Julia examples

Each example constructs its own data. The documentation build executes these examples, and the test suite executes the README's Julia fitting examples.

Logistic regression

using SpectralMM, Distributions, GLM, Random
rng = MersenneTwister(71)
X = randn(rng, 100, 3)
y = Float64.(rand(rng, 100) .< 0.5)
model = SpectralMM.glm(X, y, Bernoulli(), LogitLink())
@assert length(coef(model)) == 4
@assert length(predict(model, X)) == 100
coef(model)

4-element Vector{Float64}:
 -0.02006501003624904
 -0.37997911492894837
  0.001758992754949947
  0.18761977918256972

Sparse inputs

The response is generated from the same design, so its length always matches the number of observations. Sparse storage is retained during fitting; optional covariance computation uses dense coefficient matrices.

using SpectralMM, Distributions, GLM, SparseArrays, Random
rng = MersenneTwister(72)
X = sprandn(rng, 200, 8, 0.3)
mu = 0.4 .+ X * fill(0.15, size(X, 2))
y = mu + randn(rng, size(X, 1))
model = SpectralMM.glm(X, y, Normal(), IdentityLink(); inference=false)
@assert length(y) == size(X, 1)
@assert length(coef(model)) == size(X, 2) + 1
@assert all(isfinite, predict(model, X))
coef(model)

9-element Vector{Float64}:
  0.4196078121909915
 -0.03291313092308612
  0.009195761516944463
  0.08271954384842548
 -0.13216709096518722
  0.06399964227077326
  0.21908939016452683
  0.25259856966556554
  0.5081425374547809

Robust regression

using SpectralMM, Random
rng = MersenneTwister(73)
X = randn(rng, 100, 3)
y = 0.5 .+ X * [0.2, -0.3, 0.1] + randn(rng, 100)
model = SpectralMM.fit(X, y, PseudoHuber(); inference=false)
@assert length(coef(model)) == 4
coef(model)

4-element Vector{Float64}:
  0.5667256712582192
  0.10941105340430869
 -0.2750593482205364
  0.12300997249428401