Package {SimplicialComplex}

Construct an Alpha Complex

Description

Construct an Alpha Complex

Usage

AlphaComplex(points, epsilon = Inf)

Arguments

Value

A list of class "alpha_complex" with elements:

simplices

A list of integer vectors, every included simplex (all dimensions, not just the maximal ones - membership in the alpha complex is not simply generated by cliques).

filtration

A numeric vector, the alpha value of each simplex in simplices, in the same order.

Pass the result to as_filtration for the same list(simplex=, t=) format used by build_filtration and every persistence function, or just call build_filtration(points, method = "Alpha", eps_max = epsilon).

Examples

points <- matrix(c(0, 1, 1, 0, 0, 0, 1, 1), ncol = 2)
alpha_complex <- AlphaComplex(points, epsilon = 1)

Construct a Cech Complex

Description

Construct a Cech Complex

Usage

CechComplex(points, epsilon)

Arguments

Value

A list with:

network

An igraph object: the 1-skeleton of the candidate graph (edges where distance \le 2\epsilon).

simplices

A list of integer vectors, each the vertex indices of a maximal clique of network.

Examples

points <- matrix(c(0, 1, 1, 0, 0, 0, 1, 1), ncol = 2)
cech_complex <- CechComplex(points, epsilon = 0.8)

Construct a Delaunay Complex

Description

Construct a Delaunay Complex

Usage

DelaunayComplex(points)

Arguments

Value

A list of class c("delaunay_complex", "alpha_complex"); see AlphaComplex for the element description.

Examples

points <- matrix(c(0, 1, 1, 0, 0, 0, 1, 1), ncol = 2)
delaunay_complex <- DelaunayComplex(points)

Construct a Vietoris–Rips Complex (1-skeleton + maximal simplices)

Description

Construct a Vietoris–Rips Complex (1-skeleton + maximal simplices)

Usage

VietorisRipsComplex(points, epsilon)

Arguments

Details

The Vietoris–Rips complex at scale \epsilon includes a simplex for every finite set of points with pairwise distances < \epsilon. This function constructs the 1-skeleton (edges only) and then uses maximal cliques in that graph as the maximal simplices.

Value

A list with:

network

An igraph object representing the 1-skeleton.

simplices

A list of integer vectors, each the vertex indices of a maximal simplex.

Examples

points <- matrix(c(0, 1, 1, 0, 0, 0, 1, 1), ncol = 2)
epsilon <- 1.5
vr_complex <- VietorisRipsComplex(points, epsilon)

Construct a Witness Complex

Description

Construct a Witness Complex

Usage

WitnessComplex(points, landmarks, epsilon, nu = 1)

Arguments

Value

A list with:

network

An igraph object: the 1-skeleton on the landmark vertex set.

simplices

A list of integer vectors (indices into landmarks / landmark_indices), each the vertices of a maximal simplex.

landmarks

The landmark coordinate matrix.

landmark_indices

Row indices into points, or NULL if landmarks was given as coordinates rather than indices.

edge_birth

A symmetric matrix giving, for every pair of landmarks, the exact scale at which their edge is witnessed (used by build_filtration to time every simplex exactly, rather than only checking membership at the single scale epsilon).

Examples

points <- matrix(rnorm(200), ncol = 2)
witness_complex <- WitnessComplex(points, landmarks = 15, epsilon = 0.5)

Convert a flood_complex object to a filtration list

Description

Convert a flood_complex object to a filtration list

Usage

as_filtration(fc)

Arguments

Value

A filtration list compatible with boundary_info.

Build the augmented (n+m) x (n+m) assignment cost matrix for two diagrams

Description

Build the augmented (n+m) x (n+m) assignment cost matrix for two diagrams

Usage

augmented_cost_matrix(X, Y, ground = c("L2", "Linf"), power = 1)

Arguments

Value

A square numeric matrix of size nrow(X) + nrow(Y).

Safely compute the rank of a sparse matrix

Description

This helper function wraps Matrix::rankMatrix() to safely handle empty matrices (i.e., with 0 rows or columns).

Usage

betti_number(simplices, bound_dim, tol = NULL)

Arguments

Value

An integer representing the rank of the matrix.

