Package 'mlfeaturer'

Machine Learning and Feature Engineering for Regression Models

Description

This package provides preprocessing tools using S4 classes for data transformation, scaling, and inverse transformation.

Author(s)

Maintainer: Thomas Petzoldt [email protected] (ORCID)

Authors:

• Thomas Petzoldt [email protected] (ORCID)

See Also

Useful links:

• https://tpetzoldt.github.io/mlfeaturer/

---

Convert feature_data to feature_data_list

Description

Coerces a 'feature_data' object to a 'feature_data_list' object.

Coerces a 'feature_data' object to a standard R 'list' object.

Arguments

from	A 'feature_data' object.

Value

A 'feature_data_list' object.

A 'list' object.

---

Convert feature_data to a list

Description

Coerces a 'feature_data' object to a 'list' object.

Usage

## S4 method for signature 'feature_data'
as.list(x)

Arguments

x	A 'feature_data' object.

Value

A 'list' object.

---

Create Preprocessed Data Object

Description

Constructs a 'feature_data' object, calculates scaling parameters, and optionally applies initial data transformations.

Usage

create_preprocessed_data(
  data,
  target_col,
  id_col = NULL,
  split_col = NULL,
  scale_option = c("train", "test", "all"),
  scale_method = c("zscore", "minmax", "none"),
  fun_transform = NULL,
  fun_inverse = NULL
)

Arguments

data	A data frame containing the raw data.
target_col	Character vector specifying the name of the target variable column.
id_col	Character vector specifying the name of the ID column (optional).
split_col	Character vector specifying the name of the column used for train/test split.
scale_option	Character string specifying the scaling option. Must be one of "train", "test", or "all". Defaults to "train".
scale_method	Character string specifying the scaling method. Must be one of "zscore" (standardization) or "minmax" (normalization). Defaults to "zscore".
fun_transform	A named list of functions to apply to the data *before* scaling. The names of the list elements should correspond to the columns to transform.
fun_inverse	A named list of functions representing the inverse transformations of 'fun_transform'. These are applied during the 'inverse_transform' method.

Details

This function calculates scaling parameters (mean, standard deviation, min, max) based on the specified 'scale_option' and 'scale_method'. It then creates a 'feature_data' object, storing the original data and the calculated parameters. If 'fun_transform' is provided, it applies the specified transformations to the data before creating the object.

If no 'id_col' is given, an additional column 'id' with row numers is created.

If a 'split_col' is missing, an random 'split' column is created.

Value

A 'feature_data' object.

Examples

# Example usage (replace with your actual data and column names)
df <- data.frame(
  id = 1:10,
  x = runif(10),
  y = rnorm(10),
  z = 1:10,
  split = sample(c(TRUE, FALSE), 10, replace = TRUE)
)

transformations <- list(
  x = log,
  y = \(y) sqrt(5 + y),
  z = \(z) z^2
)

inverse_transformations <- list(
  x = exp,
  y = \(y) y^2 - 5,
  z = sqrt
)

prep_data <- create_preprocessed_data(
  df, id_col = "id", target_col = "y", split_col = "split",
  fun_transform = transformations, fun_inverse = inverse_transformations)
print(prep_data)

---

Growthrate of Algae

Description

Growthrates of four different phytoplankton species, dependent on temperature and light.

Format

Data frame with the following columns:

species

Phytoplankton species.

temperature

Temperature in °C.

light

Photosynthetic active radiation in µ E m^-2s^-1.

growthrate

Population growth rate in d^-1.

Source

The example data were taken manually from the figures of Dauta et al. 1990.

References

Dauta, A, Devaux, J, Piquemal, F and Boumnich, L (1990) Growth Rate of Four Algae in Relation to Light and Temperature. Hydrobiologia 207: 221–26. doi:10.1007/BF00041459

---

List Representation of Preprocessed Data

Description

An S4 class to represent the preprocessed data as a list.

Slots

x_train

Training data (X).

x_test

Test data (X).

x_all

All data (X).

y_train

Training data (Y).

y_test

Test data (Y).

y_all

All data (Y).

x_train_scaled

Scaled training data (X).

x_test_scaled

Scaled test data (X).

x_all_scaled

Scaled all data (X).

