InvalidCastException

[paulh-rnd]

I'm getting an InvalidCastException on the following line, when using the code unmodified from this documentation:

        LinearRegressionModelParameters originalModelParameters =
            ((ISingleFeaturePredictionTransformer<object>)trainedModel).Model as LinearRegressionModelParameters;

The full exception message is:

System.InvalidCastException: 'Unable to cast object of type 'Microsoft.ML.Data.TransformerChain`1[Microsoft.ML.ITransformer]' to type 'Microsoft.ML.ISingleFeaturePredictionTransformer`1[System.Object]'.'

I'm using Microsoft.ML v1.4.0 in a .Net Core 2.1 console app.

---

Document Details

⚠ Do not edit this section. It is required for docs.microsoft.com ➟ GitHub issue linking.

• ID: 0366cc46-bc07-44fd-72d9-d37b40bd5eca
• Version Independent ID: 5c20d041-f946-d7c8-b0b5-7bee96e90282
• Content: Re-train a model - ML.NET
• Content Source: docs/machine-learning/how-to-guides/retrain-model-ml-net.md
• Product: dotnet-ml
• GitHub Login: @luisquintanilla
• Microsoft Alias: luquinta

[luisquintanilla]

Hi @paulh-rnd ,

Sorry to hear you're running into issues. I was not able to replicate the exception mentioned in this issue. One thing I would mention is, not all trainers can be retrained. In the article there is a list of the ones you can retrain. Here is working code that aligns with the samples in the how-to guide.

.NET SDK Version: 2.1
ML.NET Version: 1.4.0

Data Schema

class HousingData
{
    public float Size { get; set; }
    [VectorType(3)]
    public float[] HistoricalPrices { get; set; }
    [ColumnName("Label")]
    public float CurrentPrice { get; set; }
}

Main Program

static void Main(string[] args)
{
    // Create MLContext
    MLContext mlContext = new MLContext();

    // Train and Save Model
    TrainAndSaveModel(mlContext);

    // Retrain model
    RetrainModel(mlContext);
}

TrainAndSaveModel Method

static void TrainAndSaveModel(MLContext mlContext)
{
    HousingData[] housingData = new HousingData[]
    {
        new HousingData
        {
            Size = 600f,
            HistoricalPrices = new float[] { 100000f, 125000f, 122000f },
            CurrentPrice = 170000f
        },
        new HousingData
        {
            Size = 1000f,
            HistoricalPrices = new float[] { 200000f, 250000f, 230000f },
            CurrentPrice = 225000f
        },
        new HousingData
        {
            Size = 1000f,
            HistoricalPrices = new float[] { 126000f, 130000f, 200000f },
            CurrentPrice = 195000f
        }
    };
    
    // Load Data
    IDataView data = mlContext.Data.LoadFromEnumerable<HousingData>(housingData);

    // Define data preparation estimator
    IEstimator<ITransformer> dataPrepPipeline =
        mlContext.Transforms.Concatenate("Features", new string[] { "Size", "HistoricalPrices" })
            .Append(mlContext.Transforms.NormalizeMinMax("Features"));

    // Fir data preparation estimator to data
    ITransformer preprocessor = dataPrepPipeline.Fit(data);

    // Define model training pipeline
    var trainingPipeline = mlContext.Regression.Trainers.OnlineGradientDescent();

    // Preprocess data
    IDataView preprocessedData = preprocessor.Transform(data);

    // Train model
    ITransformer trainedModel = trainingPipeline.Fit(preprocessedData);

    // Save data preparation pipeline
    mlContext.Model.Save(preprocessor, data.Schema, "data_preparation_pipeline.zip");

    // Save trained model
    mlContext.Model.Save(trainedModel, preprocessedData.Schema, "ogd_model.zip");
}

RetrainModel Method

static void RetrainModel(MLContext mlContext)
{
    // Load saved data prep pipeline
    ITransformer dataPrepPipeline = mlContext.Model.Load("data_preparation_pipeline.zip", out DataViewSchema dataPrepPipelineSchema);

    // Load trained model
    ITransformer trainedModel = mlContext.Model.Load("ogd_model.zip", out DataViewSchema modelSchema);

    // Extract trained model parameters
    LinearRegressionModelParameters originalModelParameters =
        ((ISingleFeaturePredictionTransformer<object>)trainedModel).Model as LinearRegressionModelParameters;

    // Create new data
    HousingData[] housingData = new HousingData[]
    {
        new HousingData
        {
            Size = 850f,
            HistoricalPrices = new float[] { 150000f,175000f,210000f },
            CurrentPrice = 205000f
        },
        new HousingData
        {
            Size = 900f,
            HistoricalPrices = new float[] { 155000f, 190000f, 220000f },
            CurrentPrice = 210000f
        },
        new HousingData
        {
            Size = 550f,
            HistoricalPrices = new float[] { 99000f, 98000f, 130000f },
            CurrentPrice = 180000f
        }
    };

    // Load new data into an IDataView
    IDataView newData = mlContext.Data.LoadFromEnumerable<HousingData>(housingData);

    // Apply data prep transforms to new data
    IDataView transformedNewData = dataPrepPipeline.Transform(newData);

    // Use the new data to train a new model
    var retrainedModel = mlContext.Regression.Trainers.OnlineGradientDescent().Fit(transformedNewData, originalModelParameters);

    // Extract Model Parameters of retrained model
    LinearRegressionModelParameters retrainedModelParameters = retrainedModel.Model as LinearRegressionModelParameters;

    // Inspect change in weights after retraining
    var weightDiffs =
        originalModelParameters.Weights.Zip(
            retrainedModelParameters.Weights, (original, retrained) => original - retrained).ToArray();

