Add DialoguePrediction

src/Plugins/BotSharp.Plugin.RoutingSpeeder/BotSharp.Plugin.RoutingSpeeder.csproj

@@ -11,4 +11,10 @@
     <ProjectReference Include="..\..\Infrastructure\BotSharp.Abstraction\BotSharp.Abstraction.csproj" />
   </ItemGroup>
 
+  <ItemGroup>
+    <PackageReference Include="FastText.NetWrapper" Version="1.3.0" />
+    <PackageReference Include="TensorFlow.Keras" Version="0.11.2" />
+    <PackageReference Include="TensorFlow.NET" Version="0.110.2" />
+  </ItemGroup>
+
 </Project>

src/Plugins/BotSharp.Plugin.RoutingSpeeder/Providers/DialogueClassifier.cs

@@ -0,0 +1,144 @@
+using System;
+using System.IO;
+using System.Text;
+using System.Collections.Generic;
+using Tensorflow;
+using static Tensorflow.KerasApi;
+using Tensorflow.Keras.Engine;
+using Tensorflow.NumPy;
+using static Tensorflow.Binding;
+using Tensorflow.Keras.Callbacks;
+using System.Text.RegularExpressions;
+using BotSharp.Plugin.RoutingSpeeder.Settings;
+using BotSharp.Abstraction.MLTasks;
+using BotSharp.Plugin.RoutingSpeeder.Providers.Models;
+using Microsoft.Extensions.DependencyInjection;
+using System.Linq;
+using Tensorflow.Keras;
+
+namespace BotSharp.Plugin.RoutingSpeeder.Providers;
+
+public class DialogueClassifier
+{
+    private readonly IServiceProvider _services;
+    Model _model;
+    public Model model => _model;
+    private bool _isModelReady;
+    public bool isModelReady => _isModelReady;
+    private classifierSetting _settings;
+
+    public DialogueClassifier(IServiceProvider services, classifierSetting settings)
+    {
+        _services = services;
+        _settings = settings;
+    }
+
+    private void Reset()
+    {
+        keras.backend.clear_session();
+        _isModelReady = false;
+    }
+
+    private void Build()
+    {
+        if (_isModelReady)
+        {
+            return;
+        }
+
+        var layers = new List<ILayer>
+        {
+            keras.layers.InputLayer((300), name: "Input"),
+            keras.layers.Dense(256, activation:"relu"),
+            keras.layers.Dense(256, activation:"relu"),
+            keras.layers.Dense(_settings.labelMappingDict.Count, activation: keras.activations.Softmax)
+        };
+        _model = keras.Sequential(layers);
+
+#if DEBUG
+        Console.WriteLine();
+        _model.summary();
+#endif
+        _isModelReady = true;
+    }
+
+    private void Fit(NDArray x, NDArray y, TrainingParams trainingParams)
+    {
+        // release more memory
+        var vector = _services.GetRequiredService<ITextEmbedding>();
+        // vector.UnloadModel();
+
+        _model.compile(optimizer: keras.optimizers.Adam(trainingParams.LearningRate),
+            loss: keras.losses.SparseCategoricalCrossentropy(),
+            metrics: new[] { "accuracy" }
+            );
+
+        CallbackParams callback_parameters = new CallbackParams
+        {
+            Model = _model,
+            Epochs = trainingParams.Epochs,
+            Verbose = 1,
+            Steps = 10
+        };
+
+        ICallback earlyStop = new EarlyStopping(callback_parameters, "accuracy");
+
+        var callbacks = new List<ICallback>() { earlyStop };
+
+        var weights = LoadWeights();
+
+        _model.fit(x, y,
+            batch_size: trainingParams.BatchSize,
+            epochs: trainingParams.Epochs,
+            callbacks: callbacks,
+            // validation_split: 0.1f,
+            shuffle: true);
+
+        _model.save_weights(weights);
+
+        _isModelReady = true;
+    }
+
+    public string LoadWeights()
+    {
+        var weightsFile = Path.Combine(_settings.MODEL_DIR, $"wo-dialogue-classifier.h5");
+        if (File.Exists(weightsFile))
+        {
+            _model.load_weights(weightsFile);
+            Console.WriteLine($"Successfully load the weights!");
+        }
+        else
+        {
+            Console.WriteLine("No available weights.");
+        }
+        return weightsFile;
+    }
+
+    public (NDArray x, NDArray y) Vectorize(List<DialoguePredictionModel> items)
+    {
+        var x = np.zeros((items.Count, 300), dtype: np.float32);
+        var y = np.zeros((items.Count, 1), dtype: np.float32);
+
+        var vector = _services.GetRequiredService<ITextEmbedding>();
+
+        for (int i = 0; i < items.Count; i++)
+        {
+            x[i] = vector.GetVector(TextClean(items[i].text));
+            if (_settings.labelMappingDict.ContainsKey(items[i].label))
+            {
+                y[i] = _settings.labelMappingDict[items[i].label];
+            }
+        }
+        return (x, y);
+    }
+    public string TextClean(string text)
+    {
+        // Remove punctuation
+        // Remove digits
+        // To lowercase
+        var processedText = Regex.Replace(text, "[AB0-9]", " ");
+        processedText = string.Join("", processedText.Select(c => char.IsPunctuation(c) ? ' ' : c).ToList());
+        processedText = processedText.Replace("  ", " ").ToLower();
+        return processedText;
+    }
+}

