Skip to contents

persistence_landscape() computes the persistence landscape (Bubenik 2015) of one homology dimension of a persistence diagram, evaluated on a grid. Each (birth, death) pair contributes a tent function $$\Lambda_{(b,d)}(t) = \max(0, \min(t - b, d - t)),$$ and the \(k\)-th landscape function \(\lambda^{(k)}(t)\) is the \(k\)-th largest of \(\{\Lambda_{(b_i,d_i)}(t)\}_i\) at each \(t\). It takes a persistent_homology() result or a data frame of persistence pairs.

At each filtration value \(t\), a tent is the distance from \(t\) to the nearer end of a feature's interval (0 outside it), so the first landscape function follows the feature that \(t\) lies deepest inside, the second the feature with the second largest tent, and so on. This need not be the most persistent feature alive: a long-lived feature near its birth or death can sit below a shorter one near its midpoint. Landscapes are stable: the sup-norm distance between two landscapes is at most the bottleneck_distance() between their diagrams. They also live in a Banach space, where they can be averaged (Bubenik 2015).

Essential classes are left out, because the tent of an infinitely lived class is undefined on a finite grid: pairs with death = Inf (VR mode) and pairs with death = 0 but birth > 0 (the clique-mode encoding of an essential class) are dropped before the landscape is built. When no finite pair remains in the requested dimension, every landscape function is zero on the grid.

Usage

persistence_landscape(ph, k_max = 5L, dimension = 1L, t_grid = NULL)

# S3 method for class 'persistence_landscape'
print(x, ...)

# S3 method for class 'persistence_landscape'
plot(x, ...)

Arguments

ph

The diagram: a persistent_homology() result, or a data frame with the columns dimension, birth and death.

k_max

The number of landscape functions to compute, a positive whole number. Functions beyond the number of pairs are zero. Default 5.

dimension

The homology dimension, a single non-negative whole number. Default 1.

t_grid

The filtration values to evaluate the landscape at. Default NULL: 200 evenly spaced points from the smallest birth to the largest death of the remaining pairs (each pair is ordered so that birth comes first, since clique filtrations descend), or from 0 to 1 when no pair remains. Pass the same grid to landscapes you want to compare.

x

For the print() and plot() methods: a persistence_landscape object.

...

Not used.

Value

A persistence_landscape object. Printing it shows the dimension, k_max, the grid range and the sup norm (largest value) of each landscape function; plot() draws the functions as lines over the grid. The object holds:

landscape

A data frame with one row per landscape function and grid point, and the columns k, t and value.

dimension

The homology dimension.

k_max

The number of landscape functions.

t_grid

The evaluation grid.

The print() method returns its input invisibly, and plot() a ggplot object with one line per landscape function.

References

Bubenik, P. (2015). Statistical topological data analysis using persistence landscapes. Journal of Machine Learning Research 16, 77-102.

Examples

mat <- matrix(c(0, .6, .5, .6, 0, .4, .5, .4, 0), 3, 3)
rownames(mat) <- colnames(mat) <- c("A","B","C")
ph <- persistent_homology(mat, n_steps = 5)
pl <- persistence_landscape(ph, k_max = 3, dimension = 0)