Add contours of estimated point density to a ternary plot

Source: R/Contours.R

Use two-dimensional kernel density estimation to plot contours of point density.

TernaryDensityContour(
  coordinates,
  bandwidth,
  resolution = 25L,
  tolerance = -0.2/resolution,
  edgeCorrection = TRUE,
  direction = getOption("ternDirection", 1L),
  ...
)

Arguments

coordinates

A list, matrix, data.frame or vector in which each element (or row) specifies the three coordinates of a point in ternary space.

bandwidth

Vector of bandwidths for x and y directions. Defaults to normal reference bandwidth (see MASS::bandwidth.nrd). A scalar value will be taken to apply to both directions.

resolution

The number of triangles whose base should lie on the longest axis of the triangle. Higher numbers will result in smaller subdivisions and smoother colour gradients, but at a computational cost.

tolerance

Numeric specifying how close to the margins the contours should be plotted, as a fraction of the size of the triangle. Negative values will cause contour lines to extend beyond the margins of the plot.

edgeCorrection

Logical specifying whether to correct for edge effects (see details).

direction

(optional) Integer specifying the direction that the current ternary plot should point: 1, up; 2, right; 3, down; 4, left.

...

Further parameters to pass to `contour.

Details

This function is modelled on MASS::kde2d(), which uses "an axis-aligned bivariate normal kernel, evaluated on a square grid".

This is to say, values are calculated on a square grid, and contours fitted between these points. This produces a couple of artefacts. Firstly, contours may not extend beyond the outermost point within the diagram, which may fall some distance from the margin of the plot if a low resolution is used. Setting a negative tolerance parameter allows these contours to extend closer to (or beyond) the margin of the plot.

Individual points cannot fall outside the margins of the ternary diagram, but their associated kernels can. In order to sample regions of the kernels that have 'bled' outside the ternary diagram, each point's value is calculated by summing the point density at that point and at equivalent points outside the ternary diagram, 'reflected' across the margin of the plot (see function ReflectedEquivalents). This correction can be disabled by setting the edgeCorrection parameter to FALSE.

A model based on a triangular grid may be more appropriate in certain situations, but is non-trivial to implement; if this distinction is important to you, please let the maintainers known by opening a Github issue.

See also

Other contour plotting functions: ColourTernary(), TernaryContour(), TernaryPointValues()

Author

Adapted from MASS::kde2d() by Martin R. Smith

Examples


TernaryPlot(axis.labels = seq(0, 10, by = 1))

nPoints <- 400L
coordinates <- cbind(abs(rnorm(nPoints, 2, 3)),
                     abs(rnorm(nPoints, 1, 1.5)),
                     abs(rnorm(nPoints, 1, 0.5)))

ColourTernary(TernaryDensity(coordinates, resolution = 10L))
TernaryPoints(coordinates, col = 'red', pch = '.')
TernaryDensityContour(coordinates, resolution = 30L)