Not able to re-train model [Multiclassification(AveragedPerceptron)]

[StefanStroescu]

Context

Hello! I am new to ML.Net, I have decided to to try using it in building a dispatcher. Basically I want it to be able to classify text in one of multiple categories. Due to the high volume of data, I want that when a prediction is confirmed by the users wrong to add it to its database(or re-train the model)

I have used AutoML to generate a base model. The algorithm with the best results chose by the AutoML for multiclassification is AveragedPerceptron. I have checked this page in order to make sure that is re-trainable.

I am able to get the first model, but struggling to re-train it.

• First time I have created the model (simulate all the steps generated by AutoML)

// First Phase: Create the model
           
            var mlContext = new MLContext(seed: 1);


            // BuildTrainingPipeline

            // Load Data
            var data = mlContext.Data.LoadFromTextFile<ModelInput>(
                                            path: TRAIN_DATA_FILEPATH,
                                            hasHeader: false,
                                            separatorChar: '\t',
                                            allowQuoting: true,
                                            allowSparse: false);


            // Data process configuration with pipeline data transformations
            var dataProcessPipeline = mlContext.Transforms.Conversion.MapValueToKey("col0", "col0")
                                      .Append(mlContext.Transforms.Text.FeaturizeText("col1_tf", "col1"))
                                      .Append(mlContext.Transforms.CopyColumns("Features", "col1_tf"))
                                      .Append(mlContext.Transforms.NormalizeMinMax("Features", "Features"))
                                      .AppendCacheCheckpoint(mlContext);



            // Set the training algorithm 
            var trainer = mlContext.MulticlassClassification.Trainers.OneVersusAll(mlContext.BinaryClassification.Trainers
                                     .AveragedPerceptron(labelColumnName: "col0", numberOfIterations: 10, featureColumnName: "Features"), labelColumnName: "col0")
                                      .Append(mlContext.Transforms.Conversion.MapKeyToValue("PredictedLabel", "PredictedLabel"));


            IEstimator<ITransformer> trainingPipeline = dataProcessPipeline.Append(trainer);



            // Train and save Model


            // Create model here
            ITransformer firstModel = trainingPipeline.Fit(data);

            // Save the model
            mlContext.Model.Save(firstModel, data.Schema, MODEL_FILEPATH);

• Then I presume I have new data to train the model with

/// Second Phase - Re-training the model

            // New Data
            ModelInput[] ticketData = new ModelInput[]
            {

                  new ModelInput
                  {
                      Col0 = "Category 3",
                      Col1 = "Text to classify 1"
                  },

                  new ModelInput
                  {
                      Col0 = "Category 2",
                      Col1 = "Text to classify 2"
                  },

                  new ModelInput
                  {
                      Col0 = "Category 3",
                      Col1 = "Text to classify 3"
                  },

                  new ModelInput
                  {
                      Col0 = "Category 2",
                      Col1 = "Text to classify 4"
                  },

                  new ModelInput
                  {
                      Col0 = "Category 1",
                      Col1 = "Text to classify 5"
                  },

            };



            // Create MLContext
            MLContext mlContext = new MLContext();

            // Define DataViewSchema  trained model
            DataViewSchema modelSchema;

            // Load trained model
            var trainedModel = mlContext.Model.Load(MODEL_FILEPATH, out modelSchema);

            //Load New Data
            IDataView newData = mlContext.Data.LoadFromEnumerable<ModelInput>(ticketData);
     


           // And here I get stuck. Because I don't know how to retrain the model with new data. 

• Issue

  I have tried to follow the guidance from this topics: here, here or changes suggested here but with no result.

  My issues are due to multiclassification I think, because the trainer is of type EstimatorChain and my model is of type TransformerChain.
  My trainer.Fit doesn't take 2 arguments.

[michaelgsharp]

Are you able to upload the model/sample data you used so we can take a look?

[StefanStroescu]

MultiClassification_AveragePerceptron.zip

Sorry for the late replay. Please see the RetrainModel.cs

This is the data I used initially for training.
DummyData.txt