---

Preprocessed Data

Description

An S4 class to store the preprocessed data and its associated parameters.

Slots

data

Data frame containing the preprocessed data.

params

'feature_params' object storing the parameters used in preprocessing.

---

Preprocessing Parameters

Description

An S4 class to store the parameters used in the preprocessing pipeline.

Slots

id_col

Character vector specifying the name of the ID column.

target_col

Character vector specifying the name of the target variable column.

split_col

Character vector specifying the name of the column used for train/test split.

scale_option

Character string specifying the scaling option ("train", "test", "all").

scale_method

Character string specifying the scaling method ("scale", "norm").

mean_vals

Numeric vector storing the mean values for z-score scaling.

sd_vals

Numeric vector storing the standard deviation values for z-score scaling.

min_vals

Numeric vector storing the minimum values for minmax normalization.

max_vals

Numeric vector storing the maximum values for minmax normalization.

t_mean_vals

Numeric vector storing the mean of transformed values for z-score scaling.

t_sd_vals

Numeric vector storing the standard deviation of transformed values for z-score scaling.

t_min_vals

Numeric vector storing the minimum values of transformed for minmax normalization.

t_max_vals

Numeric vector storing the maximum values of transformed for minmax normalization.

fun_transform

List of functions for data transformation.

fun_inverse

List of functions for inverse data transformation.

transformed

Boolean value, 'TRUE' if object contains transformed values, 'NA' if it is the result of an inverse transformation.

---

Get All Data (X and Y)

Description

Access complete data set from the preprocessed data.

Usage

get_data(object, ...)

## S4 method for signature 'feature_data'
get_data(
  object,
  prep = c("both", "scale", "transform", "none"),
  as_matrix = TRUE
)

Arguments

object	A 'feature_data' object.
...	Additional arguments (currently not used).
prep	Character argument if transformed data ("transform"), scaled data ("scale" ), transformed and scaled data ("both") or original raw data ("none") will be returned.
as_matrix	logical TRUE if the function should return matrix, or a data frame or tibble otherwise.

Value

A data frame or matrix containing all data, including 'id_col' and 'split_col'.

Methods (by class)

• get_data(feature_data): Method for extracting all data.

---

Get ID Columns for all data

Description

Extracts ID columns (ID) from the preprocessed data.

Usage

get_id_all(object, ...)

## S4 method for signature 'feature_data'
get_id_all(object, as_matrix = TRUE)

Arguments

object	A 'feature_data' object.
...	Additional arguments (currently not used).
as_matrix	logical TRUE if the function should return matrix, or a data frame or tibble otherwise.

Value

A data frame or matrix containing the ID column(s).

Methods (by class)

• get_id_all(feature_data): Method extracting id columns for all data

---

Get ID Columns for test data

Description

Extracts ID columns (ID) from the preprocessed data.

Usage

get_id_test(object, ...)

## S4 method for signature 'feature_data'
get_id_test(object, as_matrix = TRUE)

Arguments

object	A 'feature_data' object.
...	Additional arguments (currently not used).
as_matrix	logical TRUE if the function should return matrix, or a data frame or tibble otherwise.

Value

A data frame or matrix containing the ID column(s).

Methods (by class)

• get_id_test(feature_data): Method for extracting id columns for the test data

---

Get ID Columns for training data

Description

Extracts ID columns (ID) from the preprocessed data.

Usage

get_id_train(object, ...)

## S4 method for signature 'feature_data'
get_id_train(object, as_matrix = TRUE)

Arguments

object	A 'feature_data' object.
...	Additional arguments (currently not used).
as_matrix	logical TRUE if the function should return matrix, or a data frame or tibble otherwise.

Value

A data frame or matrix containing the ID column(s).

Methods (by class)

• get_id_train(feature_data): Method for extracting id columns for the training data

---

Get All Data (X)

Description

Extracts all the data (X) from the preprocessed data.

Usage

get_x_all(object, ...)

## S4 method for signature 'feature_data'
get_x_all(
  object,
  prep = c("both", "scale", "transform", "none"),
  as_matrix = TRUE
)

Arguments

object	A 'feature_data' object.
...	Additional arguments (currently not used).
prep	Character argument if transformed data ("transform"), scaled data ("scale" ), transformed and scaled data ("both") or original raw data ("none") will be returned.
as_matrix	logical TRUE if the function should return matrix, or a data frame or tibble otherwise.

