Refactor interface

Project.toml

@@ -7,6 +7,7 @@ version = "0.1.0"
 ColumnSelectors = "9cc86067-7e36-4c61-b350-1ac9833d277f"
 DensityRatioEstimation = "ab46fb84-d57c-11e9-2f65-6f72e4a7229f"
 GeoStatsBase = "323cb8eb-fbf6-51c0-afd0-f8fba70507b2"
+GeoStatsModels = "ad987403-13c5-47b5-afee-0a48f6ac4f12"
 GeoStatsTransforms = "725d9659-360f-4996-9c94-5f19c7e4a8a6"
 GeoTables = "e502b557-6362-48c1-8219-d30d308dcdb0"
 LossFunctions = "30fc2ffe-d236-52d8-8643-a9d8f7c094a7"
@@ -18,9 +19,11 @@ Transducers = "28d57a85-8fef-5791-bfe6-a80928e7c999"
 ColumnSelectors = "0.1"
 DensityRatioEstimation = "1.2"
 GeoStatsBase = "0.42"
+GeoStatsModels = "0.2"
 GeoStatsTransforms = "0.2"
 GeoTables = "1.14"
 LossFunctions = "0.11"
-StatsLearnModels = "0.2"
+Meshes = "0.37"
+StatsLearnModels = "0.3"
 Transducers = "0.4"
 julia = "1.9"

src/GeoStatsValidation.jl

@@ -5,7 +5,8 @@ using GeoTables
 using Transducers
 using DensityRatioEstimation
 
-using StatsLearnModels: Learn
+using GeoStatsModels: GeoStatsModel
+using StatsLearnModels: Learn, StatsLearnModel, input, output
 using GeoStatsTransforms: Interpolate, InterpolateNeighbors
 
 using ColumnSelectors: selector

src/cverror.jl

@@ -9,37 +9,52 @@ A method for estimating cross-validatory error.
 """
 abstract type ErrorMethod end
 
-struct LearnSetup{M}
+abstract type ErrorSetup end
+
+struct LearnSetup{M} <: ErrorSetup
   model::M
   input::Vector{Symbol}
   output::Vector{Symbol}
 end
 
-struct InterpSetup{I,M}
+struct InterpSetup{I,M,K} <: ErrorSetup
   model::M
+  kwargs::K
 end
 
 """
-    cverror(Learn, model, incols => outcols, geotable, method)
-    cverror(Interpolate, model, geotable, method)
-    cverror(InterpolateNeighbors, model, geotable, method)
+    cverror(model::GeoStatsModel, geotable, method; kwargs...)
+
+Estimate error of `model` in a given `geotable` with
+error estimation `method` using `Interpolate` or `InterpolateNeighbors`
+depending on the passed `kwargs`.
+
+    cverror(model::StatsLearnModel, geotable, method)
+    cverror((model, invars => outvars), geotable, method)
 
 Estimate error of `model` in a given `geotable` with
-error estimation `method`.
+error estimation `method` using the `Learn` transform.
 """
 function cverror end
 
-function cverror(::Type{Learn}, model, (incols, outcols)::Pair, geotable::AbstractGeoTable, method::ErrorMethod)
+cverror((model, cols)::Tuple{Any,Pair}, geotable::AbstractGeoTable, method::ErrorMethod) =
+  cverror(StatsLearnModel(model, first(cols), last(cols)), geotable, method)
+
+function cverror(model::StatsLearnModel, geotable::AbstractGeoTable, method::ErrorMethod)
   names = setdiff(propertynames(geotable), [:geometry])
-  input = selector(incols)(names)
-  output = selector(outcols)(names)
-  cverror(LearnSetup(model, input, output), geotable, method)
+  invars = input(model)(names)
+  outvars = output(model)(names)
+  setup = LearnSetup(model, invars, outvars)
+  cverror(setup, geotable, method)
 end
 
-const Interp = Union{Interpolate,InterpolateNeighbors}
+const INTERPNEIGHBORS = (:minneighbors, :maxneighbors, :neighborhood, :distance)
 
-cverror(::Type{I}, model::M, geotable::AbstractGeoTable, method::ErrorMethod) where {I<:Interp,M} =
-  cverror(InterpSetup{I,M}(model), geotable, method)
+function cverror(model::M, geotable::AbstractGeoTable, method::ErrorMethod; kwargs...) where {M<:GeoStatsModel}
+  I = any(∈(INTERPNEIGHBORS), keys(kwargs)) ? InterpolateNeighbors : Interpolate
+  setup = InterpSetup{I,M,typeof(kwargs)}(model, kwargs)
+  cverror(setup, geotable, method)
+end
 