Examples

simplices <- list(c(1, 2), c(3, 4), c(2, 1, 3), c(4, 2))
betti_number(simplices, 0, tol=0.1)

Bottleneck distance between two persistence diagrams

Description

The bottleneck distance between the points of two persistence diagrams in a given homological dimension, allowing points to be matched to the diagonal: the infimum, over all matchings, of the largest single point-to-point distance (Cohen-Steiner, Edelsbrunner and Harer (2007), "Stability of Persistence Diagrams") - the p \to \infty limit of wasserstein_distance. Essential (death = Inf) classes are kept rather than dropped - see wasserstein_distance's Details for why and how. Computed exactly by binary search over the (finitely many) candidate distance values for the smallest one admitting a perfect matching, checked with a standard augmenting-path bipartite matcher, on the same augmented assignment problem wasserstein_distance uses (see augmented_cost_matrix).

Usage

bottleneck_distance(df1, df2, dimension = 0, ground = c("Linf", "L2"))

Arguments

Value

A single non-negative number, the bottleneck distance.

Examples

df1 <- data.frame(dim = c(0, 0), birth = c(0, 0), death = c(1, 2))
df2 <- data.frame(dim = c(0, 0), birth = c(0, 0), death = c(1, 3))
bottleneck_distance(df1, df2, dimension = 0)

Compute the boundary operator for a simplicial complex

Description

Compute the boundary operator for a simplicial complex

Usage

boundary(simplices, bound_dim)

Arguments

Details

\partial_k \sigma = \sum_i (-1)^i [v_0 v_1 \ldots \hat{v}_i \ldots v_k]

Value

A sparse matrix representing \partial_{k}.

Examples

simplices <- list(c(1, 2), c(3, 4), c(2, 1, 3), c(4, 2))
boundary(simplices, 0)

Get the boundary matrix and its reduction information in matrix form

Description

Get the boundary matrix and its reduction information in matrix form

Usage

boundary_info(filist, max_dimension = NULL)

Arguments

Value

A list containing the boundary matrix, the last boundary row, the pivot owner for persistence extraction, and filist - the exact (possibly max_dimension-restricted) filtration list the other three elements were computed from, re-tagged with the same "max_dimension" attribute so extract_persistence_pairs can auto-detect it too. Always pass THIS filist back into extract_persistence_pairs, not your original one - see its Details for why.

Examples

points <- matrix(c(0, 1, 1, 0, 0, 0, 1, 1), ncol = 2)
filtration <- build_filtration(points, method = "VR", eps_max = 1.2)
res <- boundary_info(filtration)

Build the clique (flag) complex of a graph on a fixed vertex set

Description

Shared by VietorisRipsComplex, CechComplex and WitnessComplex: each of them reduces to "connect vertices that are close enough (in whatever sense that complex uses), then take the maximal cliques of that graph as the maximal simplices."

Usage

build_clique_complex(n, edges)

Arguments

Value

A list with network (an igraph object, the 1-skeleton) and simplices (a list of integer vectors, the vertex sets of the maximal cliques, each sorted).

Build a cubical (grid) filtration from an image

Description

Build a cubical (grid) filtration from an image

Usage

build_cubical_filtration(image, superlevel = FALSE)

Arguments

Value

A filtration list: one list(simplex = <integer vector of pixel ids>, t = <numeric>) per simplex, sorted by (t, dimension, lexicographic order) - the same format as build_filtration.

Examples

# a ring (value 1) around a hole (value 5) around a background (value 9):
# the hole is born once the ring closes and dies once its center fills in
image <- matrix(c(
  9, 9, 9, 9, 9,
  9, 1, 1, 1, 9,
  9, 1, 5, 1, 9,
  9, 1, 1, 1, 9,
  9, 9, 9, 9, 9
), nrow = 5, byrow = TRUE)
filtration <- build_cubical_filtration(image)
pairs <- persistence_pairs(filtration)
pairs[pairs$dim == 1, ] # one H1 bar: birth = 1 (ring closes), death = 5
plot_persistence(pairs)

Build a filtration from a point cloud, for any of five complex types

Description

Build a filtration from a point cloud, for any of five complex types

Usage

build_filtration(
  points,
  method = c("VR", "Delaunay", "Alpha", "Cech", "Witness"),
  eps_max = NULL,
  landmarks = NULL,
  nu = 1,
  max_dimension = NULL
)

Arguments

Value

A filtration list, as described above. When max_dimension is set, the list carries it as a "max_dimension" attribute (see above); otherwise the attribute is absent, exactly as before this parameter existed.