Value

A data frame or matrix containing all the data (X).

Methods (by class)

• get_x_all(feature_data): Method for extracting all data (X).

---

Get Test Data (X)

Description

Extracts the test data (X) from the preprocessed data.

Usage

get_x_test(object, ...)

## S4 method for signature 'feature_data'
get_x_test(
  object,
  prep = c("both", "scale", "transform", "none"),
  as_matrix = TRUE
)

Arguments

object	A 'feature_data' object.
...	Additional arguments (currently not used).
prep	Character argument if transformed data ("transform"), scaled data ("scale" ), transformed and scaled data ("both") or original raw data ("none") will be returned.
as_matrix	logical TRUE if the function should return matrix, or a data frame or tibble otherwise.

Value

A data frame or matrix containing the test data (X).

Methods (by class)

• get_x_test(feature_data): Method for extracting test data (X).

---

Get Training Data (X)

Description

Extracts the training data (X) from the preprocessed data.

Usage

get_x_train(object, ...)

## S4 method for signature 'feature_data'
get_x_train(
  object,
  prep = c("both", "scale", "transform", "none"),
  as_matrix = TRUE
)

Arguments

object	A 'feature_data' object.
...	Additional arguments (currently not used).
prep	Character argument if transformed data ("transform"), scaled data ("scale" ), transformed and scaled data ("both") or original raw data ("none") will be returned.
as_matrix	logical TRUE if the function should return matrix, or a data frame or tibble otherwise.

Value

A data frame or matrix containing the training data (X).

Methods (by class)

• get_x_train(feature_data): Method for extracting training data (X).

---

Get All Data (Y)

Description

Extracts all the data (Y) from the preprocessed data.

Usage

get_y_all(object, ...)

## S4 method for signature 'feature_data'
get_y_all(
  object,
  prep = c("both", "scale", "transform", "none"),
  as_matrix = TRUE
)

Arguments

object	A 'feature_data' object.
...	Additional arguments (currently not used).
prep	Character argument if transformed data ("transform"), scaled data ("scale" ), transformed and scaled data ("both") or original raw data ("none") will be returned.
as_matrix	logical TRUE if the function should return matrix, or a data frame or tibble otherwise.

Value

A data frame or matrix containing all the data (Y).

Methods (by class)

• get_y_all(feature_data): Method for extracting all data (Y).

---

Get Test Data (Y)

Description

Extracts the test data (Y) from the preprocessed data.

Usage

get_y_test(object, ...)

## S4 method for signature 'feature_data'
get_y_test(
  object,
  prep = c("both", "scale", "transform", "none"),
  as_matrix = TRUE
)

Arguments

object	A 'feature_data' object.
...	Additional arguments (currently not used).
prep	Character argument if transformed data ("transform"), scaled data ("scale" ), transformed and scaled data ("both") or original raw data ("none") will be returned.
as_matrix	logical TRUE if the function should return matrix, or a data frame or tibble otherwise.

Value

A data frame or matrix containing the test data (Y).

Methods (by class)

• get_y_test(feature_data): Method for extracting test data (Y).

---

Get Training Data (Y)

Description

Extracts the training data (Y) from the preprocessed data.

Usage

get_y_train(object, ...)

## S4 method for signature 'feature_data'
get_y_train(
  object,
  prep = c("both", "scale", "transform", "none"),
  as_matrix = TRUE
)

Arguments

object	A 'feature_data' object.
...	Additional arguments (currently not used).
prep	Character argument if transformed data ("transform"), scaled data ("scale" ), transformed and scaled data ("both") or original raw data ("none") will be returned.
as_matrix	logical TRUE if the function should return matrix, or a data frame or tibble otherwise.

Value

A data frame or matrix containing the training data (Y).

Methods (by class)

• get_y_train(feature_data): Method for extracting training data (Y).

---

Inverse Transform Data

Description

Applies the inverse transformation to the data.

Usage

inv_transform_data(object, params, funs, ...)

