Merge pull request #24 from evanfields/ef_update_07

Ef update 0.7

Author: tlnagy

.travis.yml

@@ -3,8 +3,7 @@ os:
     - linux
     - osx
 julia:
-    - 0.5
-    - 0.6
+    - 0.7
     - nightly
 notifications:
     email: false

README.md

@@ -1,8 +1,8 @@
 # Loess
 
 [![Build Status](https://travis-ci.org/JuliaStats/Loess.jl.svg?branch=master)](https://travis-ci.org/JuliaStats/Loess.jl)
-[![Loess](http://pkg.julialang.org/badges/Loess_0.5.svg)](http://pkg.julialang.org/?pkg=Loess)
 [![Loess](http://pkg.julialang.org/badges/Loess_0.6.svg)](http://pkg.julialang.org/?pkg=Loess)
+[![Loess](http://pkg.julialang.org/badges/Loess_0.7.svg)](http://pkg.julialang.org/?pkg=Loess)
 
 This is a pure Julia loess implementation, based on the fast kd-tree based
 approximation described in the original Cleveland, et al papers, implemented

REQUIRE

@@ -1,4 +1,2 @@
-julia 0.5
-Compat 0.17
+julia 0.7.0-beta2
 Distances
-IterTools

src/Loess.jl

@@ -1,18 +1,18 @@
-isdefined(Base, :__precompile__) && __precompile__()
+__precompile__()
 
-module Loess
 
-using Compat
+module Loess
 
-import IterTools.product
 import Distances.euclidean
 
+using Statistics
+
 export loess, predict
 
 include("kd.jl")
 
 
-type LoessModel{T <: AbstractFloat}
+mutable struct LoessModel{T <: AbstractFloat}
     xs::AbstractMatrix{T} # An n by m predictor matrix containing n observations from m predictors
     ys::AbstractVector{T} # A length n response vector
     bs::Matrix{T}         # Least squares coefficients
@@ -37,8 +37,8 @@ Returns:
   A fit `LoessModel`.
 
 """
-function loess{T <: AbstractFloat}(xs::AbstractMatrix{T}, ys::AbstractVector{T};
-	                           normalize::Bool=true, span::T=0.75, degree::Int=2)
+function loess(xs::AbstractMatrix{T}, ys::AbstractVector{T};
+           normalize::Bool=true, span::T=0.75, degree::Int=2) where T <: AbstractFloat
     if size(xs, 1) != size(ys, 1)
 	error("Predictor and response arrays must of the same length")
     end
@@ -54,7 +54,7 @@ function loess{T <: AbstractFloat}(xs::AbstractMatrix{T}, ys::AbstractVector{T};
     end
 
     kdtree = KDTree(xs, 0.05 * span)
-    verts = Array{T}(length(kdtree.verts), m)
+    verts = Array{T}(undef, length(kdtree.verts), m)
 
     # map verticies to their index in the bs coefficient matrix
     verts = Dict{Vector{T}, Int}()
@@ -63,13 +63,13 @@ function loess{T <: AbstractFloat}(xs::AbstractMatrix{T}, ys::AbstractVector{T};
     end
 
     # Fit each vertex
-    ds = Array{T}(n) # distances
+    ds = Array{T}(undef, n) # distances
     perm = collect(1:n)
-    bs = Array{T}(length(kdtree.verts), 1 + degree * m)
+    bs = Array{T}(undef, length(kdtree.verts), 1 + degree * m)
 
     # TODO: higher degree fitting
-    us = Array{T}(q, 1 + degree * m)
-    vs = Array{T}(q)
+    us = Array{T}(undef, q, 1 + degree * m)
+    vs = Array{T}(undef, q)
     for (vert, k) in verts
         # reset perm
 	for i in 1:n
@@ -82,7 +82,7 @@ function loess{T <: AbstractFloat}(xs::AbstractMatrix{T}, ys::AbstractVector{T};
 	end
 
