recode-replace: Recode and Replace Values in Matrix-Like Objects

In collapse: Advanced and Fast Data Transformation

Recode and Replace Values in Matrix-Like Objects

Description

A small suite of functions to efficiently perform common recoding and replacing tasks in matrix-like objects.

Usage

recode_num(X, ..., default = NULL, missing = NULL, set = FALSE)

recode_char(X, ..., default = NULL, missing = NULL, regex = FALSE,
            ignore.case = FALSE, fixed = FALSE, set = FALSE)

replace_na(X, value = 0, cols = NULL, set = FALSE, type = "const")

replace_inf(X, value = NA, replace.nan = FALSE, set = FALSE)

replace_outliers(X, limits, value = NA,
                 single.limit = c("sd", "mad", "min", "max"),
                 ignore.groups = FALSE, set = FALSE)

Arguments

X	a vector, matrix, array, data frame or list of atomic objects. replace_outliers has internal methods for grouped and indexed data.
...	comma-separated recode arguments of the form: value = replacement, `2` = 0, Secondary = "SEC" etc. recode_char with regex = TRUE also supports regular expressions i.e. `^S|D$` = "STD" etc.
default	optional argument to specify a scalar value to replace non-matched elements with.
missing	optional argument to specify a scalar value to replace missing elements with. Note that to increase efficiency this is done before the rest of the recoding i.e. the recoding is performed on data where missing values are filled!
set	logical. TRUE does replacements by reference (i.e. in-place modification of the data) and returns the result invisibly.
type	character. One of "const", "locf" (last non-missing observation carried forward) or "focb" (first non-missing observation carried back). The latter two ignore value.
regex	logical. If TRUE, all recode-argument names are (sequentially) passed to grepl as a pattern to search X. All matches are replaced. Note that NA's are also matched as strings by grepl.
value	a single (scalar) value to replace matching elements with. In replace_outliers setting value = "clip" will replace outliers with the corresponding threshold values. See Examples.
cols	select columns to replace missing values in using a function, column names, indices or a logical vector.
replace.nan	logical. TRUE replaces NaN/Inf/-Inf. FALSE (default) replaces only Inf/-Inf.
limits	either a vector of two-numeric values c(minval, maxval) constituting a two-sided outlier threshold, or a single numeric value:
single.limit	character, controls the behavior if length(limits) == 1: "sd"/"mad": limits will be interpreted as a (two-sided) outlier threshold in terms of (column) standard deviations/median absolute deviations. For the standard deviation this is equivalent to X[abs(fscale(X)) > limits] <- value. Since fscale is S3 generic with methods for 'grouped_df', 'pseries' and 'pdata.frame', the standardizing will be grouped if such objects are passed (i.e. the outlier threshold is then measured in within-group standard deviations) unless ignore.groups = TRUE. The same holds for median absolute deviations. "min"/"max": limits will be interpreted as a (one-sided) minimum/maximum threshold. The underlying code is equivalent to X[X </> limits] <- value.
ignore.groups	logical. If length(limits) == 1 and single.limit %in% c("sd", "mad") and X is a 'grouped_df', 'pseries' or 'pdata.frame', TRUE will ignore the grouped nature of the data and calculate outlier thresholds on the entire dataset rather than within each group.
ignore.case, fixed	logical. Passed to grepl and only applicable if regex = TRUE.

Details

• recode_num and recode_char can be used to efficiently recode multiple numeric or character values, respectively. The syntax is inspired by dplyr::recode, but the functionality is enhanced in the following respects: (1) when passed a data frame / list, all appropriately typed columns will be recoded. (2) They preserve the attributes of the data object and of columns in a data frame / list, and (3) recode_char also supports regular expression matching using grepl.

• replace_na efficiently replaces NA/NaN with a value (default is 0). data can be multi-typed, in which case appropriate columns can be selected through the cols argument. For numeric data a more versatile alternative is provided by data.table::nafill and data.table::setnafill.

• replace_inf replaces Inf/-Inf (or optionally NaN/Inf/-Inf) with a value (default is NA). It skips non-numeric columns in a data frame.

• replace_outliers replaces values falling outside a 1- or 2-sided numeric threshold or outside a certain number of standard deviations or median absolute deviation with a value (default is NA). It skips non-numeric columns in a data frame.

Note

These functions are not generic and do not offer support for factors or date(-time) objects. see dplyr::recode_factor, forcats and other appropriate packages for dealing with these classes.

Simple replacing tasks on a vector can also effectively be handled by, setv / copyv. Fast vectorized switches are offered by package kit (functions iif, nif, vswitch, nswitch) as well as data.table::fcase and data.table::fifelse. Using switches is more efficient than recode_*, as recode_* creates an internal copy of the object to enable cross-replacing.

Function TRA, and the associated TRA ('transform') argument to Fast Statistical Functions also has option "replace_na", to replace missing values with a statistic computed on the non-missing observations, e.g. fmedian(airquality, TRA = "replace_na") does median imputation.

See Also

pad, Efficient Programming, Collapse Overview

Examples

recode_char(c("a","b","c"), a = "b", b = "c")
recode_char(month.name, ber = NA, regex = TRUE)
mtcr <- recode_num(mtcars, `0` = 2, `4` = Inf, `1` = NaN)
replace_inf(mtcr)
replace_inf(mtcr, replace.nan = TRUE)
replace_outliers(mtcars, c(2, 100))                 # Replace all values below 2 and above 100 w. NA
replace_outliers(mtcars, c(2, 100), value = "clip") # Clipping outliers to the thresholds
replace_outliers(mtcars, 2, single.limit = "min")   # Replace all value smaller than 2 with NA
replace_outliers(mtcars, 100, single.limit = "max") # Replace all value larger than 100 with NA
replace_outliers(mtcars, 2)                         # Replace all values above or below 2 column-
                                                    # standard-deviations from the column-mean w. NA
replace_outliers(fgroup_by(iris, Species), 2)       # Passing a grouped_df, pseries or pdata.frame
                                                    # allows to remove outliers according to
                                                    # in-group standard-deviation. see ?fscale

collapse documentation built on Nov. 3, 2024, 9:08 a.m.