（11）評(píng)估模型，保存模型

Imports System Imports System.Collections Imports System.IO Imports System.Linq Imports Microsoft.Data.DataView Imports Microsoft.ML Imports Microsoft.ML.Data Imports Microsoft.ML.Trainers Imports Microsoft.ML.Transforms.TextModule ProgramPrivate ReadOnly _dataPath As String = Path.Combine(Environment.CurrentDirectory, "data", "yelp_labelled.txt")Private ReadOnly _modelPath As String = Path.Combine(Environment.CurrentDirectory, "data", "model.zip")Public ReadOnly Property DataPath As StringGetReturn _dataPathEnd GetEnd PropertyPublic ReadOnly Property ModelPath As StringGetReturn _modelPathEnd GetEnd PropertySub Main(args As String())'創(chuàng)建上下文ML作業(yè)Dim mlConText As New MLContextDim splitDataView As TrainCatalogBase.TrainTestData = LoadData(mlConText)Dim model As ITransformer = BuildAndTrainModel(mlConText, splitDataView.TrainSet)Evaluate(mlConText, model, splitDataView.TestSet)End SubPublic Function BuildAndTrainModel(mlContext As MLContext, splitTrainSet As IDataView) As ITransformer'將文本列特征化為機(jī)器學(xué)習(xí)算法使用的名為Features的數(shù)值向量的FeaturizeText，再將決策樹算法追加到管道Dim pipleline = mlContext.Transforms.Text.FeaturizeText(outputColumnName:=DefaultColumnNames.Features, inputColumnName:=NameOf(SentimentData.SentimentText)).Append(mlContext.BinaryClassification.Trainers.FastTree(numLeaves:=50, numTrees:=50, minDatapointsInLeaves:=20))Dim model = pipleline.Fit(splitTrainSet)Return modelEnd FunctionPublic Function LoadData(mlContext As MLContext) As TrainCatalogBase.TrainTestData'加載數(shù)據(jù)，將數(shù)據(jù)集分為訓(xùn)練集與測(cè)試集并返回'加載數(shù)據(jù)集通過(guò)基本的數(shù)據(jù)管道dataviewDim dataView As IDataView = mlContext.Data.LoadFromTextFile(Of SentimentData)(_dataPath, hasHeader:=False)'拆分?jǐn)?shù)據(jù)集進(jìn)行模型訓(xùn)練和測(cè)試,20%的測(cè)試集Dim splitDataView As TrainCatalogBase.TrainTestData = mlContext.BinaryClassification.TrainTestSplit(dataView, testFraction:=0.2)Return splitDataViewEnd FunctionPublic Sub Evaluate(mlContext As MLContext, model As ITransformer, splitTestSet As IDataView)'加載測(cè)試數(shù)據(jù)集，創(chuàng)建分類計(jì)算器，評(píng)估模型并創(chuàng)建指標(biāo)，顯示效果指標(biāo)Console.WriteLine("===用測(cè)試數(shù)據(jù)評(píng)估模型正確率===")'返回預(yù)測(cè)Dim predictions As IDataView = model.Transform(splitTestSet)'計(jì)算預(yù)測(cè)模型質(zhì)量指標(biāo)Dim metrics As CalibratedBinaryClassificationMetrics = mlContext.BinaryClassification.Evaluate(predictions, "label")'顯示模型驗(yàn)證指標(biāo)Console.WriteLine("正確率：" & metrics.Accuracy)Console.WriteLine("AUC：" & metrics.Auc)Console.WriteLine("F1Score：" & metrics.F1Score)SaveModelAsFile(mlContext, model)End SubPrivate Sub SaveModelAsFile(mlContext As MLContext, model As ITransformer)Throw New NotImplementedException()Using fs As New FileStream(ModelPath, FileMode.Create, FileAccess.Write, FileShare.Write))mlContext.Model.Save(model, fs)Console.WriteLine("模型存入" & ModelPath)End UsingEnd Sub End Module

總結(jié)

以上是生活随笔為你收集整理的vb.net 2019-机器学习ml.net情绪分析(3)的全部?jī)?nèi)容，希望文章能夠幫你解決所遇到的問題。

如果覺得生活随笔網(wǎng)站內(nèi)容還不錯(cuò)，歡迎將生活随笔推薦給好友。