<< Statistical Visualization Statistical Visualization nan_boxplot >>

NaN Toolbox >> NaN Toolbox > Statistical Visualization > nan_andrewsplot

nan_andrewsplot

Andrews plot for multivariate data.

Calling Sequence

nan_andrewsplot(X)
nan_andrewsplot(X,groups)
nan_andrewsplot(X,groups,quant)
nan_andrewsplot(X,groups,quant,stdize)

Parameters

X :

Rows of X correspond to observations, columns to variables.

groups:

plots the data in different groups with different colors, must be a numeric array or character matrix,

quant:

if it is not empty it plots only the median and the quant and (1-quant) quantiles of f(t) at each value of t. This is useful if X contains many observations.

stdize = 'off' :

do nothing

stdize ='on' :

normalize each column of X (mean = 0, deviation = 1)

stdize = 'pca' :

instead of X the principal component scores of X is used, in order of decreasing eigenvalue.

stdize = 'pcastd' :

uses the standardized principal component scores.

Description

An Andrews plot is a graphical representation of the observed values ​​of several metric scaled variables X_j (j = 1, ..., p) in a two-dimensional coordinate system in which each multivariate data point is represented by a function.

f(t) is defined for the j-th observation as

f(t) = X(j,1)/sqrt(2) + X(j,2)*sin(2*pi*t) + X(j,3)*cos(2*pi*t) + ...

Elements with similar values ​​will have similar curves. This is possible because the Andrews plot preserves the distances of data points and the mean values ​​of the variables. The Andrews plot is suitable for the detection of outliers and for finding clusters.

nan_andrewsplot treats NaNs in X as missing values, and remove the corresponding rows.

Examples

// make a grouped plot of the raw data
loadmatfile("demos/data/iris.mat");
nan_andrewsplot(meas,species);

// plot only the median and quartiles of each group
nan_andrewsplot(meas,species,.25);

Bibliography

Andrews, D. F. (1972), Plots of high-dimensional data, Biometrics 28, S.125-136

See also


<< Statistical Visualization Statistical Visualization nan_boxplot >>