jwutil (version 1.2.3)

factor_nosort: Fast Factor Generation

Description

This function generates factors more quickly, without leveraging fastmatch. The speed increase with fastmatch for ICD-9 codes was about 33 using Rcpp, and a hashed matching algorithm.

Usage

factor_nosort(x, levels = NULL, labels = levels)

Arguments

x

An object of atomic type integer, numeric, character or logical.

levels

An optional character vector of levels. Is coerced to the same type as x. By default, we compute the levels as sort(unique.default(x)).

labels

A set of labels used to rename the levels, if desired.

Details

NaNs are converted to NA when used on numeric values. Extracted from https://github.com/kevinushey/Kmisc.git

These feature from base R are missing: exclude = NA, ordered = is.ordered(x), nmax = NA

I don't think there is any requirement for factor levels to be sorted in advance, especially not for ICD-9 codes where a simple alphanumeric sorting will likely be completely wrong.

Examples

# NOT RUN {
pts <- icd:::random_unordered_patients(1e7)
u <- unique.default(pts$code)
# this shows that stringr (which uses stringi) sort takes 50% longer than
# built-in R sort.
microbenchmark::microbenchmark(sort(u), str_sort(u))

# this shows that \code{factor_} is about 50% faster than \code{factor} for
# big vectors of strings

# without sorting is much faster:
microbenchmark::microbenchmark(factor(pts$code),
  # factor_(pts$code),
  factor_nosort(pts$code),
  times = 25
)
# }