## S4 method for signature 'feature_data,ANY,ANY'
inv_transform_data(object, params, funs, ...)

## S4 method for signature 'feature_data,missing,list'
inv_transform_data(object, params, funs, ...)

## S4 method for signature 'data.frame,feature_params,missing'
inv_transform_data(object, params, funs, ...)

Arguments

object	A 'feature_data' object.
params	A 'feature_params' object with transformation and scaling parameters
funs	A list of inverse data transformation functions.
...	Additional arguments (currently not used).

Value

A 'feature_data' object or a 'data.frame' with the inverse transformed data.

Methods (by class)

• inv_transform_data(object = feature_data, params = ANY, funs = ANY): Method for inverse transforming data in a 'feature_data' object.

• inv_transform_data(object = feature_data, params = missing, funs = list): Method for transforming data in a 'feature_data' object and user-defined functions (deprecated).

• inv_transform_data( object = data.frame, params = feature_params, funs = missing ): Method for a 'data frame'.

---

Union Class of List or NULL

Description

Class to slots that can be empty

---

Evaluate Machine Learning Model Performance

Description

Calculates goodness-of-fit metrics (MSE, RMSE, MAE, R2) on preprocessed data, supporting fine-grained control over feature and target preprocessing states.

Usage

ml_evaluate(object, ...)

## S4 method for signature 'feature_data'
ml_evaluate(
  object,
  model,
  subset = c("all", "test", "train"),
  xprep = c("both", "scale", "transform", "none"),
  yprep = c("both", "scale", "transform", "none"),
  to_original_scale = FALSE,
  ...
)

Arguments

object	A 'feature_data' object.
...	Additional arguments passed to 'predict()'.
model	A fitted machine learning model (e.g., Keras, nnet, lm).
subset	Data subset to evaluate ('"all"', '"train"', or '"test"').
xprep	Preprocessing state for predictor features ('"both"', '"scale"', '"transform"', '"none"').
yprep	Preprocessing state for target variable ('"both"', '"scale"', '"transform"', '"none"').
to_original_scale	Logical; if 'TRUE', evaluates performance on original physical units (overrides 'yprep' to '"none"' for target comparison).

Details

Bias (Mean Error) is calculated as mean(observed - predicted). Positive values indicate systematic underprediction by the model.

Value

A named numeric vector of evaluation metrics (MSE, RMSE, MAE, R2).

Methods (by class)

• ml_evaluate(feature_data): Evaluate performance metrics for a 'feature_data' object.

---

Model Predictions with 'feature_data' Objects

Description

Makes predictions from a machine learning model with a 'feature_data' object.

Usage

## S4 method for signature 'feature_data'
predict(
  object,
  model,
  subset = c("all", "test", "train"),
  xprep = c("both", "scale", "transform", "none"),
  yprep = c("both", "scale", "transform", "none"),
  to_original_scale = FALSE,
  as_matrix = TRUE,
  ...
)

Arguments

object	A 'feature_data' object.
model	A fitted model object.
subset	Subset of the 'preproc' data set.
xprep	Character argument if transformed data ("transform"), scaled data ("scale" ), transformed and scaled data ("both") or original raw data ("none") are used for x.
yprep	Character argument if transformed data ("transform"), scaled data ("scale" ), transformed and scaled data ("both") or original raw data ("none") are used for y.
to_original_scale	Logical value to force re-transformation of predictions to the original scale.
as_matrix	logical TRUE if the function should return matrix, or a data frame or tibble otherwise.
...	Additional arguments (currently not used).

Details

Currently only certain Keras models can be used as fitted models.

Value

A matrix or vector with the predictions.

---

Extract Model Residuals Using 'feature_data' Objects

Description

Extract residuals from a machine learning model compared to a 'feature_data' object.

