Package 'ImpactEffectsize'

ImpactEffectsize-package

Description

Calculation and visualization of the Impact effect size measure. A non-parametric effect size measure capturing changes in central tendency or shape of data distributions for feature selection preceding machine-learning. The package provides the necessary functions to calculate and plot the Impact effect size measure between two groups.

References

Lotsch, J., and Ultsch, A. (2019): Impact – An R Package for calculation and visualisation of the Impact distance and data distribution-shape based effect size measure.

---

Example data of bimodal CD79 expression.

Description

Data set of CD79 expression in patients with B lymphoma (class 1) and in controls (class 2).

Usage

data("BcellLymphomaCD79")

Details

Size 258429 x 2 , Dimensions 1, stored in BcellLymphomaCD79$Data

Classes 2, stored in BcellLymphomaCD79$Classes

Examples

data(BcellLymphomaCD79)
str(BcellLymphomaCD79)

---

Example data with two groups and the Impact effet size measure.

Description

Dataset with 2 classes and 20 variables that allow class separation at varous degrees of difficulty.

Usage

data("FeatureselectionData")

Details

Size 2000 x 20 , Dimensions 1, stored in FeatureselectionData$Var0001,...,FeatureselectionData$Var0020

Classes 2, stored in FeatureselectionData$Classes

Examples

data(FeatureselectionData)
str(FeatureselectionData)

---

Example data of hematologic marker expression.

Description

Data set of 8 flow cytometry-based lymphoma makers from 1,494 cells from healthy subjects (class 1) and 1,302 cells from lymphoma patients (class 2).

Usage

data("FlowcytometricData")

Details

Size 2796 x 9 , Dimensions 1, stored in FlowcytometricData$$[CD3,CD4,CD8,CD11,CD19,CD103,CD200,IgM]

Classes 2, stored in FlowcytometricData$Classes

Examples

data(FlowcytometricData)
str(FlowcytometricData)

---

Impact effect size measure

Description

Calculates the Impact effect size measure that is based on the group distance and the difference in the shape of the data distribution between two groups.

Usage

Impact(Data, Cls, PlotIt = FALSE, pde = TRUE, 
  col = c("red","blue"), meanLines = FALSE, medianLines = FALSE, ...)

Arguments

Data	Numeric vector containing the values of both groups.
Cls	Grouping vector or factor of same length as Data, containing exactly two distinct classes.
PlotIt	Logical; if TRUE, plots the probability density function (PDF) of the two groups using Pareto density estimation or kernel density as fallback.
pde	Logical; if TRUE, attempts Pareto Density Estimation (PDE) for the PDFs in the plot. Ignored if PlotIt = FALSE.
col	Character vector of length two specifying colors for the two groups in the plot. Ignored if PlotIt = FALSE.
meanLines	Logical; if TRUE, draws vertical lines at group means in the plot. Ignored if PlotIt = FALSE.
medianLines	Logical; if TRUE, draws vertical lines at group medians in the plot. Ignored if PlotIt = FALSE.
...	Further graphical parameters passed to the plotting function if PlotIt = TRUE.

Details

The Impact effect size measure combines central tendency differences (based on group medians) and morphological differences (based on Pareto density estimation). If Pareto density estimation fails or is disabled, the density plots fallback to standard kernel density estimates. The function can optionally plot these densities along with mean and/or median reference lines.

Value

Returns a list with the following components:

Impact

Numeric scalar; the combined effect size measure based on difference in medians and distribution shapes.

MorphDiff

Numeric scalar; the extent of difference in shapes of the probability density functions.

CTDiff

Numeric scalar; the extent of difference in group medians.

density_df

Data frame with columns PDEKernels, pde_Cls1, pde_Cls2 containing the density kernel points and Pareto densities for each group. This may be empty if Pareto density estimation was not successful or disabled.

Author(s)

Jorn Lotsch and Alfred Ultsch

References

Lotsch, J., and Ultsch, A. (2019): ImpactEffectsize – an R Package for calculation and visualisation of the Impact distance and shape based effect size measure.

Examples

## Example 1: Use Impact with plotting
data("FeatureselectionData")
ImpactSize <- Impact(Data = FeatureselectionData$Var0011, 
  Cls = FeatureselectionData$Classes, PlotIt = TRUE)
  
## Example 2: Impact without plotting
ImpactSize <- Impact(Data = FeatureselectionData$Var0011, 
  Cls = FeatureselectionData$Classes, PlotIt = FALSE)

## example 2
data("BcellLymphomaCD79")
data("FeatureselectionData")
data("FlowcytometricData")
data("SameMeansData")
data("StocksFluctuation")

---

Example artificial data with two groups of same means but different data distribution shapes.

Description

Dataset with 2 classes six variables were both classes have the same means but different shapes of the distribution.

Usage

data("SameMeansData")

Details

Size 2000 x 7 , Dimensions 1, stored in SameMeansData$NOchangeInMandS,...,SameMeansData$NegChi2andGauss

Classes 2, stored in SameMeansData$Classes

Examples

data(SameMeansData)
str(SameMeansData)

---

Example data of stock fluctuation.

Description

Data set of Log ratios of daily changes of n =5,522 for 10 German stocks with low fluctuation (class 1) or high fluctuation (class 2).

Usage

data("StocksFluctuation")

Details

Size 5522 x 2 , Dimensions 1, stored in StocksFluctuation$logFluctuation

Classes 2, stored in StocksFluctuation$Classes

Examples

data(StocksFluctuation)
str(StocksFluctuation)

---