mlnettest1.fsx

#r @"..\packages\NETStandard.Library.NETFramework.2.0.0-preview2-25405-01\build\net461\lib\netstandard.dll"



#r @"..\packages\Google.Protobuf.3.5.1\lib\net45\Google.Protobuf.dll" //Google.Protobuf
#r @"..\packages\Microsoft.ML.0.1.0\lib\netstandard2.0\Microsoft.ML.dll" //Microsoft.ML
#r @"..\packages\Microsoft.ML.0.1.0\lib\netstandard2.0\Microsoft.ML.Api.dll" //Microsoft.ML.Api
#r @"..\packages\Microsoft.ML.0.1.0\lib\netstandard2.0\Microsoft.ML.Core.dll" //Microsoft.ML.Core
#r @"..\packages\Microsoft.ML.0.1.0\lib\netstandard2.0\Microsoft.ML.CpuMath.dll" //Microsoft.ML.CpuMath
#r @"..\packages\Microsoft.ML.0.1.0\lib\netstandard2.0\Microsoft.ML.Data.dll" //Microsoft.ML.Data
#r @"..\packages\Microsoft.ML.0.1.0\lib\netstandard2.0\Microsoft.ML.FastTree.dll" //Microsoft.ML.FastTree
#r @"..\packages\Microsoft.ML.0.1.0\lib\netstandard2.0\Microsoft.ML.InternalStreams.dll" //Microsoft.ML.InternalStreams
#r @"..\packages\Microsoft.ML.0.1.0\lib\netstandard2.0\Microsoft.ML.KMeansClustering.dll" //Microsoft.ML.KMeansClustering
#r @"..\packages\Microsoft.ML.0.1.0\lib\netstandard2.0\Microsoft.ML.Maml.dll" //Microsoft.ML.Maml
#r @"..\packages\Microsoft.ML.0.1.0\lib\netstandard2.0\Microsoft.ML.PCA.dll" //Microsoft.ML.PCA
#r @"..\packages\Microsoft.ML.0.1.0\lib\netstandard2.0\Microsoft.ML.PipelineInference.dll" //Microsoft.ML.PipelineInference
#r @"..\packages\Microsoft.ML.0.1.0\lib\netstandard2.0\Microsoft.ML.ResultProcessor.dll" //Microsoft.ML.ResultProcessor
#r @"..\packages\Microsoft.ML.0.1.0\lib\netstandard2.0\Microsoft.ML.StandardLearners.dll" //Microsoft.ML.StandardLearners
#r @"..\packages\Microsoft.ML.0.1.0\lib\netstandard2.0\Microsoft.ML.Sweeper.dll" //Microsoft.ML.Sweeper
#r @"..\packages\Microsoft.ML.0.1.0\lib\netstandard2.0\Microsoft.ML.Transforms.dll" //Microsoft.ML.Transforms
#r @"..\packages\Microsoft.ML.0.1.0\lib\netstandard2.0\Microsoft.ML.UniversalModelFormat.dll" //Microsoft.ML.UniversalModelFormat
#r @"..\packages\Newtonsoft.Json.10.0.3\lib\net45\Newtonsoft.Json.dll" //Newtonsoft.Json
#r @"..\packages\NuGet.Client.4.0.0\lib\net45\NuGet.Client.dll" //NuGet.Client
#r @"..\packages\NuGet.Common.4.0.0\lib\net45\NuGet.Common.dll" //NuGet.Common
#r @"..\packages\NuGet.ContentModel.4.0.0\lib\net45\NuGet.ContentModel.dll" //NuGet.ContentModel
#r @"..\packages\NuGet.Frameworks.4.0.0\lib\net45\NuGet.Frameworks.dll" //NuGet.Frameworks
#r @"..\packages\NuGet.Packaging.4.0.0\lib\net45\NuGet.Packaging.dll" //NuGet.Packaging
#r @"..\packages\NuGet.Packaging.Core.4.0.0\lib\net45\NuGet.Packaging.Core.dll" //NuGet.Packaging.Core
#r @"..\packages\NuGet.Packaging.Core.Types.4.0.0\lib\net45\NuGet.Packaging.Core.Types.dll" //NuGet.Packaging.Core.Types
#r @"..\packages\NuGet.Repositories.4.0.0\lib\net45\NuGet.Repositories.dll" //NuGet.Repositories
#r @"..\packages\NuGet.RuntimeModel.4.0.0\lib\net45\NuGet.RuntimeModel.dll" //NuGet.RuntimeModel
#r @"..\packages\NuGet.Versioning.4.0.0\lib\net45\NuGet.Versioning.dll" //NuGet.Versioning
#r @"..\packages\System.CodeDom.4.4.0\lib\net461\System.CodeDom.dll" //System.CodeDom
#r @"..\packages\System.Threading.Tasks.Dataflow.4.8.0\lib\netstandard2.0\System.Threading.Tasks.Dataflow.dll" //System.Threading.Tasks.Dataflow
#r @"..\packages\System.ValueTuple.4.4.0\lib\net47\System.ValueTuple.dll" //System.ValueTuple


