Skip to contents

Grouped K-NN or PCA Imputation

Source: R/group_imp.R
group_imp.Rd

Perform K-NN or PCA imputation independently within feature groups.

Usage

group_imp(
  obj,
  group,
  subset = NULL,
  allow_unmapped = FALSE,
  k = NULL,
  ncp = NULL,
  method = NULL,
  cores = 1,
  .progress = TRUE,
  min_group_size = NULL,
  colmax = NULL,
  post_imp = NULL,
  dist_pow = NULL,
  scale = NULL,
  coeff.ridge = NULL,
  threshold = NULL,
  row.w = NULL,
  seed = NULL,
  nb.init = NULL,
  maxiter = NULL,
  miniter = NULL,
  solver = NULL,
  lobpcg_control = NULL,
  clamp = NULL,
  pin_blas = FALSE,
  na_check = TRUE,
  on_infeasible = c("error", "skip", "mean")
)

Arguments

obj

A numeric matrix with samples in rows and features in columns.

group

Specification of how features should be grouped for imputation. Accepted formats are:

  • A character scalar naming a supported Illumina platform; see Note.

  • A long-format data.frame with columns group and feature.

  • A list-column data.frame with a feature list-column. Optional list-columns are aux, for auxiliary feature names, and parameters, for group-specific parameter lists.

subset

Optional character vector of feature names to impute. If NULL, all grouped features are imputed. Features in a group but not in subset are demoted to auxiliary columns for that group. Groups left with zero features after demotion are dropped with a message.

allow_unmapped

Logical. If FALSE, every column in colnames(obj) must appear in group. If TRUE, columns with no group assignment are left untouched and are not used as auxiliary columns.

k

Integer or NULL. Number of nearest neighbors for K-NN imputation. If NULL, k may be supplied through group$parameters.

ncp

Integer or NULL. Number of components for PCA imputation. If NULL, ncp may be supplied through group$parameters.

method

Character or NULL. For K-NN imputation, one of "euclidean" or "manhattan". For PCA imputation, one of "regularized" or "EM". If NULL, the corresponding backend default is used unless overridden by group$parameters.

cores

Integer. Number of cores for K-NN imputation only. For PCA imputation, use mirai::daemons() to parallelize across groups.

.progress

Logical. If TRUE, show progress.

min_group_size

Integer or NULL. Minimum total number of columns per group, counting both features and auxiliary columns. Groups smaller than this are padded with randomly sampled columns from obj.

colmax

Numeric scalar between 0 and 1. Columns with a missing-data proportion greater than colmax are excluded from the main imputation method. Excluded columns are left unchanged unless post_imp = TRUE, in which case remaining missing values are replaced by column means when possible.

post_imp

Logical. If TRUE, replace missing values remaining after the main imputation method with column means when possible.

dist_pow

Numeric. Power used to penalize more distant neighbors in the weighted average. dist_pow = 0 gives an unweighted average of the nearest neighbors.

scale

Logical. If TRUE, columns are scaled to unit variance.

coeff.ridge

Numeric. Ridge regularization, used only when method = "regularized". Values < 1 move toward EM PCA; values > 1 move toward mean imputation.

threshold

Numeric. Convergence threshold.

row.w

Row weights, normalized to sum to 1. NULL (equal weights), a positive numeric vector of length nrow(obj), or "n_miss" (down-weight rows with more missing values).

seed

Integer, numeric, or NULL. Random seed for reproducibility.

nb.init

Integer. Number of random initializations. The first initialization is always mean imputation.

maxiter

Integer. Maximum number of iterations.

miniter

Integer. Minimum number of iterations.

solver

Character. Eigensolver: "auto" (default), "exact", or "lobpcg". "auto" runs a short timed probe and picks "lobpcg" only when clearly faster. Consecutive EM calls warm-start LOBPCG with both the previous eigenblock and search direction. When nb.init > 1, the auto choice from the first init is reused. See Performance tips.

lobpcg_control

A list of LOBPCG eigensolver control options, usually created by lobpcg_control(). A plain named list is also accepted. Ignored when solver = "exact".

clamp

Optional numeric vector c(lower, upper) bounding PCA-imputed values (use -Inf/Inf for one-sided, NULL for none). E.g., c(0, 1) for DNAm beta values. Observed values are not clamped.

pin_blas

Logical. If TRUE, pin BLAS threads to 1 to reduce contention when using parallel PCA on systems linked with multithreaded BLAS.

na_check

Logical. If TRUE, check whether the returned matrix still contains missing values.

on_infeasible

Character. One of "error", "skip", or "mean". Controls behavior when a group is infeasible for imputation, for example when k or ncp exceeds the number of usable columns after applying colmax.

Value

A numeric matrix of the same dimensions as obj, with missing values imputed. The returned object has class slideimp_results.

Details

group_imp() performs K-NN or PCA imputation on feature groups independently, which can substantially reduce runtime for large matrices.

Specify k and related arguments to use K-NN imputation, or ncp and related arguments to use PCA imputation. If both k and ncp are NULL, group$parameters must supply either k or ncp for every group.