Examples

points <- matrix(c(0, 1, 1, 0, 0, 0, 1, 1), ncol = 2)
vr_filtration <- build_filtration(points, method = "VR", eps_max = 1.2)
alpha_filtration <- build_filtration(points, method = "Alpha")
delaunay_filtration <- build_filtration(points, method = "Delaunay")
cech_filtration <- build_filtration(points, method = "Cech", eps_max = 0.8)
# cap at H1: persistence_pairs()/etc. need no max_dimension of their own
vr_capped <- build_filtration(points, method = "VR", eps_max = 1.2,
                               max_dimension = 1)
pairs <- persistence_pairs(vr_capped)

Flood filtration in SimplicialComplex format

Description

Wraps flood_complex and returns the filtration in the same format as build_filtration: a list of list(simplex = <integer vector>, t = <numeric>), sorted by (time, dimension, lexicographic order). The result plugs directly into boundary_info() / extract_persistence_pairs() / plot_persistence(), as well as into the faster flood_persistence.

Usage

build_flood_filtration(points, landmarks, ...)

Arguments

Value

A filtration list compatible with boundary_info.

Examples

## Not run: 
pts <- matrix(rnorm(2000), ncol = 2)
filtration <- build_flood_filtration(pts, landmarks = 30)
pairs <- flood_persistence(filtration)

## End(Not run)

Circumsphere of an affinely independent point set

Description

The unique sphere through points whose center lies in their affine hull - i.e. the minimal-radius sphere with all of points on its boundary. Used by AlphaComplex/DelaunayComplex to compute alpha values (unlike min_enclosing_ball, which minimizes radius over ALL enclosing balls, this fixes every point to lie exactly on the sphere - the two coincide only for simplices whose circumcenter is not "hidden" inside the simplex in a way that admits a smaller covering ball).

Usage

circumsphere(points)

Arguments

Value

A list with center (numeric vector) and radius.

Compare simplicial complex constructions on one point cloud

Description

Builds the same point cloud's filtration under several build_filtration methods and summarizes, side by side, how expensive each one was to build and what persistent homology it found - useful for seeing directly how Cech's dimension blow-up, Alpha's Delaunay-bounded dimension, VR's cheap-but-approximate cliques, and Witness's landmark subsampling trade off against each other on the same data (see Otter et al. (2017), Section 5.2, for the underlying trade-offs).

Usage

compare_complexes(
  points,
  methods = c("VR", "Delaunay", "Alpha", "Cech", "Witness"),
  eps_max = NULL,
  landmarks = NULL,
  nu = 1
)

Arguments

Value

A list with:

summary

A data frame, one row per method, with the number of vertices and simplices, the highest simplex dimension reached, build and persistence-computation time in seconds, and the number of finite/essential persistence pairs found.

diagrams

A named list of persistence diagram data frames (one per method, from persistence_pairs), for further comparison (e.g. with wasserstein_distance/ bottleneck_distance, or plot_persistence).

filtrations

A named list of the raw filtration lists.

Examples

points <- matrix(c(0, 1, 1, 0, 0, 0, 1, 1, 0.5, 0.5), ncol = 2, byrow = TRUE)
cmp <- compare_complexes(points, methods = c("VR", "Alpha", "Delaunay"), eps_max = 1.2)
cmp$summary

Wasserstein/bottleneck distance between two point clouds, via one chosen complex

Description

Picks a single build_filtration method, builds the persistence diagram of each point cloud with it, and compares the two diagrams with either wasserstein_distance or bottleneck_distance. Useful for asking "how different are these two datasets topologically", holding the complex construction fixed so the comparison is apples-to-apples.

Usage

complex_distance(
  points1,
  points2,
  method = c("VR", "Delaunay", "Alpha", "Cech", "Witness"),
  distance = c("wasserstein", "bottleneck"),
  dimension = 0,
  p = 2,
  ground = NULL,
  eps_max = NULL,
  landmarks = NULL,
  nu = 1
)

Arguments

Value

A list with:

distance

The computed distance (a single number).

method, distance_type, dimension

Echoed back for reference.

diagram1, diagram2

The two persistence diagrams (data frames) the distance was computed from.

Examples

set.seed(1)
cloud_a <- matrix(rnorm(24), ncol = 2)
cloud_b <- matrix(rnorm(24), ncol = 2) + 0.1
complex_distance(cloud_a, cloud_b, method = "VR", distance = "bottleneck",
                  dimension = 0, eps_max = 0.6)

Optimal point matching between two persistence diagrams

Description

Computes the same optimal matching that wasserstein_distance/bottleneck_distance reduce to internally, and returns it decoded into a readable table instead of just the distance - what plot_matching draws. Every point of both diagrams appears in exactly one row: matched to a point of the other diagram (type = "real-real"), or matched to the diagonal, i.e. effectively unmatched (type = "x-diagonal" for a point of df1, "y-diagonal" for a point of df2).

