using Jchemo, JchemoData
using JLD2, DataFrames, CairoMakie, GLMakie
using LinearAlgebra, Random
using ManifoldLearning

Data simulation

n = 1000
noise = .5  # "vertical" noise (axis3)
#noise = 2
segments = 4
hlims = (-10.0, 10.0)  # seems to impact only axis2
rng = TaskLocalRNG()
#rng = MersenneTwister(1234)
Xt, L = ManifoldLearning.swiss_roll(n, noise; segments, hlims, rng)
@head X = Xt'

3×3 Matrix{Float64}:
 12.9293   -8.31783  0.672177
 -8.07359   1.00253  4.28689
  2.7955    7.80959  6.91284
... (1000, 3)

labs = vec(L)
tab(labs)

OrderedCollections.OrderedDict{Int64, Int64} with 4 entries:
  0 => 155
  1 => 339
  2 => 335
  3 => 171

CairoMakie.activate!()  
#GLMakie.activate!()    # for interactive axe-rotation

mks = 10
i = 1
plotxyz(X[:, i], X[:, i + 1], X[:, i + 2], labs; size = (700, 500), markersize = 10, 
    xlabel = string("x", i), ylabel = string("x", i + 1), zlabel = string("x", i + 2), 
    title = "Swiss roll").f

Pca

nlv = 2
model = pcasvd(; nlv)
fit!(model, X)
@head T = model.fitm.T

3×2 Matrix{Float64}:
 -7.48938  -7.48381
  3.81531  10.2805
 -5.31822   3.6551
... (1000, 2)

i = 1
plotxy(T[:, i], T[:, i + 1]; color = labs,
    xlabel = string("LV", i), ylabel = string("LV", i + 1), title = "Pca").f

Umap

nlv = 2
n_neighbors = 15; min_dist = .5 
model = umap(; nlv, n_neighbors, min_dist)
fit!(model, X)
@head T = model.fitm.T

3×2 Matrix{Float64}:
 -9.62654   0.796588
  3.01404  -4.81665
  9.02768  -2.63106
... (1000, 2)

i = 1
plotxy(T[:, i], T[:, i + 1]; color = labs,
    xlabel = string("LV", i), ylabel = string("LV", i + 1), title = "Umap").f

Tsne

p = 30
maxoutdim = 2
M = ManifoldLearning.fit(TSNE, X'; p, maxoutdim) 
Tt =  ManifoldLearning.predict(M) 
@head T = Tt'

3×2 Matrix{Float64}:
  11.8462   -39.3409
 -28.3948    12.1073
  -1.34849   39.8881
... (1000, 2)

i = 1
plotxy(T[:, i], T[:, i + 1]; color = labs,
    xlabel = string("LV", i), ylabel = string("LV", i + 1), title = "t-SNE").f

Kpca

nlv = 2
kern = :krbf; gamma = 1e-2
model = kpca(; nlv, kern, gamma)
fit!(model, X)
@head T = model.fitm.T

3×2 Matrix{Float64}:
 -0.202924   0.193962
  0.344041  -0.551969
 -0.304295  -0.322843
... (1000, 2)

i = 1
plotxy(T[:, i], T[:, i + 1]; color = labs,
    xlabel = string("LV", i), ylabel = string("LV", i + 1), title = "Kpca").f

nlv = 2
kern = :kpol
gamma = 1; degree = 3; coef0 = 10
model = kpca(; nlv, kern, degree, gamma, coef0)
fit!(model, X)
@head T = model.fitm.T  
i = 1
plotxy(T[:, i], T[:, i + 1]; color = labs,
    xlabel = string("LV", i), ylabel = string("LV", i + 1), title = "Kpca").f

3×2 Matrix{Float64}:
 -929.434  -1099.51
  347.469   -189.924
 -453.346    459.152
... (1000, 2)

---