commented out unused GGN

src/curvature/functions.jl

@@ -32,49 +32,49 @@ function gradients(curvature::CurvatureInterface, X::AbstractArray, y::Union{Num
     return 𝐠
 end
 
-"Constructor for Generalized Gauss Newton."
-struct GGN <: CurvatureInterface
-    model::Any
-    likelihood::Symbol
-    loss_fun::Function
-    params::AbstractArray
-    factor::Union{Nothing,Real}
-end
+# "Constructor for Generalized Gauss Newton."
+# struct GGN <: CurvatureInterface
+#     model::Any
+#     likelihood::Symbol
+#     loss_fun::Function
+#     params::AbstractArray
+#     factor::Union{Nothing,Real}
+# end
 
-function GGN(model::Any, likelihood::Symbol, params::AbstractArray)
+# function GGN(model::Any, likelihood::Symbol, params::AbstractArray)
 
-    @error "GGN not yet implemented."
+#     @error "GGN not yet implemented."
 
-    # Define loss function:
-    loss_fun = get_loss_fun(likelihood, model)
-    factor = likelihood == :regression ? 0.5 : 1.0
+#     # Define loss function:
+#     loss_fun = get_loss_fun(likelihood, model)
+#     factor = likelihood == :regression ? 0.5 : 1.0
 
-    GGN(model, likelihood, loss_fun, params, factor)
-end
+#     GGN(model, likelihood, loss_fun, params, factor)
+# end
 
-"""
-    full(curvature::GGN, d::Union{Tuple,NamedTuple})
+# """
+#     full(curvature::GGN, d::Union{Tuple,NamedTuple})
 
-Compute the full GGN.
-"""
-function full(curvature::GGN, d::Tuple)
-    x, y = d
+# Compute the full GGN.
+# """
+# function full(curvature::GGN, d::Tuple)
+#     x, y = d
 
-    loss = curvature.factor * curvature.loss_fun(x, y)
+#     loss = curvature.factor * curvature.loss_fun(x, y)
 
-    𝐉, fμ = jacobians(curvature, x)
+#     𝐉, fμ = jacobians(curvature, x)
 
-    if curvature.likelihood == :regression
-        H = 𝐉 * 𝐉'
-    else
-        p = outdim(curvature.model) > 1 ? softmax(fμ) : sigmoid(fμ)
-        H = map(j -> j * (diagm(p) - p * p') * j', eachcol(𝐉))
-        println(H)
-    end
+#     if curvature.likelihood == :regression
+#         H = 𝐉 * 𝐉'
+#     else
+#         p = outdim(curvature.model) > 1 ? softmax(fμ) : sigmoid(fμ)
+#         H = map(j -> j * (diagm(p) - p * p') * j', eachcol(𝐉))
+#         println(H)
+#     end
 
-    return loss, H
+#     return loss, H
 
-end
+# end
 
 "Constructor for Empirical Fisher."
 struct EmpiricalFisher <: CurvatureInterface