initial build

.gitignore

@@ -4,3 +4,4 @@
 /Manifest.toml
 /docs/Manifest.toml
 /docs/build/
+.DS_Store

src/maxnet.jl

@@ -1,5 +1,11 @@
-module Maxnet
+module MaxNet
+using Lasso
+using StatsModels
+using Missings
 
 # Write your package code here.
 
+include("utils.jl")
+include("terms.jl")
+include("maxnet_function.jl")
 end

src/maxnet_function.jl

@@ -0,0 +1,31 @@
+function addsamples(presences, predictors) # this should be a general sdm thing
+    ret = Int[]
+    for i in eachindex(presences)
+        if presences[i]
+            add = true
+            for j in eachindex(presences)
+                if (!presences[j]) && predictors[j, :] == predictors[i, :]
+                    add = false
+                    break
+                end
+            end
+            add && push!(ret, i)
+        end
+    end
+    vcat(presences, zeros(length(ret))), vcat(predictors, predictors[ret, :])
+end
+
+function maxnet(presences, predictors, 
+                formula = maxnet_formula(presences, predictors),
+                regularization_multiplier = 1.0,
+                regularization_function = maxnet_default_regularization,
+                addsamplestobackground = true)
+
+    disallowmissing!(predictors)
+    if addsamplestobackground
+        presences, predictors = addsamples(presences, predictors)
+    end
+
+    #... TODO: This function is not yet finished
+
+end

src/terms.jl

@@ -0,0 +1,37 @@
+using StatsBase, StatsModels
+
+struct ThresholdTerm{T,N} <: AbstractTerm
+        term::T
+        nknots::N
+end
+
+thresholds(t::Symbol, nknots::Int) = ThresholdTerm(term(t), term(nknots))
+
+Base.show(io::IO, t::ThresholdTerm) = print(io, "thresholds($(t.term), nknots = $(t.nknots))")
+
+function StatsModels.apply_schema(t::FunctionTerm{typeof(thresholds)},
+                                sch::StatsModels.Schema,
+                                Mod::Type{<:Any})
+    apply_schema(ThresholdsTerm(t.args_parsed...), sch, Mod)
+end
+
+# apply_schema to internal Terms and check for proper types
+function StatsModels.apply_schema(t::ThresholdTerm,
+                                sch::StatsModels.Schema,
+                                Mod::Type{<:Any})
+    term = apply_schema(t.term, sch, Mod)
+    isa(term, ContinuousTerm) ||
+    throw(ArgumentError("ThresholdTerm only works with continuous terms (got $term)"))
+    isa(t.nknots, ConstantTerm) ||
+    throw(ArgumentError("ThresholdTerm nknots must be a number (got $t.nknots)"))
+    ThresholdTerm(term, t.nknots.n)
+end
+
+StatsModels.terms(p::ThresholdTerm) = terms(p.term)
+StatsModels.termvars(p::ThresholdTerm) = StatsModels.termvars(p.term)
+StatsModels.width(p::ThresholdTerm) = 1
+
+function StatsModels.modelcols(p::ThresholdTerm, d::NamedTuple)
+    col = modelcols(p.term, d)
+    reduce(hcat, [col.^n for n in 1:p.deg])
+end

src/utils.jl

@@ -0,0 +1,22 @@
+function hinge(x, nknots = 50)
+    mi, ma = extrema(x)
+    k = range(mi, ma, length = nknots)
+    lh = hingeval.(x, k[1:end-1]', [ma])
+    rh = hingeval.(x, [mi], k[2:end]')
+    [lh rh]
+end
+hingeval(x, mi, ma) = clamp((x - mi) / (ma - mi), 0, 1)
+
+function thresholds(x, nknots = 50)
+    k = range(extrema(x)..., length = nknots + 2)[2:nknots + 1]
+    Int.(x .>= k')
+end
+thresholdval(x, knot) = Int.(x .>= knot)
+
+categorical(x) = Int.(x .== unique(x)')
+categoricalval(x, category) = Int.(x .== category)
+
+function thresholds(x, nknots = 50)
+    k = range(extrema(x)..., length = nknots + 2)[2:nknots + 1]
+    Int.(x .>= k')
+end

test/runtests.jl

@@ -4,3 +4,14 @@ using Test
 @testset "Maxnet.jl" begin
     # Write your tests here.
 end
+
+# Test stub based on the R package
+using DataFrames, CSV, CategoricalArrays
+bradypus = CSV.read("data/bradypus.tsv", DataFrame)
+bradypus.ecoreg = categorical(bradypus.ecoreg)
+p = bradypus.presence
+data = bradypus[:, 2:end]
+mod = maxnet(p, data)
+plot(mod, type = "cloglog")
+mod = amxnet(p, data, maxnet_formula(p, data, classes = "lq"))
+plot(mod, "tmp6190_ann")