Usage

diagram_matching(
  df1,
  df2,
  dimension,
  distance = c("wasserstein", "bottleneck"),
  p = 2,
  ground = NULL
)

Arguments

Value

A list with distance (the matching's cost, equal to what wasserstein_distance/bottleneck_distance would return), matches (a data frame with columns x_birth, x_death, x_essential, y_birth, y_death, y_essential, type; a NA pair on one side means that row's point matched the diagonal), and X, Y, essential_X, essential_Y (the two diagrams' points after the same essential-pair capping wasserstein_distance uses - see its Details - and which of them were essential before capping).

Examples

df1 <- data.frame(dim = c(0, 0), birth = c(0, 0), death = c(1, 2))
df2 <- data.frame(dim = c(0, 0), birth = c(0, 0), death = c(1, 3))
diagram_matching(df1, df2, dimension = 0, distance = "wasserstein", p = 2)

Compute the Euler characteristic \chi of a simplicial complex

Description

Compute the Euler characteristic \chi of a simplicial complex

Usage

euler_characteristic(simplices, tol)

Arguments

Details

The Euler characteristic is computed as:

\chi = \sum_{k=0}^{k_{\max}} (-1)^k \beta_k

where \beta_k is the kth Betti number, and k_{\max} is the highest dimension of any simplex in the complex.

Interpretation of values:

• \chi = 2: Sphere-like surfaces

• \chi = 1: Disk-like spaces

• \chi = 0: Torus-like or circle-like spaces

• \chi < 0: Surfaces with multiple handles or genus

Value

An integer representing the Euler characteristic \chi.

See Also

betti_number

Examples

simplices <- list(c(1, 2), c(3, 4), c(2, 1, 3), c(4, 2))
euler_characteristic(simplices, tol=0.1)

This function extracts the persistence from combining the boundary matrix and its filtration

Description

This function extracts the persistence from combining the boundary matrix and its filtration

Usage

extract_persistence_pairs(filist, last_1, pivot_owner, max_dimension = NULL)

Arguments

Details

boundary_info() and extract_persistence_pairs() are two halves of one computation - last_1/pivot_owner only mean anything relative to the exact filist boundary_info() used internally. Whenever max_dimension is involved, always call as:

res <- boundary_info(filtration, max_dimension = k)
pairs <- extract_persistence_pairs(res$filist, res$last_1, res$pivot_owner,
                                    max_dimension = k)

A length mismatch between filist and last_1/pivot_owner is refused with an error rather than silently producing wrong pairs - see persistence_pairs for a one-call alternative that cannot run into this.

Value

A data frame with columns: dimension, birth, and death.

Examples

points <- matrix(c(0, 1, 1, 0, 0, 0, 1, 1), ncol = 2)
filtration <- build_filtration(points, method = "VR", eps_max = 1.2)
res <- boundary_info(filtration)
pairs <- extract_persistence_pairs(filtration, res$last_1, res$pivot_owner)

Generate all unique faces of a given dimension from simplices

Description

Generate all unique faces of a given dimension from simplices

Usage

faces(simplices, target_dim)

Arguments

Details

The function generates all possible subsets (combinations) of each simplex, removes duplicates, and filters them to only include those of length target_dim + 1.

For example, a 2-simplex c(1, 2, 3) has three 1-dimensional faces (edges): c(1,2), c(1,3), and c(2,3), and three 0-dimensional faces (vertices): 1, 2, and 3.

Value

A list of faces (each a numeric vector) of dimension target_dim.

Examples

simplices <- list(c(1, 2), c(3, 4), c(2, 1, 3), c(4, 2))
faces(simplices, target_dim=0)

Construct a Flood complex

Description

Builds the Flood complex of a point cloud: a Delaunay complex on a set of landmarks whose simplices are filtered by their covering radius with respect to the full point cloud ("flood time").

Usage

flood_complex(
  points,
  landmarks,
  max_dimension = NULL,
  points_per_edge = 30,
  backend = c("auto", "cpu", "torch"),
  batch_points = 2^22,
  delaunay = NULL
)

Arguments

Value

A list of class "flood_complex" with elements simplices (list of integer vectors, landmark indices), filtration (numeric vector of flood times), landmarks (matrix), landmark_indices (or NULL).

