mimar implements a compact chained-imputation workflow in R for
missing-data analysis, artificial amputation, native and learner-backed
single and multiple imputation, diagnostic evaluation, and
post-imputation pooling.
The package is built around a complete missing-data workflow: describe the missingness, create benchmark amputations when needed, impute with native or learner-backed update rules, inspect diagnostics, evaluate recovered cells when truth is available, and pool post-fit quantities. The goal is a concise grammar for the whole workflow, not a replacement for every specialist feature in larger imputation systems.
The package owns the imputation loop. Every imputer, whether implemented natively or backed by a learner package, is called the same way:
impute(data, imputer = "pmm", m = 5, maxit = 5, seed = 1)
impute(data, imputer = "rf", m = 5, seed = 1)
impute(data, imputer = "xgboost", m = 5, seed = 1)There is no dependency on funcml. Learner-backed imputers call their
original packages directly, and those backend packages are hard
dependencies so users can run any registered imputer without manually
resolving learner installations.
Install the development version from GitHub:
install.packages("remotes")
remotes::install_github("ielbadisy/mimar")Then load the package:
library(mimar)For normal use, impute() is the only function you need. The input data
can contain NA, and the completed outputs returned by complete() do
not. Set verbose = TRUE when you want a concise progress log for the
chained imputation workflow.
i <- impute(a, imputer = "knn", m = 5, maxit = 5, seed = 1)
complete(i, 1)
complete(i, "all")describe()
ampute()
imputer_registry()
imputer()
impute()
complete()
evaluate()
pool()
plot()library(mimar)
set.seed(1)
dat <- data.frame(
age = rnorm(120, 50, 10),
bmi = rnorm(120, 25, 4),
sex = factor(sample(c("F", "M"), 120, TRUE)),
group = factor(sample(c("A", "B", "C"), 120, TRUE)),
smoker = sample(c(TRUE, FALSE), 120, TRUE)
)
a <- ampute(
dat,
prop = 0.25,
mechanism = "MAR",
target = c("bmi", "group"),
by = c("age", "sex"),
seed = 1
)
i <- impute(a, imputer = "knn", m = 5, maxit = 5, seed = 1, ncore = 2)
complete(i, 1)## # A tibble: 120 × 5
## age bmi sex group smoker
## <dbl> <dbl> <fct> <fct> <lgl>
## 1 43.7 23.0 M C FALSE
## 2 51.8 30.4 F A FALSE
## 3 41.6 24.1 F B TRUE
## 4 66.0 24.3 F C TRUE
## 5 53.3 24.6 F A FALSE
## 6 41.8 27.9 M C TRUE
## 7 54.9 24.7 M B TRUE
## 8 57.4 24.8 F B FALSE
## 9 55.8 22.3 F B FALSE
## 10 46.9 30.8 M A FALSE
## # ℹ 110 more rows
summary(i)## mimar imputation summary
## # A tibble: 1 × 11
## rows columns n_imputations imputer maxit ncore stochastic
## <int> <int> <int> <chr> <dbl> <int> <lgl>
## 1 120 5 5 knn 5 2 TRUE
## # ℹ 4 more variables: total_missing_before <int>, total_imputed <int>,
## # remaining_missing <int>, variables_imputed <int>
##
## Variables:
## # A tibble: 5 × 9
## variable type method n_missing_before prop_missing_before n_imputed
## <chr> <chr> <chr> <int> <dbl> <int>
## 1 age numeric none 0 0 0
## 2 bmi numeric knn 26 0.217 26
## 3 sex factor none 0 0 0
## 4 group factor knn 27 0.225 27
## 5 smoker logical none 0 0 0
## # ℹ 3 more variables: prop_imputed <dbl>, remaining_missing <int>,
## # between_imputation_sd <dbl>
evaluate(i)## mimar imputation evaluation
## # A tibble: 1 × 4
## n_imputations imputer total_missing evaluated_cells
## <int> <chr> <int> <int>
## 1 5 knn 53 53
plot(i, type = "density")
Inspect available imputers with:
imputer_registry()## # A tibble: 23 × 10
## imputer implementation package supports_numeric supports_binary
## <chr> <chr> <chr> <lgl> <lgl>
## 1 mean mimar internal TRUE TRUE
## 2 median mimar internal TRUE TRUE
## 3 mode mimar internal TRUE TRUE
## 4 naive mimar internal TRUE TRUE
## 5 norm mimar internal TRUE TRUE
## 6 pmm mimar internal TRUE TRUE
## 7 spmm mimar internal TRUE TRUE
## 8 logreg mimar internal TRUE TRUE
## 9 polyreg mimar internal TRUE TRUE
## 10 rf wrapped ranger TRUE TRUE
## # ℹ 13 more rows
## # ℹ 5 more variables: supports_multiclass <lgl>, stochastic <lgl>,
## # description <chr>, available <lgl>, status <chr>
describe("imputers")## mimar available imputers