Usage

## S4 method for signature 'feature_data'
residuals(
  object,
  model,
  subset = c("all", "test", "train"),
  xprep = c("both", "scale", "transform", "none"),
  yprep = c("both", "scale", "transform", "none"),
  to_original_scale = FALSE,
  as_matrix = TRUE,
  ...
)

Arguments

object	A 'feature_data' object.
model	A fitted model object.
subset	Subset of the 'preproc' data set.
xprep	Character argument if transformed data ("transform"), scaled data ("scale" ), transformed and scaled data ("both") or original raw data ("none") are used for x.
yprep	Character argument if transformed data ("transform"), scaled data ("scale" ), transformed and scaled data ("both") or original raw data ("none") are used for y.
to_original_scale	Logical value to force re-transformation of predictions to the original scale.
as_matrix	logical TRUE if the function should return matrix, or a data frame or tibble otherwise.
...	Additional arguments (currently not used).

Details

Currently only certain Keras models can be used as fitted models.

Value

A matrix or vector with the residuals.

---

Nonlinear Coefficient of Determination Using 'feature_data' Objects

Description

Estimates the nonlinear R2 from a machine learning model compared to a 'feature_data' object.

Usage

rsquared(object, ...)

## S4 method for signature 'feature_data'
rsquared(
  object,
  model,
  subset = c("all", "test", "train"),
  prep = c("both", "scale", "transform", "none"),
  ...
)

Arguments

object	A 'feature_data' object.
...	Additional arguments passed down to 'predict()'.
model	A fitted model object.
subset	Subset of the 'preproc' dataset ('"all"', '"train"', or '"test"').
prep	Character argument if transformed data ("transform"), scaled data ("scale"), transformed and scaled data ("both"), or original raw data ("none") will be returned.

Value

The coefficient of determination as an un-named scalar numeric value.

Methods (by class)

• rsquared(feature_data): Method for estimating the coefficient of determination of a model for a 'feature_data' object.

---

Scale Data

Description

Scales the data using the stored parameters.

Usage

scale_data(object, params, ...)

## S4 method for signature 'feature_data,missing'
scale_data(object, params, ...)

## S4 method for signature 'data.frame,feature_params'
scale_data(object, params, transformed = FALSE, ...)

Arguments

object	A 'feature_data' object or a data frame.
params	n 'feature_params' object with scaling parameters
...	Additional arguments (currently not used).
transformed	'TRUE' to use parameters of transformed data.

Value

A 'feature_data' object with the scaled data.

Methods (by class)

• scale_data(object = feature_data, params = missing): Method for scaling data in a 'feature_data' object.

• scale_data(object = data.frame, params = feature_params): Method for scaling data in a 'feature_data' object.

---

Set Transformations

Description

Sets the transformation functions for a 'feature_data' object.

Usage

set_transformations(object, fun_transform = NULL, fun_inverse = NULL)

## S4 method for signature 'feature_data'
set_transformations(object, fun_transform = NULL, fun_inverse = NULL)

Arguments

object	A 'feature_data' object.
fun_transform	A named list of functions to apply to the data. T he names should correspond to the columns, and the values should be the functions.
fun_inverse	A named list of functions representing the inverse transformations. The structure should be the same as 'fun_transform'.

Value

A 'feature_data' object with the updated transformation functions.

Methods (by class)

• set_transformations(feature_data): Method for setting transformations in a 'feature_data' object.

---

Transform Data

Description

Transforms the data using the provided functions.

Usage

transform_data(object, params, funs, ...)

## S4 method for signature 'feature_data,ANY,ANY'
transform_data(object, params, funs, ...)

## S4 method for signature 'feature_data,missing,list'
transform_data(object, params, funs, ...)

## S4 method for signature 'data.frame,feature_params,missing'
transform_data(object, params, funs, ...)

## S4 method for signature 'data.frame,missing,list'
transform_data(object, params, funs, ...)

Arguments

object	A 'feature_data' object or a data frame.
params	A 'feature_params' object with transformation and scaling parameters
funs	A list of data transformation functions .
...	Additional arguments (currently not used).

Value

A 'feature_data' object or a 'data.frame' with the transformed data.

Methods (by class)

• transform_data(object = feature_data, params = ANY, funs = ANY): Method for transforming data in a 'feature_data' object.

• transform_data(object = feature_data, params = missing, funs = list): Method for transforming data in a 'feature_data' object and user-defined functions (deprecated).

• transform_data(object = data.frame, params = feature_params, funs = missing): Method for a 'data frame'.

• transform_data(object = data.frame, params = missing, funs = list): Method for a 'data frame'.

---