Examples

## Not run: 
pts <- matrix(rnorm(3000), ncol = 2)
fc <- flood_complex(pts, landmarks = 40)

## End(Not run)

Persistence pairs via sparse boundary reduction

Description

Computes persistence pairs from a filtration list with the standard column-reduction algorithm, but on sparse columns (integer index vectors over GF(2)) instead of a dense matrix. Produces the same output format as extract_persistence_pairs(filist, res$last_1, res$pivot_owner) while scaling to the much larger complexes produced by build_flood_filtration.

Usage

flood_persistence(filist, max_dimension = NULL)

Arguments

Value

A data frame with columns dim, birth, death.

Farthest-Point Sampling of landmarks

Description

Selects n_lms landmarks from a point cloud via (exact) Farthest-Point Sampling. Equivalent to flooder::generate_landmarks (which uses an approximate bucket-FPS; the exact version below gives the same qualitative coverage).

Usage

generate_landmarks(points, n_lms, start_idx = 1)

Arguments

Value

A list with landmarks (n_lms x d matrix) and indices (row indices into points).

Examples

## Not run: 
pts <- matrix(rnorm(2000), ncol = 2)
lms <- generate_landmarks(pts, 50)

## End(Not run)

Minimum enclosing ball of a finite point set (Welzl's algorithm)

Description

Used by CechComplex: the Cech complex includes a simplex \sigma at scale \epsilon exactly when the balls of radius \epsilon centered at its vertices have a common point, which happens if and only if the minimum enclosing ball of \sigma's vertices has radius at most \epsilon (this is the standard reduction used e.g. by GUDHI's Cech complex; see Cavanna, Jahanseir and Sheehy (2017)).

Usage

min_enclosing_ball(points)

Arguments

Details

Randomized, expected linear-time in the number of points; the point sets passed in here are the vertices of a single simplex, so they are always small.

Value

A list with center (numeric vector) and radius.

Pairwise Euclidean distance matrix

Description

Pairwise Euclidean distance matrix

Usage

pairwise_dist(points, query = NULL)

Arguments

Value

A numeric distance matrix.

Compute the persistence landscape of a persistence diagram

Description

Converts a persistence diagram data frame (the direct output of extract_persistence_pairs(), persistence_pairs(), or flood_persistence()) into its persistence landscape: a collection of continuous, piecewise-linear functions \lambda(k, \cdot), obtained by overlaying the "tent" function of every birth-death pair and, at each time t, taking the kth largest tent value (Bubenik, 2015; Chazal and Michel, 2021, Section 5.4).

Usage

persistence_landscape(
  df,
  dimension = 0,
  k_max = NULL,
  resolution = 500,
  t_range = NULL
)

Arguments

Value

A data frame with columns t, k, value: the discretized landscape functions, one row per (grid point, level) pair.

Examples

points <- matrix(c(0, 1, 1, 0, 0, 0, 1, 1), ncol = 2)
filtration <- build_filtration(points, method = "VR", eps_max = 1.2)
res <- boundary_info(filtration)
pairs <- extract_persistence_pairs(filtration, res$last_1, res$pivot_owner)
landscape <- persistence_landscape(pairs, dimension = 0)

Persistence pairs via sparse boundary reduction (for large filtrations)

Description

Same standard column-reduction algorithm as boundary_info() + extract_persistence_pairs(), but each column is stored as a sorted integer vector over GF(2) instead of a row of a dense n x n matrix. Memory drops from O(n^2) to O(total number of non-zeros), which is what makes filtrations with thousands to hundreds of thousands of simplices (e.g. Flood or large Vietoris-Rips complexes) feasible. Output is identical to the dense pipeline.

Usage

persistence_pairs(filist, max_dimension = NULL)

Arguments

Value

A data frame with columns: dim, birth, and death.

Examples

points <- matrix(c(0, 1, 1, 0, 0, 0, 1, 1), ncol = 2)
filtration <- build_filtration(points, method = "VR", eps_max = 1.2)
pairs <- persistence_pairs(filtration)
pairs_h0_only <- persistence_pairs(filtration, max_dimension = 0)
# or bake the cap into the filtration itself, and drop the argument here:
capped <- build_filtration(points, method = "VR", eps_max = 1.2,
                            max_dimension = 0)
pairs_h0_only2 <- persistence_pairs(capped)

Plot a Persistence Landscape

Description

Plot a Persistence Landscape

Usage

plot_landscape(landscape_df)

Arguments

Value

A ggplot2 object with one line per landscape level \lambda(k, \cdot).