src/Plugins/BotSharp.Plugin.RoutingSpeeder/Providers/Models/DialoguePredictionModel.cs

@@ -0,0 +1,13 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace BotSharp.Plugin.RoutingSpeeder.Providers.Models;
+
+public class DialoguePredictionModel
+{
+    public int Id { get; set; }
+    public string text { get; set; }
+    public string? label { get; set; }
+    public string? prediction { get; set; }
+}

src/Plugins/BotSharp.Plugin.RoutingSpeeder/Providers/fastTextEmbeddingProvider.cs

@@ -0,0 +1,70 @@
+using System;
+using System.Collections.Generic;
+using System.IO;
+using System.Runtime;
+using System.Text;
+using System.Text.RegularExpressions;
+using BotSharp.Abstraction.MLTasks;
+using BotSharp.Plugin.RoutingSpeeder.Settings;
+using FastText.NetWrapper;
+
+namespace BotSharp.Plugin.RoutingSpeeder.Providers;
+
+public class fastTextEmbeddingProvider : ITextEmbedding
+{
+    private FastTextWrapper _fastText;
+    private readonly fastTextSetting _settings;
+
+    public int Dimension
+    {
+        get
+        {
+            if (!_fastText.IsModelReady())
+            {
+                _fastText.LoadModel(_settings.ModelPath);
+            }
+            return _fastText.GetModelDimension();
+        }
+    }
+
+    public fastTextEmbeddingProvider(fastTextSetting settings)
+    {
+        _settings = settings;
+
+    }
+
+    public float[] GetVector(string text)
+    {
+        LoadModel();
+        return _fastText.GetSentenceVector(text);
+    }
+
+    public List<float[]> GetVectors(List<string> texts)
+    {
+        LoadModel();
+        var vectors = new List<float[]>();
+        for (int i = 0; i < texts.Count; i++)
+        {
+            vectors.Add(GetVector(texts[i]));
+        }
+        return vectors;
+    }
+
+    private void LoadModel()
+    {
+        if (_fastText == null)
+        {
+            if (!File.Exists(_settings.ModelPath))
+            {
+                throw new FileNotFoundException($"Can't load pre-trained word vectors from {_settings.ModelPath}.\n Try to download from https://fasttext.cc/docs/en/english-vectors.html.");
+            }
+
+            _fastText = new FastTextWrapper();
+
+            if (!_fastText.IsModelReady())
+            {
+                _fastText.LoadModel(_settings.ModelPath);
+            }
+        }
+    }
+}

src/Plugins/BotSharp.Plugin.RoutingSpeeder/RoutingConversationHook.cs

@@ -1,13 +1,29 @@
 using BotSharp.Abstraction.Conversations;
 using BotSharp.Abstraction.Conversations.Models;
+using BotSharp.Abstraction.MLTasks;
+using Microsoft.Extensions.DependencyInjection;
+using System;
+using System.Linq;
 using System.Threading.Tasks;
+using FastText.NetWrapper;
+using BotSharp.Plugin.RoutingSpeeder.Settings;
 
