F Sharp implementation of the Kaggle digit recognizer exercise.

LJFSharp

open System.IO
open System
open System.Text

type DataWithDigit = {  digit:int
                        data:int array }
type Data = {data:int array}

let convertToTrainingData (x:string) =
    let split = x.Split(',')
    { digit = (int)split.[0]
      data = split.[1..] |> Array.map(int) }

let convertToTestData (x:string) =
    { data = x.Split(',') |> Array.map(int) }

let readData file converter=
    (File.ReadAllLines file).[1..] |> Array.Parallel.map converter

let GetData trainingFile testFile =
    readData trainingFile convertToTrainingData, (readData testFile convertToTestData)

let PutData data =
    File.WriteAllLines("results.csv", data |> Array.map(string))
    
let euclideanDistance x y =
   (x,y)||> Array.fold2(fun a b c -> a + pown (b-c) 2) 0

let classifyEuclideanDistance (x:DataWithDigit array) (y:Data) = 
    x |> Array.minBy(fun a -> euclideanDistance (y.data) (a.data))

//do    
let data = GetData (__SOURCE_DIRECTORY__ + @"\..\" + "train.csv") (__SOURCE_DIRECTORY__ + @"\..\" + "test.csv")

let classifyFunction = classifyEuclideanDistance (fst data)
let results = (snd data) |> Array.Parallel.map(classifyFunction)

PutData (results |> Array.map(fun x-> x.digit.ToString()))
//let comparisonData = (fst data).[1000..1500], results
//let compareResults =  comparisonData ||> Array.map2(fun x y->x, y, x.digit = y.digit)
//printfn "%d" (compareResults |> Array.filter(fun (x,y,z)->not z)|>Array.length)