1 |
76072df0
|
Roman Kalivoda
|
//
|
2 |
|
|
// Author: Roman Kalivoda
|
3 |
|
|
//
|
4 |
|
|
|
5 |
|
|
using System;
|
6 |
|
|
using System.Collections.Generic;
|
7 |
|
|
using log4net;
|
8 |
|
|
using Microsoft.ML;
|
9 |
|
|
|
10 |
|
|
namespace ServerApp.Predictor
|
11 |
|
|
{
|
12 |
|
|
abstract class AbstractClassificationPredictor : IPredictor
|
13 |
|
|
{
|
14 |
|
|
private static readonly ILog _log = LogManager.GetLogger(typeof(AbstractClassificationPredictor));
|
15 |
|
|
|
16 |
|
|
/// <summary>
|
17 |
|
|
/// Context of the ML.NET framework.
|
18 |
|
|
/// </summary>
|
19 |
|
|
protected MLContext _mlContext;
|
20 |
|
|
|
21 |
|
|
/// <summary>
|
22 |
|
|
/// Model instance
|
23 |
|
|
/// </summary>
|
24 |
|
|
protected ITransformer _trainedModel;
|
25 |
|
|
|
26 |
|
|
protected PredictionEngine<ModelInput, ModelOutput> _predictionEngine;
|
27 |
|
|
|
28 |
|
|
protected IDataView _trainingDataView;
|
29 |
|
|
|
30 |
|
|
public void Evaluate(IEnumerable<ModelInput> modelInputs)
|
31 |
|
|
{
|
32 |
|
|
var testDataView = this._mlContext.Data.LoadFromEnumerable(modelInputs);
|
33 |
|
|
var data = _trainedModel.Transform(testDataView);
|
34 |
|
|
var testMetrics = _mlContext.MulticlassClassification.Evaluate(data);
|
35 |
|
|
|
36 |
|
|
Console.WriteLine($"*************************************************************************************************************");
|
37 |
|
|
Console.WriteLine($"* Metrics for Multi-class Classification model - Test Data ");
|
38 |
|
|
Console.WriteLine($"*------------------------------------------------------------------------------------------------------------");
|
39 |
|
|
Console.WriteLine($"* MicroAccuracy: {testMetrics.MicroAccuracy:0.###}");
|
40 |
|
|
Console.WriteLine($"* MacroAccuracy: {testMetrics.MacroAccuracy:0.###}");
|
41 |
|
|
Console.WriteLine($"* LogLoss: {testMetrics.LogLoss:#.###}");
|
42 |
|
|
Console.WriteLine($"* LogLossReduction: {testMetrics.LogLossReduction:#.###}");
|
43 |
|
|
Console.WriteLine($"* Confusion Matrix: {testMetrics.ConfusionMatrix.GetFormattedConfusionTable()}");
|
44 |
|
|
Console.WriteLine($"*************************************************************************************************************");
|
45 |
|
|
}
|
46 |
|
|
|
47 |
|
|
public abstract void Fit(IEnumerable<ModelInput> trainInput);
|
48 |
|
|
|
49 |
|
|
public void Load(string filename)
|
50 |
|
|
{
|
51 |
|
|
DataViewSchema modelSchema;
|
52 |
|
|
this._trainedModel = _mlContext.Model.Load(filename, out modelSchema);
|
53 |
|
|
// TODO check if the loaded model has valid input and output schema
|
54 |
|
|
this._predictionEngine = _mlContext.Model.CreatePredictionEngine<ModelInput, ModelOutput>(this._trainedModel);
|
55 |
|
|
}
|
56 |
|
|
|
57 |
|
|
public string Predict(ModelInput input)
|
58 |
|
|
{
|
59 |
|
|
_log.Debug($"Predicting for input: {input}");
|
60 |
|
|
return this._predictionEngine.Predict(input).PredictedLabel;
|
61 |
|
|
}
|
62 |
|
|
|
63 |
|
|
public void Save(string filename)
|
64 |
|
|
{
|
65 |
|
|
if (this._trainingDataView is null)
|
66 |
|
|
{
|
67 |
|
|
throw new NullReferenceException("DataView is not set.");
|
68 |
|
|
}
|
69 |
|
|
if (this._trainedModel is null)
|
70 |
|
|
{
|
71 |
|
|
throw new NullReferenceException("Trained model instance does not exist. This predictor has not been trained yet.");
|
72 |
|
|
}
|
73 |
|
|
if (filename is null)
|
74 |
|
|
{
|
75 |
|
|
throw new ArgumentNullException(nameof(filename));
|
76 |
|
|
}
|
77 |
|
|
this._mlContext.Model.Save(this._trainedModel, this._trainingDataView.Schema, filename);
|
78 |
|
|
}
|
79 |
|
|
}
|
80 |
|
|
}
|