na.tools is a comprehensive library for handling missing (NA) values. It has several goals:
stats::na.*()
functions,In this package, there are methods for the detection, removal,
replacement, –imputation–, recollection, etc. of missing values
(NAs
). This libraries focus is on vectors (atomics). For
tidy/dplyr compliant methods operating
on tables and lists, please use the tidyimpute
package which depends on this package.
devtools::install_github( "decisionpatterns/na.tools")
install.packages("na.tools")
na.*
functions found in the
stats package.n_na
,
pct_na
na.rm
na.
return a transformed
version of the input vector with missing values imputes/x <- 1:3
x[2] <- NA_real_
any_na(x)
all_na(x)
which_na(x)
n_na(x)
pct_na(x)
na.rm(x)
na.replace(x, 2)
na.replace(x, mean) # error
na.replace(x, na.mean) # Works
na.zero(x)
na.mean(x)
na.cumsum(x)
na.n
- Count mising valuesna.pct
- Calculate pct of missing valueswhich.na
- Return logical or character indicating which
elements are missingall.na
(na.all
) - test if all elements are
missingany.na
(na.any
) - test if any elements are
missingna.rm
- remove NA
s (with tables is
equivalent to drop_cols_all_na
)na.trim
- remove NA
s from beginning or end
(non-commutative/order matters)There are two types of imputation methods for plain vectors. They are distinguished by their replacement values.
In “constant” imputation methods, missing values are replaced by an a priori selected constant value. No calculation are performed to derive replacement values and all missing value assume the same transformied value.
na.zero
: Replace NA
s with 0na.true
| na.false
: …
TRUE
na.inf
/ na.neginf
: … Inf
/
-Inf
na.constant
: constant value .na
In functional imputation, the value is calculated from the vector containing the missing value(s) – and only that vector. Missing values may impute to different values. Replacement values may (or may not) be affected by the ording of the vector.
Cummatative functions
Commutative functions provide the same result irregarless of the ordering of the input vectors. Therefore, these functions do not depend on the ordering of elements of the input vector.
(When imputing in a table, imputation by function is also called column-based imputation since replacement values derive from the single column. Table-based imputation is found in the tidyimpute package.)
na.max
- maximumna.min
- minumumna.mean
- meanna.median
- median valuena.quantile
- quantile valuena.sample
/na.random
- randomly sampled
value** Non-commulative functions **s
na.cummax
- cumulative maxna.cummin
- cumulative minna.cumsum
- cumulative sumna.cumprod
- cumulative prodGeneral Imputation
na.replace
/na.explicit
- atomic vectors
only. General replacement functionna.unreplace
/na.implicit
- turn explicit
values back into NAsA number of other packages have methods for working with missing values and/or imputation. Here is a short, incomplete and growing list:
randomForest::na.roughfix()
- imputes with
median
zoo::na.*
- collection of non-commutative
imputation techniques for time series data.mitools provides tools for multiple imputation, mice provides multivariate imputation by chained equations mvnmle provides ML estimation for multivariate normal data with missing values, mix provides multiple imputation for mixed categorical and continuous data. pan provides multiple imputation for missing panel data. VIM provides methods for the visualisation as well as imputation of missing data. aregImpute() and transcan() from Hmisc provide further imputation methods. monomvn deals with estimation models where the missing data pattern is monotone.