DataVisualizations (version 1.4.0)

MDstrips: High-dimensional Density Strips based on Pareto Density Estimation

Description

MDstrips visualizes the distribution of each variable in a dataset as a vertical strip of colored tiles. Density is estimated by Pareto Density Estimation (PDE). Low-density regions are shown in blue/green, mid-density regions in yellow, and high-density regions in orange/red. This is a strip-based alternative to MDplot for the high-dimensional case of d>100 in which violins are easy visible due to restricted widht of screen.

Usage

MDstrips(Data, Ordering = "Default", Scaling = "None",
         QuantityThreshold = 50, UniqueValuesThreshold = 12, SampleSize = 5e+05,
         LabelThreshold = 100, LabelMax = 40, LabelEvery = NULL, LabelVariables = NULL,
         SizeOfJitteredPoints=1,
         palette = c("blue","green","yellow","orange","red"), ylab, main, BW = TRUE)

Value

Returns a ggplot object visualizing the density strips.

Arguments

Data

Numeric matrix containing the data. Each column is one variable.

Ordering

Optional:x String specifying ordering of variables on the axis. Options are "Default", "Columnwise", "AsIs", "Alphabetical", "Average", "Bimodal", "Variance", "Statistics".

Scaling

Optional, Data scaling method. Options: "None", "Percentalize", "CompleteRobust", "Robust", "Log".

QuantityThreshold

Optional: Minimum number of finite values required to estimate a distribution.

UniqueValuesThreshold

Optional: Minimum number of unique values required to estimate a distribution.

SampleSize

Optional: Maximum sample size. Larger datasets are subsampled for faster computation.

LabelThreshold

Optional: If number of variables exceeds this threshold, not all x-axis labels are shown.

LabelMax

Optional: Maximum number of x-axis labels shown when many variables are present.

LabelEvery

Optional: Integer. Show every \(k\)th variable label (overrides LabelMax).

LabelVariables

Optional: Character vector of variables to label explicitly (overrides other options).

SizeOfJitteredPoints

Optional: scalar. If not enough unique values for density estimation are given, data points are jittered. This parameter defines the size of the points.

palette

Optional: Vector of colors for density scale, from low to high density.

ylab

Optional: Label for the y-axis (range of values in which PDE is estimated).

main

Optional: Title of the plot.

BW

Optional: Logical. If TRUE, use black-and-white theme and remove grid lines.

Author

Michael Thrun

Details

This function is intended for high-dimensional but univariate distribution visualization where MDplot, violin plots or boxplots become unreadable. By using density-colored strips, a large number of variables can be compared side-by-side.

Pareto Density Estimation (PDE) is used for robustness of univariate density estimation.

If too few values are present for PDE estimation (less than QuantityThreshold or less than UniqueValuesThreshold unique values), jittered points are drawn instead.

See Also

MDplot for Mirrored-density plot and the explaination of Ordering and Scaling in Detail.

Examples

# Example with toy data
set.seed(123)
X <- matrix(rnorm(1000), ncol = 10)
colnames(X) <- paste0("V", 1:10)

# Default plot
MDstrips(X)

# With statistical ordering and black-and-white theme
MDstrips(X, Ordering = "Statistics", BW = TRUE)

# Show only every 2nd variable label
MDstrips(X, LabelEvery = 2)

# Show only specific variables
MDstrips(X, LabelVariables = c("V1","V5","V10"))