## # A tibble: 23 × 10
## imputer implementation package supports_numeric supports_binary
## <chr> <chr> <chr> <lgl> <lgl>
## 1 mean mimar internal TRUE TRUE
## 2 median mimar internal TRUE TRUE
## 3 mode mimar internal TRUE TRUE
## 4 naive mimar internal TRUE TRUE
## 5 norm mimar internal TRUE TRUE
## 6 pmm mimar internal TRUE TRUE
## 7 spmm mimar internal TRUE TRUE
## 8 logreg mimar internal TRUE TRUE
## 9 polyreg mimar internal TRUE TRUE
## 10 rf wrapped ranger TRUE TRUE
## # ℹ 13 more rows
## # ℹ 5 more variables: supports_multiclass <lgl>, stochastic <lgl>,
## # description <chr>, available <lgl>, status <chr>
Core native imputers:
mean,median,modenaive: median/mode chained baselinenorm: linear normal drawpmm,spmm: predictive mean matchinglogreg: binary logistic regression drawpolyreg: one-vs-rest multinomial drawknn: nearest-neighbor donor imputationhotdeck: stochastic donor imputation
Learner-backed imputers:
rf: MissForest-style chained random forest imputer throughrangerranger: random forest throughrangerrpart: tree imputer throughrpartnbayes: naive Bayes throughnaivebayessvm: support vector machine throughe1071bart: Bayesian additive regression trees throughBARTglmnet: penalized regression throughglmnetgbm: gradient boosting throughgbmxgboost: gradient boosted trees throughxgboostfamd: FAMD-assisted donor imputation throughmissMDAsuperlearner,sl: cross-validated Super Learner-style ensemble imputer
Imputer names are strict: use the names shown by imputer_registry().
Learner-backed imputers are applied as requested to numeric, binary, and
multiclass targets; mimar does not silently swap them for another
imputer inside benchmark runs.
The ncore argument runs independent completed datasets in parallel.
The parallel boundary is the outer imputation index: each completed
dataset gets a deterministic seed offset, so a fixed seed, m,
maxit, and imputer remain reproducible.
i <- impute(a, imputer = "knn", m = 5, maxit = 5, seed = 1, ncore = 2)Use ncore = 1 for sequential execution, small examples, and the most
conservative behavior in constrained environments.
Learner-backed imputers expose their hyperparameters through imputer()
or directly through ... in impute(). Donor-based imputers use the
explicit donors argument.
rf_spec <- imputer("rf", num.trees = 500)
xgb_spec <- imputer("xgboost", nrounds = 100, max_depth = 3)
i1 <- impute(a, imputer = rf_spec, m = 5, maxit = 5, seed = 1)
i2 <- impute(a, imputer = "xgboost", m = 5, maxit = 5, seed = 1,
nrounds = 100, max_depth = 3)
i3 <- impute(a, imputer = "knn", m = 5, maxit = 5, seed = 1, donors = 10)The same hyperparameter set is reused across all incomplete variables that a given imputer supports, which keeps the full chained-imputation pipeline reproducible and easy to tune.
superlearner combines candidate imputers by cross-validating them on
observed cells, assigning non-negative loss-based weights, and using the
weighted ensemble inside the chained-imputation loop.
sl <- imputer(
"superlearner",
library = c("pmm", "knn", "rpart"),
folds = 5,
metalearner = "inverse_loss"
)
i_sl <- impute(a, imputer = sl, m = 5, maxit = 5, seed = 1)The short alias imputer = "sl" is equivalent to
imputer = "superlearner".
plot() methods return ggplot objects. For mimar_imputation
objects, the main diagnostic types are:
plot(i) # imputed cell counts
plot(i, type = "missing") # observed/imputed cell map
plot(i, type = "trace", statistic = "mean") # convergence-screening trace
plot(i, type = "density", variable = "bmi") # line-only density overlays
plot(i, type = "boxplot", variable = "bmi") # observed vs imputation 1:m
plot(i, type = "strip", variable = "bmi") # individual values by imputation
Formula diagnostics are available for bivariate and categorical checks:
plot(i, type = "xy", formula = bmi ~ age | sex)
plot(i, type = "proportion", formula = group ~ sex)
For type = "xy", formulas use y ~ x or y ~ x | group. For
type = "proportion", formulas use categorical_variable ~ strata.
Density diagnostics use line-only overlays so several imputations remain
visible rather than obscuring each other with filled areas.