 	# copy the q nearest points to vert into X
-	select!(perm, q, by=i -> ds[i])
+	partialsort!(perm, q, by=i -> ds[i])
 	dmax = maximum([ds[perm[i]] for i = 1:q])
 
 	for i in 1:q
@@ -105,8 +105,8 @@ function loess{T <: AbstractFloat}(xs::AbstractMatrix{T}, ys::AbstractVector{T};
     LoessModel{T}(xs, ys, bs, verts, kdtree)
 end
 
-function loess{T <: AbstractFloat}(xs::AbstractVector{T}, ys::AbstractVector{T};
-	                           normalize::Bool=true, span::T=0.75, degree::Int=2)
+function loess(xs::AbstractVector{T}, ys::AbstractVector{T};
+           normalize::Bool=true, span::T=0.75, degree::Int=2) where T <: AbstractFloat
     loess(reshape(xs, (length(xs), 1)), ys, normalize=normalize, span=span, degree=degree)
 end
 
@@ -125,12 +125,12 @@ end
 # Returns:
 #   A length n' vector of predicted response values.
 #
-function predict{T <: AbstractFloat}(model::LoessModel{T}, z::T)
+function predict(model::LoessModel{T}, z::T) where T <: AbstractFloat
 	predict(model, T[z])
 end
 
 
-function predict{T <: AbstractFloat}(model::LoessModel{T}, zs::AbstractVector{T})
+function predict(model::LoessModel{T}, zs::AbstractVector{T}) where T <: AbstractFloat
     m = size(model.xs, 2)
 
     # in the univariate case, interpret a non-singleton zs as vector of
@@ -163,8 +163,8 @@ function predict{T <: AbstractFloat}(model::LoessModel{T}, zs::AbstractVector{T}
 end
 
 
-function predict{T <: AbstractFloat}(model::LoessModel{T}, zs::AbstractMatrix{T})
-	ys = Array{T}(size(zs, 1))
+function predict(model::LoessModel{T}, zs::AbstractMatrix{T}) where T <: AbstractFloat
+	ys = Array{T}(undef, size(zs, 1))
 	for i in 1:size(zs, 1)
 		# the vec() here is not necessary on 0.5 anymore
 		ys[i] = predict(model, vec(zs[i,:]))
@@ -226,7 +226,7 @@ Args:
 Modifies:
   `xs`
 """
-function tnormalize!{T <: AbstractFloat}(xs::AbstractMatrix{T}, q::T=0.1)
+function tnormalize!(xs::AbstractMatrix{T}, q::T=0.1) where T <: AbstractFloat
     n, m = size(xs)
     cut = ceil(Int, (q * n))
     for j in 1:m

src/kd.jl

@@ -1,22 +1,19 @@
-using Compat
-import Compat.view
-
 # Simple static kd-trees.
 
-@compat abstract type KDNode end
+abstract type KDNode end
 
-immutable KDLeafNode <: KDNode
+struct KDLeafNode <: KDNode
 end
 
-immutable KDInternalNode{T <: AbstractFloat} <: KDNode
+struct KDInternalNode{T <: AbstractFloat} <: KDNode
     j::Int             # dimension on which the data is split
     med::T             # median value where the split occours
     leftnode::KDNode
     rightnode::KDNode
 end
 
 
-immutable KDTree{T <: AbstractFloat}
+struct KDTree{T <: AbstractFloat}
     xs::AbstractMatrix{T} # A matrix of n, m-dimensional observations
     perm::Vector{Int}     # permutation of data to avoid modifying xs
     root::KDNode          # root node
@@ -42,14 +39,14 @@ Returns:
   A `KDTree` object
 
 """
-function KDTree{T <: AbstractFloat}(xs::AbstractMatrix{T},
-	                                leaf_size_factor=0.05,
-	                                leaf_diameter_factor=0.0)
+function KDTree(xs::AbstractMatrix{T},
+                leaf_size_factor=0.05,
+                leaf_diameter_factor=0.0) where T <: AbstractFloat
 
     n, m = size(xs)
     perm = collect(1:n)
 
