From 9d1c074f3ae937dad5ad5fc9fbcce0a8f7acc2f2 Mon Sep 17 00:00:00 2001
From: pevnak <pevnak@gmail.com>
Date: Sat, 26 May 2018 07:17:40 +0200
Subject: [PATCH] added bandwidth function, which has been removed from
 statsbase. Updated to comply 0.6

---
 src/bandwidth.jl   | 23 +++++++++++++++++++++++
 src/classifier.jl  |  8 ++++----
 src/kNN.jl         |  1 +
 src/regress.jl     |  4 ++--
 test/classifier.jl |  6 +++---
 test/regress.jl    |  2 +-
 6 files changed, 34 insertions(+), 10 deletions(-)
 create mode 100644 src/bandwidth.jl

diff --git a/src/bandwidth.jl b/src/bandwidth.jl
new file mode 100644
index 0000000..f60ff16
--- /dev/null
+++ b/src/bandwidth.jl
@@ -0,0 +1,23 @@
+# Silverman's rule of thumb for KDE bandwidth selection
+function bandwidth(data::Vector{Float64}, alpha::Float64 = 0.9)
+    # Determine length of data
+    ndata = length(data)
+
+    # Calculate width using variance and IQR
+    var_width = std(data)
+    q25, q75 = quantile(data, [0.25, 0.75])
+    quantile_width = (q75 - q25) / 1.34
+
+    # Deal with edge cases with 0 IQR or variance
+    width = min(var_width, quantile_width)
+    if width == 0.0
+        if var_width == 0.0
+            width = 1.0
+        else
+            width = var_width
+        end
+    end
+
+    # Set bandwidth using Silverman's rule of thumb
+    return alpha * width * ndata^(-0.2)
+end
diff --git a/src/classifier.jl b/src/classifier.jl
index 97a4daa..f238f75 100644
--- a/src/classifier.jl
+++ b/src/classifier.jl
@@ -1,19 +1,19 @@
 immutable kNNClassifier
-    t::NaiveNeighborTree
+    t::BruteTree
     y::Vector
 end
 
 function knn(X::Matrix,
              y::Vector;
              metric::Metric = Euclidean())
-    return kNNClassifier(NaiveNeighborTree(X, metric), y)
+    return kNNClassifier(BruteTree(X, metric), y)
 end
 
 # TODO: Don't construct copy of model.y just to extract majority vote
 function StatsBase.predict(model::kNNClassifier,
                            x::Vector,
                            k::Integer = 1)
-    inds, dists = nearest(model.t, x, k)
+    inds, dists = NearestNeighbors.knn(model.t, x, k)
     return majority_vote(model.y[inds])
 end
 
@@ -33,7 +33,7 @@ end
 function StatsBase.predict(model::kNNClassifier,
                            X::Matrix,
                            k::Integer = 1)
-    predictions = Array(eltype(model.y), size(X, 2))
+    predictions = Array{eltype(model.y)}(size(X, 2))
     predict!(predictions, model, X, k)
     return predictions
 end
diff --git a/src/kNN.jl b/src/kNN.jl
index 9444ac6..5a415cb 100644
--- a/src/kNN.jl
+++ b/src/kNN.jl
@@ -9,6 +9,7 @@ module kNN
     using NearestNeighbors
 	using SmoothingKernels
 
+    include("bandwidth.jl")
     include("majority_vote.jl")
     include("classifier.jl")
     include("regress.jl")
diff --git a/src/regress.jl b/src/regress.jl
index 51cb561..a68fcae 100644
--- a/src/regress.jl
+++ b/src/regress.jl
@@ -11,7 +11,7 @@ function kernelregression{R <: Any,
                                      y::Vector{T};
                                      kernel::Symbol = :epanechnikov,
                                      bandwidth::Real = NaN,
-                                     getbandwidth::Function = StatsBase.bandwidth)
+                                     getbandwidth::Function = kNN.bandwidth)
     if isnan(bandwidth)
         h_x = getbandwidth(x)
         h_y = getbandwidth(y)
@@ -50,7 +50,7 @@ end
 
 function StatsBase.predict{T <: Real}(model::KernelRegression,
                                       xs::AbstractVector{T})
-    ys = Array(Float64, length(xs))
+    ys = Array{T}(length(xs))
     predict!(ys, model, xs)
     return ys
 end
diff --git a/test/classifier.jl b/test/classifier.jl
index 48b095f..0ef9424 100644
--- a/test/classifier.jl
+++ b/test/classifier.jl
@@ -7,9 +7,9 @@ module TestClassifier
     using Distances
     using StatsBase
 
-    iris = data("datasets", "iris")
-    X = array(iris[:, 1:4])'
-    y = array(iris[:, 5])
+    iris = dataset("datasets", "iris")
+    X = Array(iris[:, 1:4])'
+    y = Array(iris[:, 5])
     model = knn(X, y, metric = Euclidean())
 
     predict_k1 = predict(model, X, 1)
diff --git a/test/regress.jl b/test/regress.jl
index 87b3109..e6d6e6c 100644
--- a/test/regress.jl
+++ b/test/regress.jl
@@ -6,7 +6,7 @@ module TestRegress
 	srand(1)
 	n = 1_000
 	x = 10 * randn(n)
-	y = sin(x) + 0.5 * randn(n)
+	y = sin.(x) + 0.5 * randn(n)
 
 	fit = kernelregression(x, y)
 	grid = minimum(x):0.1:maximum(x)