open Microsoft.ML.Models
open Microsoft.ML.Runtime
open Microsoft.ML.Runtime.Api
open Microsoft.ML.Trainers
open Microsoft.ML.Transforms

open System.Net
open System.IO
open Microsoft.ML
open System


//CpuMathNative location
let libDir = @"C:\packages\Microsoft.ML.0.1.0\runtimes\win-x64\native"

Environment.SetEnvironmentVariable("PATH", libDir + string Path.PathSeparator + Environment.GetEnvironmentVariable("PATH")) 

let file name = 
    if File.Exists(name) then 
        name
    else
        use wc = new WebClient()
        let url = sprintf "https://raw.githubusercontent.com/dotnet/machinelearning/master/test/data/%s" name
        printfn "%s not found." name
        printfn "Downloading... %s" url
        wc.DownloadFile(url,name)
        printfn "Done"
        name

type SentimentData  =
    [<Column(ordinal = "0", name = "Label")>]
    val mutable Sentiment : double
    [<Column(ordinal = "1")>]
    val mutable SentimentText : string
    new() = {Sentiment = 0.0; SentimentText = ""}
    new(txt) = {Sentiment = 0.0; SentimentText = txt}


type SentimentPrediction  =
    [<ColumnName("PredictedLabel")>]
    val mutable Sentiment : bool
    new() = {Sentiment = false}


let sentimentDataPath = file "wikipedia-detox-250-line-data.tsv"
let sentimentTestPath = file "wikipedia-detox-250-line-test.tsv"

let dataPath = Path.GetFullPath sentimentDataPath
let pipeline = LearningPipeline();

pipeline.Add(TextLoader<SentimentData>(dataPath,useHeader = true, separator = "tab"))
pipeline.Add(TextFeaturizer("Features", "SentimentText", 
                                KeepDiacritics = false,
                                KeepPunctuations = false,
                                TextCase = TextNormalizerTransformCaseNormalizationMode.Lower,
                                OutputTokens = true,
                                StopWordsRemover = PredefinedStopWordsRemover(),
                                VectorNormalizer = TextTransformTextNormKind.L2,
                                CharFeatureExtractor = NGramNgramExtractor(NgramLength = 2, AllLengths = true),
                                WordFeatureExtractor = NGramNgramExtractor(NgramLength = 3, AllLengths = false)))


pipeline.Add(FastTreeBinaryClassifier(NumLeaves = 5, NumTrees = 5, MinDocumentsInLeafs = 2));
pipeline.Add(PredictedLabelColumnOriginalValueConverter(PredictedLabelColumn = "PredictedLabel"));

let model = pipeline.Train<SentimentData, SentimentPrediction>()

let sentiments = 
    [
        "Please refrain from adding nonsense to Wikipedia."
        "He is a CHEATER, and the article should say that."
    ] 
    |> List.map SentimentData

let predictions = model.Predict(sentiments)

let testData = new TextLoader<SentimentData>(sentimentTestPath, useHeader = true, separator = "tab")

let evaluator = new BinaryClassificationEvaluator()
let metrics = evaluator.Evaluate(model, testData)

let matrix = metrics.ConfusionMatrix