File size: 7,505 Bytes
5c61046 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 | # tda_features.jl β Vietoris-Rips β Persistence Barcodes β Feature Vectors
module TDAFeatures
using LinearAlgebra
using Statistics
export VietorisRipsComplex, PersistenceDiagram, Barcode, barcode_to_feature_vector
export compute_persistence, wasserstein_distance, bottleneck_distance
export PersistenceInterval
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Vietoris-Rips Complex
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
struct VietorisRipsComplex
points::Matrix{Float64}
max_dim::Int
epsilon::Float64
simplices::Vector{Vector{Int}}
filtration_values::Vector{Float64}
end
function VietorisRipsComplex(points::Matrix{Float64}, epsilon::Float64; max_dim::Int=2)
n = size(points, 1)
simplices = Vector{Int}[]
filt_vals = Float64[]
for i in 1:n
push!(simplices, [i])
push!(filt_vals, 0.0)
end
for i in 1:n, j in i+1:n
d = norm(points[i,:] - points[j,:])
if d <= epsilon
push!(simplices, [i, j])
push!(filt_vals, d)
end
end
if max_dim >= 2
for i in 1:n, j in i+1:n, k in j+1:n
d_ij = norm(points[i,:] - points[j,:])
d_jk = norm(points[j,:] - points[k,:])
d_ik = norm(points[i,:] - points[k,:])
if d_ij <= epsilon && d_jk <= epsilon && d_ik <= epsilon
push!(simplices, [i, j, k])
push!(filt_vals, max(d_ij, d_jk, d_ik))
end
end
end
VietorisRipsComplex(points, max_dim, epsilon, simplices, filt_vals)
end
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Persistent Homology (H0 and H1)
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
struct PersistenceInterval
dim::Int
birth::Float64
death::Float64
end
struct PersistenceDiagram
intervals::Vector{PersistenceInterval}
end
struct Barcode
H0::Vector{PersistenceInterval}
H1::Vector{PersistenceInterval}
end
function compute_persistence(vr::VietorisRipsComplex)::Barcode
n = size(vr.points, 1)
order = sortperm(vr.filtration_values)
# H0: Connected components (union-find)
parent = collect(1:n)
rank = zeros(Int, n)
function find(x)
while parent[x] != x
parent[x] = parent[parent[x]]
x = parent[x]
end
return x
end
function union!(x, y)
rx, ry = find(x), find(y)
if rx != ry
if rank[rx] < rank[ry]
parent[rx] = ry
elseif rank[rx] > rank[ry]
parent[ry] = rx
else
parent[ry] = rx
rank[rx] += 1
end
return true
end
return false
end
H0_intervals = PersistenceInterval[]
for idx in order
simp = vr.simplices[idx]
val = vr.filtration_values[idx]
if length(simp) == 2
if union!(simp[1], simp[2])
push!(H0_intervals, PersistenceInterval(0, 0.0, val))
end
end
end
max_filt = maximum(vr.filtration_values)
for i in 1:n
if find(i) == i
push!(H0_intervals, PersistenceInterval(0, 0.0, max_filt))
end
end
# H1: Cycles
H1_intervals = PersistenceInterval[]
parent_h1 = collect(1:n)
function find_h1(x)
while parent_h1[x] != x
parent_h1[x] = parent_h1[parent_h1[x]]
x = parent_h1[x]
end
return x
end
function union_h1!(x, y)
rx, ry = find_h1(x), find_h1(y)
if rx != ry
parent_h1[rx] = ry
return false
end
return true
end
for idx in order
simp = vr.simplices[idx]
val = vr.filtration_values[idx]
if length(simp) == 2
if union_h1!(simp[1], simp[2])
push!(H1_intervals, PersistenceInterval(1, val, max_filt))
end
end
end
Barcode(H0_intervals, H1_intervals)
end
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Barcode β Feature Vector
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
function barcode_to_feature_vector(bc::Barcode; n_bins::Int=50, max_filt::Float64=1.0)::Vector{Float64}
features = Float64[]
for intervals in [bc.H0, bc.H1]
if isempty(intervals)
append!(features, zeros(n_bins))
continue
end
landscape = zeros(n_bins)
for intv in intervals
mid = (intv.birth + intv.death) / 2
half_pers = (intv.death - intv.birth) / 2
for (i, t) in enumerate(range(0, max_filt, length=n_bins))
val = max(0.0, half_pers - abs(t - mid))
landscape[i] = max(landscape[i], val)
end
end
append!(features, landscape)
end
push!(features, Float64(length(bc.H0)))
push!(features, Float64(length(bc.H1)))
push!(features, sum(i.death - i.birth for i in bc.H0))
push!(features, sum(i.death - i.birth for i in bc.H1))
push!(features, maximum([i.death - i.birth for i in bc.H1]; init=0.0))
return features
end
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
# Distances Between Barcodes
# βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
function wasserstein_distance(bc1::Barcode, bc2::Barcode; p::Int=2)::Float64
dist = 0.0
for (intervals1, intervals2) in [(bc1.H0, bc2.H0), (bc1.H1, bc2.H1)]
n1, n2 = length(intervals1), length(intervals2)
if n1 == 0 && n2 == 0
continue
elseif n1 == 0
dist += sum((i.death - i.birth)^p for i in intervals2)
elseif n2 == 0
dist += sum((i.death - i.birth)^p for i in intervals1)
else
sorted1 = sort(intervals1, by=i -> i.death - i.birth, rev=true)
sorted2 = sort(intervals2, by=i -> i.death - i.birth, rev=true)
for (i1, i2) in zip(sorted1, sorted2)
dist += abs((i1.death - i1.birth) - (i2.death - i2.birth))^p
end
end
end
return dist^(1/p)
end
function bottleneck_distance(bc1::Barcode, bc2::Barcode)::Float64
wasserstein_distance(bc1, bc2; p=100)
end
end # module TDAFeatures
|