 namespace BotSharp.Plugin.RoutingSpeeder;
 
 public class RoutingConversationHook: ConversationHookBase
 {
+    private readonly IServiceProvider _services;
+    private routerSpeedSettings _settings;
+    public RoutingConversationHook(IServiceProvider service, routerSpeedSettings settings)
+    {
+        _services = service;
+        _settings = settings;
+    }
     public override async Task BeforeCompletion(RoleDialogModel message)
     {
+        var embedding = _services.GetServices<ITextEmbedding>()
+            .FirstOrDefault(x => x.GetType().FullName.EndsWith(_settings.TextEmbedding));
+
         // Utilize local discriminative model to predict intent
         message.Content = "response content";
         message.StopCompletion = true;

src/Plugins/BotSharp.Plugin.RoutingSpeeder/RoutingSpeederPlugin.cs

@@ -1,5 +1,8 @@
 using BotSharp.Abstraction.Conversations;
+using BotSharp.Abstraction.MLTasks;
 using BotSharp.Abstraction.Plugins;
+using BotSharp.Plugin.RoutingSpeeder.Settings;
+using BotSharp.Plugin.RoutingSpeeder.Providers;
 using Microsoft.Extensions.Configuration;
 using Microsoft.Extensions.DependencyInjection;
 
@@ -9,6 +12,11 @@ public class RoutingSpeederPlugin : IBotSharpPlugin
 {
     public void RegisterDI(IServiceCollection services, IConfiguration config)
     {
+        var settings = new routerSpeedSettings();
+        config.Bind("routerSpeed", settings);
+        services.AddSingleton(x => settings);
+        services.AddSingleton(x => settings.fastText);
         services.AddScoped<IConversationHook, RoutingConversationHook>();
+        services.AddSingleton<ITextEmbedding, fastTextEmbeddingProvider>();
     }
 }

src/Plugins/BotSharp.Plugin.RoutingSpeeder/Settings/classifierSetting.cs

@@ -0,0 +1,21 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace BotSharp.Plugin.RoutingSpeeder.Settings;
+
+public class classifierSetting
+{
+    
+    public Dictionary<string, float> labelMappingDict { get; set; } = new Dictionary<string, float>()
+	{
+		{"goodbye", 0f},
+		{"greeting", 1f},
+		{"other", 2f},
+		{"wo-followup", 3f},
+		{"wo-identifer", 4f},
+		{"wo-scheduler", 5}
+	};
+    public string RAW_DATA_DIR { get; set; } = "C:\\new_wenbocao\\one_brain\\WebStarter\\data\\raw_data";
+    public string MODEL_DIR { get; set; } = "C:\\new_wenbocao\\one_brain\\WebStarter\\data\\models";
+}

src/Plugins/BotSharp.Plugin.RoutingSpeeder/Settings/fastTextSetting.cs

@@ -0,0 +1,7 @@
+
+namespace BotSharp.Plugin.RoutingSpeeder.Settings;
+
+public class fastTextSetting
+{
+    public string ModelPath { get; set; }
+}

src/Plugins/BotSharp.Plugin.RoutingSpeeder/Settings/routerSpeedSettings.cs

@@ -0,0 +1,11 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace BotSharp.Plugin.RoutingSpeeder.Settings;
+
+public class routerSpeedSettings
+{
+    public fastTextSetting fastText { get; set; }
+    public string TextEmbedding { get; set; }
+}

src/Plugins/BotSharp.Plugin.RoutingSpeeder/Settings/trainingParams.cs

@@ -0,0 +1,13 @@
+using System;
+using System.Collections.Generic;
+using System.Text;
+
+namespace BotSharp.Plugin.RoutingSpeeder.Settings;
+
+public class TrainingParams
+{
+    public int ClientId { get; set; }
+    public int Epochs { get; set; } = 10;
+    public int BatchSize { get; set; } = 16;
+    public float LearningRate { get; set; } = 1.0e-4f;
+}