Let X be an n x p data frame and let R_ij = 1 when cell (i, j)
is missing. For each incomplete variable X_j:
O_j = {i : R_ij = 0}are the observed rowsM_j = {i : R_ij = 1}are the missing rows
At each chained update, mimar fits an imputer-specific model from the
observed rows and then predicts the missing rows from the current
completed data. In compact form:
fit model on X_-j, O_j -> X_j, O_j
update X_j, M_j using the fitted model
Multiple imputation repeats the same chained procedure m times with
controlled seeds, bootstrap samples of observed rows, and stochastic
prediction where supported.
Learner-backed imputers are practical stochastic update rules inside this chained workflow. They can improve predictive recovery, but users should still inspect trace, distribution, categorical-proportion, and downstream sensitivity diagnostics rather than assuming every learner automatically supplies proper multiple-imputation uncertainty for every analysis.
Input: X, R, h, m, T
Initialize: X~(0) <- init(X)
For k = 1,...,m:
X~_k(0) <- X~(0)
For t = 1,...,T:
For each incomplete variable j:
B_j <- bootstrap sample of O_j
fit h on X~_k, B_j, -j and X_Bj,j
update missing rows M_j using the fitted model
restore observed rows O_j to their original values
Return: {X~_1(T), ..., X~_m(T)}
When imputation is run on an ampute() object, evaluate() uses the
retained truth and scores only artificially removed cells. Numeric
recovery reports RMSE, MAE, bias, and correlation. Categorical recovery
reports accuracy and balanced accuracy.
pool() combines post-fit quantities estimated separately in each
completed dataset. The statistical target is the quantity itself, not a
data frame. A quantity can be a scalar, coefficient vector,
covariance-aware parameter vector, matrix of survival probabilities, or
a scalar metric. Data frames are accepted only as a tidy adapter for
scalar model output.
For scalar quantities with complete-data variance estimates, pool()
applies Rubin-style pooling:
Q_bar = mean(Q_k)
U_bar = mean(U_k)
B = sample variance of Q_k
T = U_bar + (1 + 1/m) * B
results <- data.frame(
term = rep(c("age", "bmi"), each = 3),
estimate = c(0.10, 0.11, 0.09, 0.30, 0.32, 0.29),
std.error = c(0.04, 0.05, 0.04, 0.08, 0.09, 0.08),
imputation = rep(1:3, times = 2)
)
pool(results)## mimar pooled results
## # A tibble: 2 × 14
## term estimate std.error statistic df p.value conf.low conf.high m
## <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <int>
## 1 age 0.1 0.0451 2.22 465. 0.0271 0.0114 0.189 3
## 2 bmi 0.303 0.0853 3.56 1094. 0.000393 0.136 0.471 3
## # ℹ 5 more variables: within_variance <dbl>, between_variance <dbl>,
## # total_variance <dbl>, relative_increase_variance <dbl>, rule <chr>
Direct quantity inputs are preferred when available:
pool(c(0.10, 0.11, 0.09), std.error = c(0.04, 0.05, 0.04), name = "age")## mimar pooled results
## # A tibble: 1 × 14
## term estimate std.error statistic df p.value conf.low conf.high m
## <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <int>
## 1 age 0.1 0.0451 2.22 465. 0.0271 0.0114 0.189 3
## # ℹ 5 more variables: within_variance <dbl>, between_variance <dbl>,
## # total_variance <dbl>, relative_increase_variance <dbl>, rule <chr>
betas <- list(
c(age = 0.10, bmi = 0.30),
c(age = 0.11, bmi = 0.32),
c(age = 0.09, bmi = 0.29)
)
covariances <- list(
diag(c(0.04, 0.08)^2),
diag(c(0.05, 0.09)^2),
diag(c(0.04, 0.08)^2)
)
pool(betas, covariance = covariances)## mimar pooled results
## # A tibble: 2 × 14
## term estimate std.error statistic df p.value conf.low conf.high m
## <chr> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <dbl> <int>
## 1 age 0.1 0.0451 2.22 465. 0.0271 0.0114 0.189 3
## 2 bmi 0.303 0.0853 3.56 1094. 0.000393 0.136 0.471 3
## # ℹ 5 more variables: within_variance <dbl>, between_variance <dbl>,
## # total_variance <dbl>, relative_increase_variance <dbl>, rule <chr>
When no reliable complete-data variance is supplied, as is common for
some performance metrics, pool() reports robust summaries by default:
median, interquartile range, and range across imputations.
Learner backends are hard dependencies. Installing mimar installs the
packages needed by the registered learner-backed imputers, including
ranger, rpart, naivebayes, e1071, BART, glmnet, gbm,
xgboost, and missMDA.