mikropml (version 1.7.0)

preprocess_data: Preprocess data prior to running machine learning

Description

Function to preprocess your data for input into run_ml().

Usage

preprocess_data(dataset, ...)
# S4 method for TreeSummarizedExperiment
preprocess_data(
  dataset,
  outcome_colname,
  assay.type = "counts",
  col.var = NULL,
  altexp = NULL,
  name = "preprocessed",
  ...
)
# S4 method for ANY
preprocess_data(
  dataset,
  outcome_colname,
  method = c("center", "scale"),
  remove_var = "nzv",
  collapse_corr_feats = TRUE,
  corr_method = "spearman",
  corr_thresh = 1,
  to_numeric = TRUE,
  group_neg_corr = TRUE,
  prefilter_threshold = 1,
  ...
)

Value

Named list including:

• dat_transformed: Preprocessed data.

• grp_feats: If features were grouped together, a named list of the features corresponding to each group.

• removed_feats: Any features that were removed during preprocessing (e.g. because there was zero variance or near-zero variance for those features).

If the input is TreeSummarizedExperiment, the output is added as an additional data to the input object. If the set of features match in output and input, the results are stored directly to assay slot. If they do not match, the output is stored to altExp slot of the object.

If the progressr package is installed, a progress bar with time elapsed and estimated time to completion can be displayed.

Arguments

dataset

Data frame with an outcome variable and other columns as features. Alternatively, the input can be in TreeSummarizedExperiment format.

...

All additional arguments are passed on to caret::train(), such as case weights via the weights argument or ntree for rf models. See the caret::train() docs for more details.

outcome_colname

Column name as a string of the outcome variable (default NULL; the first column will be chosen automatically).

assay.type

The name of assay from dataset when the object is in TreeSummarizedExperiment format. This assay is used as an input.

col.var

The name of sample matdata variables from colData slot of dataset when the object is in TreeSummarizedExperiment format. These variables are used as predictors.

altexp

The name of alternative experiment (altExp) from dataset when the object is in TreeSummarizedExperiment format. This can be used to select an experiment for the input.

name

Name of results used when the input is TreeSummarizedExperiment. This same name is used for assay and altExp.

method

Methods to preprocess the data, described in caret::preProcess() (default: c("center","scale"), use NULL for no normalization).

remove_var

Whether to remove variables with near-zero variance ('nzv'; default), zero variance ('zv'), or none (NULL).

collapse_corr_feats

Whether to keep only one of correlated features (see corr_method and corr_thresh)

corr_method

Correlation method. Options are the same as those supported by stats::cor: spearman, pearson, kendall. (default: spearman)

corr_thresh

group correlations above or equal to corr_thresh (range 0 to 1; default: 1).

to_numeric

Whether to change features to numeric where possible.

group_neg_corr

Whether to group negatively correlated features together (e.g. c(0,1) and c(1,0)).

prefilter_threshold

Remove features which only have non-zero & non-NA values in N rows or fewer (default: 1). Set this to -1 to keep all columns at this step. This step will also be skipped if to_numeric is set to FALSE.

More details

See the preprocessing vignette for more details.

Note that if any values in outcome_colname contain spaces, they will be converted to underscores for compatibility with caret.

Author

Zena Lapp,

Kelly Sovacool,

Examples

preprocess_data(mikropml::otu_small, "dx")

# the function can show a progress bar if you have the progressr package installed
## optionally, specify the progress bar format
progressr::handlers(progressr::handler_progress(
  format = ":message :bar :percent | elapsed: :elapsed | eta: :eta",
  clear = FALSE,
  show_after = 0
))
## tell progressor to always report progress
if (FALSE) {
progressr::handlers(global = TRUE)
## run the function and watch the live progress udpates
dat_preproc <- preprocess_data(mikropml::otu_small, "dx")

# Create TreeSE object
library(TreeSummarizedExperiment)
df <- mikropml::otu_small
assay <- df[, !colnames(df) %in% c("dx"), drop = FALSE] |> t() |> as.matrix()
tse <- TreeSummarizedExperiment(assays = SimpleList(counts = assay))
colData(tse)[["dx"]] <- df[["dx"]]

# Preprocess
tse <- preprocess_data(
  dataset = tse,
  assay.type = "counts",
  outcome_colname = "dx"
)
# The result is in assay slot
tse
}