-    bounds = Array{T}(2, m)
+    bounds = Array{T}(undef, 2, m)
     for j in 1:m
 	col = xs[:,j]
 	bounds[1, j] = minimum(col)
@@ -63,7 +60,7 @@ function KDTree{T <: AbstractFloat}(xs::AbstractMatrix{T},
     verts = Set{Vector{T}}()
 
     # Add a vertex for each corner of the hypercube
-    for vert in product([bounds[:,j] for j in 1:m]...)
+    for vert in Iterators.product([bounds[:,j] for j in 1:m]...)
 	push!(verts, T[vert...])
     end
 
@@ -116,12 +113,12 @@ Modifies:
 Returns:
   Either a `KDLeafNode` or a `KDInternalNode`
 """
-function build_kdtree{T}(xs::AbstractMatrix{T},
-	                 perm::AbstractArray,
-	                 bounds::Matrix{T},
-	                 leaf_size_cutoff::Int,
-	                 leaf_diameter_cutoff::T,
-	                 verts::Set{Vector{T}})
+function build_kdtree(xs::AbstractMatrix{T},
+                      perm::AbstractArray,
+                      bounds::Matrix{T},
+                      leaf_size_cutoff::Int,
+                      leaf_diameter_cutoff::T,
+                      verts::Set{Vector{T}}) where T
     n, m = size(xs)
 
     if length(perm) <= leaf_size_cutoff || diameter(bounds) <= leaf_diameter_cutoff
@@ -148,14 +145,14 @@ function build_kdtree{T}(xs::AbstractMatrix{T},
     # find the median and partition
     if isodd(length(perm))
 	mid = length(perm) ÷ 2
-	select!(perm, mid, by=i -> xs[i, j])
+	partialsort!(perm, mid, by=i -> xs[i, j])
 	med = xs[perm[mid], j]
 	mid1 = mid
 	mid2 = mid + 1
     else
 	mid1 = length(perm) ÷ 2
 	mid2 = mid1 + 1
-	select!(perm, mid1:mid2, by=i -> xs[i, j])
+	partialsort!(perm, mid1:mid2, by=i -> xs[i, j])
 	med = (xs[perm[mid1], j] + xs[perm[mid2], j]) / 2
     end
 
@@ -169,7 +166,7 @@ function build_kdtree{T}(xs::AbstractMatrix{T},
     rightnode = build_kdtree(xs, view(perm,mid2:length(perm)), rightbounds,
 		             leaf_size_cutoff, leaf_diameter_cutoff, verts)
 
-    coords = Array{Array}(m)
+    coords = Array{Array}(undef, m)
     for i in 1:m
 	if i == j
 	    coords[i] = [med]
@@ -178,7 +175,7 @@ function build_kdtree{T}(xs::AbstractMatrix{T},
 	end
     end
 
-    for vert in product(coords...)
+    for vert in Iterators.product(coords...)
 	push!(verts, T[vert...])
     end
 
@@ -192,7 +189,7 @@ end
 Given a bounding hypecube `bounds`, return its verticies
 """
 function bounds_verts(bounds::Matrix)
-    collect(product([bounds[:, i] for i in 1:size(bounds, 2)]...))
+    collect(Iterators.product([bounds[:, i] for i in 1:size(bounds, 2)]...))
 end
 
 
@@ -202,7 +199,7 @@ end
 Traverse the tree `kdtree` to the bottom and return the verticies of
 the bounding hypercube of the leaf node containing the point `x`.
 """
-function traverse{T}(kdtree::KDTree{T}, x::AbstractVector{T})
+function traverse(kdtree::KDTree{T}, x::AbstractVector{T}) where T
     m = size(kdtree.bounds, 2)
 
     if length(x) != m

test/runtests.jl

@@ -1,8 +1,9 @@
 # These tests don't do much except ensure that the package loads,
 # and does something sensible if it does.
 using Loess
-using Base.Test
-using Compat
+using Test
+using Random
+using Statistics
 
 srand(100)
 xs = 10 .* rand(100)