Classificação do conjunto Iris usando knn em Julia.

iris.jl

#
# iris.jl
# Classificação no conjunto de dados Iris usando knn
#
# Andrei Formiga, 2013-06-17
#

using DataFrames

train_file = "iris.train.csv"
test_file = "iris.test.csv"

classes = ["Iris-setosa", "Iris-versicolor", "Iris-virginica"]

# calculo da distancia entre itens
function distance(data::AbstractDataFrame, i::Int,
                  seplen::Float64, sepwid::Float64,
                  petlen::Float64, petwid::Float64)
    dist = 0.0
    dist += (data[i, "seplen"] - seplen) ^ 2
    dist += (data[i, "sepwid"] - sepwid) ^ 2
    dist += (data[i, "petlen"] - petlen) ^ 2
    dist += (data[i, "petwid"] - petwid) ^ 2
    dist
end

function insert_neighbor(neighbors, dists, d::Float64, n::Int)
    index = 0
    l = length(neighbors)
    for i in 1:l
        if d < dists[i]
            index = i
            break
        end
    end
    if index > 0 && index <= l
        insert!(neighbors, index, n)
        insert!(dists, index, d)
    end
    (neighbors[1:l], dists[1:l])
end

function classify_knn(k, data, seplen::Float64, sepwid::Float64,
                      petlen::Float64, petwid::Float64)
    neighbors = zeros(Int, k)
    dists = zeros(Float64, k)
    for i in 1:length(dists)
        dists[i] = Inf
    end

    for i in 1:nrow(data)
        d = distance(data, i, seplen, sepwid, petlen, petwid)
        neighbors, dist = insert_neighbor(neighbors, dists, d, i)
    end
    
    count_classes = zeros(Int, 3)
    for i in neighbors
        for c in 1:3
            if data[i, "class"] == classes[c]
                count_classes[c] += 1
            end
        end
    end

    max = 0
    maxi = 0
    for c in 1:3
        if count_classes[c] > max
            max = count_classes[c]
            maxi = c
        end
    end
    
    classes[maxi]
end

function test_knn(k::Int)
    iris_train = read_table(train_file)
    iris_test = read_table(test_file)
    hits = 0
    for i in 1:nrow(iris_test)
        class = classify_knn(k, iris_train,
                             iris_test[i, "seplen"], iris_test[i, "sepwid"],
                             iris_test[i, "petlen"], iris_test[i, "petwid"])
        if class == iris_test[i, "class"]
            hits += 1
        end
    end
    percent = hits / nrow(iris_test) * 100.0
    println("Classificação por knn com k = $k")
    println("Acertos: $hits / $(nrow(iris_test)) ($percent %)")
end

if !isinteractive()
    test_knn(3)
end