classtromech

open System
open System.IO

[<Literal>]
let trainingDataFile = @"C:\Users\ben\Downloads\digitssample.csv"

type ImageData = {
    ActualNumber: byte
    Pixels: float []
}

let readData f =
    File.ReadAllLines f |> Seq.skip 1
                        |> Seq.map (fun str -> str.Split(','))
                        |> Seq.map (Array.map Convert.ToByte)
                        |> Seq.map (fun parts -> parts.[0], parts.[1 .. parts.Length - 1])
                        |> Seq.map (fun (num, pixels) -> { ActualNumber = num; Pixels = Array.map float pixels })
                        |> Array.ofSeq

let data = readData trainingDataFile

let distanceOf (xs: float[]) (ys: float[]) =
    let mutable sum = 0.
    for x in 0..(xs.Length - 1) do
        let v = pown (xs.[x] - ys.[x]) 2
        sum <- sum + v
    sum

let compareImages imageOne imageTwo =
    let pixelsOne = imageOne.Pixels
    let pixelsTwo = imageTwo.Pixels
    distanceOf pixelsOne pixelsTwo
    
let findNearestNeighbor image neighbors = neighbors |> Array.minBy (fun n -> compareImages image n)

let classify (unknown: float[]) =
    let image = { ActualNumber = 0uy; Pixels = unknown}
    let nearestNeighbor = findNearestNeighbor image data
    nearestNeighbor.ActualNumber
 
let testData = readData @"C:\Users\ben\Downloads\digitscheck.csv"

let totalRecords = ref 0
let correctRecords = ref 0

let testAnImage image = async {
    let nearestNeighbor = classify image.Pixels
    lock testData (fun () ->
                    incr totalRecords
                    if nearestNeighbor = image.ActualNumber then
                        incr correctRecords)
}

let sw = new System.Diagnostics.Stopwatch()

sw.Start()
testData |> Seq.map testAnImage
         |> Async.Parallel
         |> Async.RunSynchronously
         |> ignore

sw.Stop()
printfn "Percent correct: %f (%d ms)" (float !correctRecords / float !totalRecords) sw.ElapsedMilliseconds