4))
vet(matrix(integer(), ncol=3), matrix(1:12, 3))

Even recursive ones:

iris.template <- abstract(iris)
levels(iris$Species)[3] <- 'Sibirica' # corrupt iris

vet(iris.template, iris[1:10,])

Note the useful error messages.

Tweet #2

We made vetr fast to mitigate overhead concerns. There is a dedicated mode for fun param vetting (2/4)

Fast

Comparable in performance to stopifnot for simple checks, and faster for complex ones:

mx.3.col.num <- matrix(numeric(), ncol=3)
mx1 <- matrix(1:12, 4)

bench_mark(times=1e4,
  vet(mx.3.col.num, mx1),
  stopifnot(is.matrix(mx1), is.numeric(mx1), ncol(mx1) == 3)
)

In Functions

vetr() streamlines function parameter vetting:

fun <- function(x, y) {
  vetr(integer(), character(1L) || NULL)
}
fun(1, 'hello')
fun(1, NULL)
fun(1, 2)

Tweet #3

Create complex vetting expressions with intuitively programmable non-standard-evaluation (3/4)

Programmable NSE

vetr implements programmable NSE via recursive substitution of language objects. This allows you to construct complex vetting expressions from simple ones:

a <- quote(integer() && . > 0)
b <- quote(logical(1L) && !is.na(.))

vet(a || b, 1:3)
vet(a || b, -1)

c <- quote(a || b)  # equivalently

vet(c, -1)

Tweet #4

On CRAN, 100% coverage with #covr https://github.com/jimhester/covr and #unitizer https://github.com/brodieG/unitizer, but under dev so feedback welcome (4/4)

R Package Announce

Announcing New Package vetr on CRAN

vetr implements a declarative template-based approach to verify that objects meet structural requirements, and auto-compose error messages when they do not. This package is intended to simplify a more formal use of S3 objects.

The template concept is borrowed from vapply:

vet(numeric(1L), 1:3) [1] "1:3 should be length 1 (is 3)" vet(numeric(1L), "hello") [1] "\"hello\" should be type \"numeric\" (is \"character\")" vet(numeric(1L), 42) [1] TRUE

There is no limit on template complexity:

vet(matrix(integer(), ncol=3), matrix(1:12, 3)) [1] "matrix(1:12, 3) should have 3 columns (has 4)"

iris.template <- abstract(iris) levels(iris$Species)[3] <- 'sibirica' vet(iris.template, iris[1:10,]) [1] "levels((iris[1:10, ])$Species)[3] should be \"virginica\" (is \"sibirica\")"

vetr implements programmable non-standard evaluation via recursive substitution of language objects:

a <- quote(integer() && . > 0) b <- quote(logical(1L) && !is.na(.)) c <- quote(a || b) vet(c, -1) [1] "At least one of these should pass:" [2] " - -1 > 0 is not TRUE (FALSE)" [3] " - -1 should be type \"logical\" (is \"double\")"

Performance is comparable to stopifnot for simple checks, and is faster for complex (template based) ones. There is a mode that further streamlines parameter vetting in functions.

The package is still under development, but the features should be mostly stable. Feedback welcome (https://github.com/brodieG/vetr).

Best regards,

Brodie.

brodieG/vetr documentation built on Jan. 31, 2023, 5:56 a.m.