Allow fitting with a single data point and provide better error

src/Loess.jl

@@ -52,12 +52,15 @@ function loess(
     Base.require_one_based_indexing(xs)
     Base.require_one_based_indexing(ys)
 
-    if size(xs, 1) != size(ys, 1)
+    if size(xs, 1) != length(ys)
         throw(DimensionMismatch("Predictor and response arrays must of the same length"))
     end
+    if isempty(ys)
+        throw(ArgumentError("input arrays are empty"))
+    end
 
     n, m = size(xs)
-    q = floor(Int, span * n)
+    q = max(1, floor(Int, span * n))
 
     # TODO: We need to keep track of how we are normalizing so we can
     # correctly apply predict to unnormalized data. We should have a normalize

src/kd.jl

@@ -9,11 +9,11 @@ end
 
 
 struct KDTree{T <: AbstractFloat}
-    xs::Matrix{T}         # A matrix of n, m-dimensional observations
-    perm::Vector{Int}     # permutation of data to avoid modifying xs
-    root::KDNode{T}       # root node
+    xs::Matrix{T}                   # A matrix of n, m-dimensional observations
+    perm::Vector{Int}               # permutation of data to avoid modifying xs
+    root::Union{Nothing, KDNode{T}} # root node
     verts::Set{Vector{T}}
-    bounds::Matrix{T}     # Top-level bounding box
+    bounds::Matrix{T}               # Top-level bounding box
 end
 
 

test/runtests.jl

@@ -107,3 +107,9 @@ end
         end
     end
 end
+
+@testset "small datasets. Issue 82" begin
+    ft = loess([1.0], [1.0])
+    @test predict(ft, 1.0) == 1.0
+    @test_throws ArgumentError loess(Float64[], Float64[])
+end