Group-specific parameters in group$parameters take priority over global arguments. Global arguments fill in any missing values. All groups must use the same imputation method.

For method-specific arguments inherited from knn_imp() or pca_imp(), NULL means the backend default is used unless overridden by group$parameters.

Per-group k is capped at the number of usable columns in the group minus one. Per-group ncp is capped at the maximum feasible number of PCA components for that group's submatrix. A warning is issued when capping occurs.

Note

A character scalar can be passed to group to name a supported Illumina platform, such as "EPICv2" or "EPICv2_deduped". This requires the optional slideimp.extra package to be installed. Supported platforms are listed in the slideimp_arrays object in the slideimp.extra package.

Parallelization

  • K-NN: use the cores argument. If mirai daemons are active, cores is automatically set to 1 to avoid nested parallelism.

  • PCA: use mirai::daemons() instead of cores.

When running PCA imputation in parallel with mirai, set pin_blas = TRUE in tune_imp() or group_imp() to prevent BLAS threads from oversubscribing CPU cores. This relies on RhpcBLASctl and works with OpenBLAS and MKL (typical on Linux, and on Windows after an OpenBLAS swap). pin_blas = TRUE may have no effect on macOS.

PCA Performance tips

Speed comes from three levers: solver (through LOBPCG with warm-start), threshold, and scale. Tune these first, then accuracy parameters (ncp, coeff.ridge) on a representative subset.

Exact vs. LOBPCG with warm-start. Whether "lobpcg" beats "exact" depends on size and low-rankness: "lobpcg" is preferred for large, approximately low-rank matrices with small ncp, and "exact" for small matrices (including slide_imp() windows), where it is faster and more robust. Separately, the warm-start makes each successive solve cheap: pca_imp() warm-starts LOBPCG with the previous eigenblock and search direction, so once imputed values stabilize, later solves converge in a few iterations. The payoff therefore grows with the number of EM iterations, independent of low-rankness. solver = "auto" (default) probes both and is a safe start.

Threshold. The default 1e-6 is conservative; 1e-5 is often faster with very similar values.

Scale. For columns on a common scale (e.g., DNAm beta values in [0, 1]), scale = FALSE can be faster and more accurate.

Parallel and BLAS. In parallel via tune_imp() or group_imp() with a multithreaded BLAS, set pin_blas = TRUE to avoid thread oversubscription. On Windows, the stock BLAS can be slow. Advanced users can swap in OpenBLAS.

See Speeding up PCA imputation for the full workflow.

See also

prep_groups(), knn_imp(), pca_imp()

Examples

set.seed(1234)
to_test <- sim_mat(10, 20, perc_total_na = 0.05, perc_col_na = 1)
obj <- to_test$input
group <- to_test$col_group
head(group)
#>    feature  group
#> 1 feature1 group2
#> 2 feature2 group1
#> 3 feature3 group1
#> 4 feature4 group2
#> 5 feature5 group1
#> 6 feature6 group1

# Simple grouped K-NN imputation
results <- group_imp(obj, group = group, k = 2, .progress = FALSE)
#> Imputing 2 groups using KNN.
#> Running mode: sequential
results
#> Method: group_imp (KNN imputation)
#> Dimensions: 10 x 20
#> 
#>          feature1  feature2  feature3  feature4   feature5  feature6
#> sample1 0.1568098 0.3213953 0.2768746 1.0000000 0.07839046 0.4219375
#> sample2 0.4098416 0.8658918 0.8221066 0.6396885 0.68926345 1.0000000
#> sample3 0.6801685 1.0000000 1.0000000 0.9953450 0.75246030 0.6958550
#> sample4 0.0000000 0.0000000 0.0000000 0.0000000 0.00000000 0.0000000
#> sample5 0.9639671 0.6409137 0.5111760 0.7184678 0.81815288 0.5883255
#> sample6 0.7031741 0.3731062 0.6782811 0.7542872 0.99910554 0.9070005
#> # Showing 6 of 10 rows and 6 of 20 columns

# Impute only a subset of features
subset_features <- sample(to_test$col_group$feature, size = 10)
knn_subset <- group_imp(
  obj,
  group = group,
  subset = subset_features,
  k = 2,
  .progress = FALSE
)
#> Imputing 2 groups using KNN.
#> Running mode: sequential

# Use prep_groups() to inspect and edit per-group parameters
prepped <- prep_groups(colnames(obj), group)
prepped$parameters <- lapply(seq_len(nrow(prepped)), function(i) list(k = 2))
prepped$parameters[[2]]$k <- 4
knn_grouped <- group_imp(obj, group = prepped, .progress = FALSE)
#> Imputing 2 groups using KNN.
#> Running mode: sequential

if (FALSE) { # interactive() && requireNamespace("mirai", quietly = TRUE)
# PCA imputation with mirai parallelism
mirai::daemons(2)
pca_grouped <- group_imp(obj, group = group, ncp = 2)
mirai::daemons(0)
pca_grouped
}