mathias-brandewinder

#r @"..\packages\Accord.2.8.1.0\lib\Accord.dll"
#r @"..\packages\Accord.Math.2.8.1.0\lib\Accord.Math.dll"
#r @"..\packages\Accord.Statistics.2.8.1.0\lib\Accord.Statistics.dll"
#r @"..\packages\Accord.MachineLearning.2.8.1.0\lib\Accord.MachineLearning.dll"

open System
open System.IO

open Accord.MachineLearning
open Accord.MachineLearning.VectorMachines

mathias-brandewinder

// This F# dojo is directly inspired by the 
// Digit Recognizer competition from Kaggle.com:
// http://www.kaggle.com/c/digit-recognizer
// The datasets below are simply shorter versions of
// the training dataset from Kaggle.
 
// The goal of the dojo will be to
// create a classifier that uses training data
// to recognize hand-written digits, and
// evaluate the quality of our classifier

mathias-brandewinder

(*
    Reading out bindings
*)

type Direction =
    | Trigger
    | In
    | Out

type Properties = Map<string,string>

mathias-brandewinder

// This is a conversion to F# of the C# code presented in
// MSDN Magazine, March 2015, by James McCaffrey:
// https://msdn.microsoft.com/en-us/magazine/dn913188.aspx

open System

let sumprod (v1:float[]) (v2:float[]) =
    Seq.zip v1 v2 |> Seq.sumBy (fun (x,y) -> x * y)

let sigmoid z = 1.0 / (1.0 + exp (- z))

mathias-brandewinder

(*
Simulation of the Lorentz attractor, using Fable.
If you want to see this code in action, just copy this code into the Fable REPL:
https://fable.io/repl/
*)

module App

open Elmish
open Elmish.React

mathias-brandewinder

type Program () =
    
    let f (x:float) = 2.0 * x + 1.0

    member this.Run (x:float) = 
        let result = f x
        printfn "Result: %.2f" result

let program = Program()
program.Run(10.0)

mathias-brandewinder

mathias-brandewinder

// This is the code sample for my #fsAdvent 2019 contribution.
// Check the corresponding blog post for some explanations :)
// https://brandewinder.com/2019/12/04/santas-mailbox/

open System
open System.Threading.Tasks
open System.Diagnostics
open System.Collections.Generic

let factorize x = 

mathias-brandewinder

namespace Discretization

// Recursive minimal entropy partitioning,
// based on Fayyad & Irani 1993. 
// See the following article, section 3.3,
// for a description of the algorithm:
// http://www.math.unipd.it/~dulli/corso04/disc.pdf
// Note: this can certainly be optimized.
module MDL =

mathias-brandewinder

CNTK + .NET Resources

• C#./.NET source: https://github.com/Microsoft/CNTK/tree/master/bindings/csharp
• C#/.NET API: https://docs.microsoft.com/en-us/cognitive-toolkit/cntk-library-managed-api
• Data file format: https://docs.microsoft.com/en-us/cognitive-toolkit/brainscript-cntktextformat-reader
• C# examples: https://github.com/Microsoft/CNTK/tree/release/latest/Examples/TrainingCSharp

Related resources

• Python API docs: https://cntk.ai/pythondocs/index.html