Examples

points <- matrix(c(0, 1, 1, 0, 0, 0, 1, 1), ncol = 2)
filtration <- build_filtration(points, method = "VR", eps_max = 1.2)
res <- boundary_info(filtration)
pairs <- extract_persistence_pairs(filtration, res$last_1, res$pivot_owner)
landscape <- persistence_landscape(pairs, dimension = 0)
plot_landscape(landscape)

Plot the optimal matching between two persistence diagrams

Description

Visualizes the point correspondence that realizes the Wasserstein or bottleneck distance between two persistence diagrams in a given homological dimension: both diagrams' points, overlaid on the same axes, joined by dashed lines to their matched partner - either a point of the other diagram, or (for a point left unmatched) its own projection onto the diagonal birth = death.

Usage

plot_matching(
  df1,
  df2,
  dimension,
  distance = c("wasserstein", "bottleneck"),
  p = 2,
  ground = NULL,
  labels = c("Diagram 1", "Diagram 2")
)

Arguments

Details

Essential (death = Inf) points are capped exactly the way wasserstein_distance does (see its Details) so they take part in the matching instead of being dropped, and are drawn as triangles at their capped height - marked "essential" in the legend - so they stay visually distinguishable from genuine finite points that happen to reach that height. Matches to the diagonal are always drawn to the point's perpendicular projection ((birth+death)/2, (birth+death)/2) regardless of ground, since that is the standard, readable way to depict an unmatched point whichever ground metric produced the matching.

Value

A ggplot2 object; the title reports the resulting distance.

Examples

df1 <- data.frame(dim = c(0, 0), birth = c(0, 0), death = c(1, 2))
df2 <- data.frame(dim = c(0, 0), birth = c(0, 0), death = c(1, 3))
plot_matching(df1, df2, dimension = 0, distance = "wasserstein", p = 2)

Plot Persistence Diagram

Description

Plot Persistence Diagram

Usage

plot_persistence(df)

Arguments

Value

A ggplot2 object representing the persistence diagram.

Examples

points <- matrix(c(0, 1, 1, 0, 0, 0, 1, 1), ncol = 2)
filtration <- build_filtration(points, method = "VR", eps_max = 1.2)
res <- boundary_info(filtration)
pairs <- extract_persistence_pairs(filtration, res$last_1, res$pivot_owner)
plot_persistence(pairs)

Restrict a filtration list to simplices up to a given dimension

Description

Drops every simplex of dimension greater than max_dimension from a filtration list; everything else (order, ties, t values) is left untouched.

Usage

restrict_filtration(filist, max_dimension)

Arguments

Value

A filtration list containing only the entries with dimension <= max_dimension, in the same relative order.

Examples

points <- matrix(c(0, 1, 1, 0, 0, 0, 1, 1), ncol = 2)
filtration <- build_filtration(points, method = "VR", eps_max = 1.2)
restrict_filtration(filtration, max_dimension = 1) # vertices + edges only

Expand maximal simplices into a sorted filtration list

Description

Shared filtration-assembly step used by every build_filtration() method: take a set of maximal simplices, generate every face of every dimension via faces, assign each face a filtration time via scale_fn, and sort by (time, dimension, lexicographic order) - the same convention used throughout the package (see faces for what "lexicographic order" means here).

Usage

simplices_to_filtration(maximal_simplices, scale_fn, max_dimension = NULL)

Arguments

Value

A filtration list: one list(simplex = <integer vector>, t = <numeric>) per face, sorted by (t, dimension, lexicographic order).

Wasserstein distance between two persistence diagrams

Description

The p-Wasserstein distance between the points of two persistence diagrams in a given homological dimension, allowing points to be matched to the diagonal (Cohen-Steiner, Edelsbrunner, Harer and Mileyko (2010), "Lipschitz Functions Have L_p-Stable Persistence"). Essential (death = Inf) classes are kept rather than dropped - see Details. Computed exactly via the Hungarian algorithm (clue::solve_LSAP) on the augmented assignment problem described in augmented_cost_matrix.

Usage

wasserstein_distance(df1, df2, dimension = 0, p = 2, ground = c("L2", "Linf"))

Arguments

Value

A single non-negative number, the p-Wasserstein distance.

Examples

df1 <- data.frame(dim = c(0, 0), birth = c(0, 0), death = c(1, 2))
df2 <- data.frame(dim = c(0, 0), birth = c(0, 0), death = c(1, 3))
wasserstein_distance(df1, df2, dimension = 0, p = 2)