 # ----------------
 # IMPLEMENTATIONS

src/cverrors/bcv.jl

@@ -26,7 +26,7 @@ end
 
 BlockValidation(sides; loss=Dict()) = BlockValidation{typeof(sides)}(sides, loss)
 
-function cverror(setup, geotable, method::BlockValidation)
+function cverror(setup::ErrorSetup, geotable::AbstractGeoTable, method::BlockValidation)
   # uniform weights
   weighting = UniformWeighting()
 

src/cverrors/drv.jl

@@ -49,7 +49,7 @@ function DensityRatioValidation(
   DensityRatioValidation{T,E,O}(k, shuffle, lambda, estimator, optlib, loss)
 end
 
-function cverror(setup::LearnSetup, geotable, method::DensityRatioValidation)
+function cverror(setup::LearnSetup, geotable::AbstractGeoTable, method::DensityRatioValidation)
   vars = setup.input
 
   # density-ratio weights

src/cverrors/kfv.jl

@@ -25,7 +25,7 @@ end
 
 KFoldValidation(k::Int; shuffle=true, loss=Dict()) = KFoldValidation(k, shuffle, loss)
 
-function cverror(setup, geotable, method::KFoldValidation)
+function cverror(setup::ErrorSetup, geotable::AbstractGeoTable, method::KFoldValidation)
   # uniform weights
   weighting = UniformWeighting()
 

src/cverrors/lbo.jl

@@ -29,7 +29,7 @@ LeaveBallOut(ball; loss=Dict()) = LeaveBallOut{typeof(ball)}(ball, loss)
 
 LeaveBallOut(radius::Number; loss=Dict()) = LeaveBallOut(MetricBall(radius), loss=loss)
 
-function cverror(setup, geotable, method::LeaveBallOut)
+function cverror(setup::ErrorSetup, geotable::AbstractGeoTable, method::LeaveBallOut)
   # uniform weights
   weighting = UniformWeighting()
 

src/cverrors/loo.jl

@@ -19,7 +19,7 @@ end
 
 LeaveOneOut(; loss=Dict()) = LeaveOneOut(loss)
 
-function cverror(setup, geotable, method::LeaveOneOut)
+function cverror(setup::ErrorSetup, geotable::AbstractGeoTable, method::LeaveOneOut)
   # uniform weights
   weighting = UniformWeighting()
 

src/cverrors/wcv.jl

@@ -33,7 +33,7 @@ end
 WeightedValidation(weighting::W, folding::F; lambda::T=one(T), loss=Dict()) where {W,F,T} =
   WeightedValidation{W,F,T}(weighting, folding, lambda, loss)
 
-function cverror(setup, geotable, method::WeightedValidation)
+function cverror(setup::ErrorSetup, geotable::AbstractGeoTable, method::WeightedValidation)
   ovars = _outputs(setup, geotable)
   loss = method.loss
   for var in ovars
@@ -86,11 +86,11 @@ _outputs(s::LearnSetup, gtb) = s.output
 function _prediction(s::InterpSetup{I}, geotable, f) where {I}
   sdat = view(geotable, f[1])
   sdom = view(domain(geotable), f[2])
-  sdat |> I(sdom, s.model)
+  sdat |> I(sdom, s.model; s.kwargs...)
 end
 
 function _prediction(s::LearnSetup, geotable, f)
   source = view(geotable, f[1])
   target = view(geotable, f[2])
-  target |> Learn(source, s.model, s.input => s.output)
+  target |> Learn(source, s.model)
 end

test/runtests.jl

@@ -19,7 +19,7 @@ using Test
 
     # dummy classifier → 0.5 misclassification rate
     for method in [LeaveOneOut(), LeaveBallOut(0.1), KFoldValidation(10), BlockValidation(0.1), DensityRatioValidation(10)]
-      e = cverror(Learn, model, :x => :y, gtb, method)
+      e = cverror((model, :x => :y), gtb, method)
       @test isapprox(e[:y], 0.5, atol=0.06)
     end
   end
@@ -34,8 +34,8 @@ using Test
     # low variance + dummy (mean) estimator → low error
     # high variance + dummy (mean) estimator → high error
     for method in [LeaveOneOut(), LeaveBallOut(0.1), KFoldValidation(10), BlockValidation(0.1)]
-      e₁ = cverror(Interpolate, model, sgtb₁, method)
-      e₂ = cverror(Interpolate, model, sgtb₂, method)
+      e₁ = cverror(model, sgtb₁, method)
+      e₂ = cverror(model, sgtb₂, method)
       @test e₁[:z] < 1
       @test e₂[:z] > 1
     end