    Console.WriteLine("Original | Retrained | Difference");
    for (int i = 0; i < weightDiffs.Count(); i++)
    {
        Console.WriteLine($"{originalModelParameters.Weights[i]} | {retrainedModelParameters.Weights[i]} | {weightDiffs[i]}");
    }
}

Running this program should yield output similar to that below:

Original | Retrained | Difference
33039.86 | 56293.76 | -23253.9
29099.14 | 49586.03 | -20486.89
28938.38 | 48609.23 | -19670.85
30484.02 | 53745.43 | -23261.41

[maolin-kooboo]

Hi @paulh-rnd ,

Sorry to hear you're running into issues. I was not able to replicate the exception mentioned in this issue. One thing I would mention is, not all trainers can be retrained. In the article there is a list of the ones you can retrain. Here is working code that aligns with the samples in the how-to guide.

.NET SDK Version: 2.1
ML.NET Version: 1.4.0

Data Schema

class HousingData
{
    public float Size { get; set; }
    [VectorType(3)]
    public float[] HistoricalPrices { get; set; }
    [ColumnName("Label")]
    public float CurrentPrice { get; set; }
}

Main Program

static void Main(string[] args)
{
    // Create MLContext
    MLContext mlContext = new MLContext();

    // Train and Save Model
    TrainAndSaveModel(mlContext);

    // Retrain model
    RetrainModel(mlContext);
}

TrainAndSaveModel Method

static void TrainAndSaveModel(MLContext mlContext)
{
    HousingData[] housingData = new HousingData[]
    {
        new HousingData
        {
            Size = 600f,
            HistoricalPrices = new float[] { 100000f, 125000f, 122000f },
            CurrentPrice = 170000f
        },
        new HousingData
        {
            Size = 1000f,
            HistoricalPrices = new float[] { 200000f, 250000f, 230000f },
            CurrentPrice = 225000f
        },
        new HousingData
        {
            Size = 1000f,
            HistoricalPrices = new float[] { 126000f, 130000f, 200000f },
            CurrentPrice = 195000f
        }
    };
    
    // Load Data
    IDataView data = mlContext.Data.LoadFromEnumerable<HousingData>(housingData);

    // Define data preparation estimator
    IEstimator<ITransformer> dataPrepPipeline =
        mlContext.Transforms.Concatenate("Features", new string[] { "Size", "HistoricalPrices" })
            .Append(mlContext.Transforms.NormalizeMinMax("Features"));

    // Fir data preparation estimator to data
    ITransformer preprocessor = dataPrepPipeline.Fit(data);

    // Define model training pipeline
    var trainingPipeline = mlContext.Regression.Trainers.OnlineGradientDescent();

    // Preprocess data
    IDataView preprocessedData = preprocessor.Transform(data);

    // Train model
    ITransformer trainedModel = trainingPipeline.Fit(preprocessedData);

    // Save data preparation pipeline
    mlContext.Model.Save(preprocessor, data.Schema, "data_preparation_pipeline.zip");

    // Save trained model
    mlContext.Model.Save(trainedModel, preprocessedData.Schema, "ogd_model.zip");
}

RetrainModel Method

static void RetrainModel(MLContext mlContext)
{
    // Load saved data prep pipeline
    ITransformer dataPrepPipeline = mlContext.Model.Load("data_preparation_pipeline.zip", out DataViewSchema dataPrepPipelineSchema);

    // Load trained model
    ITransformer trainedModel = mlContext.Model.Load("ogd_model.zip", out DataViewSchema modelSchema);

    // Extract trained model parameters
    LinearRegressionModelParameters originalModelParameters =
        ((ISingleFeaturePredictionTransformer<object>)trainedModel).Model as LinearRegressionModelParameters;

    // Create new data
    HousingData[] housingData = new HousingData[]
    {
        new HousingData
        {
            Size = 850f,
            HistoricalPrices = new float[] { 150000f,175000f,210000f },
            CurrentPrice = 205000f
        },
        new HousingData
        {
            Size = 900f,
            HistoricalPrices = new float[] { 155000f, 190000f, 220000f },
            CurrentPrice = 210000f
        },
        new HousingData
        {
            Size = 550f,
            HistoricalPrices = new float[] { 99000f, 98000f, 130000f },
            CurrentPrice = 180000f
        }
    };

    // Load new data into an IDataView
    IDataView newData = mlContext.Data.LoadFromEnumerable<HousingData>(housingData);

    // Apply data prep transforms to new data
    IDataView transformedNewData = dataPrepPipeline.Transform(newData);

    // Use the new data to train a new model
    var retrainedModel = mlContext.Regression.Trainers.OnlineGradientDescent().Fit(transformedNewData, originalModelParameters);

    // Extract Model Parameters of retrained model
    LinearRegressionModelParameters retrainedModelParameters = retrainedModel.Model as LinearRegressionModelParameters;

    // Inspect change in weights after retraining
    var weightDiffs =
        originalModelParameters.Weights.Zip(
            retrainedModelParameters.Weights, (original, retrained) => original - retrained).ToArray();

    Console.WriteLine("Original | Retrained | Difference");
    for (int i = 0; i < weightDiffs.Count(); i++)
    {
        Console.WriteLine($"{originalModelParameters.Weights[i]} | {retrainedModelParameters.Weights[i]} | {weightDiffs[i]}");
    }
}

Running this program should yield output similar to that below:

Original | Retrained | Difference
33039.86 | 56293.76 | -23253.9
29099.14 | 49586.03 | -20486.89
28938.38 | 48609.23 | -19670.85
30484.02 | 53745.43 | -23261.41

Hi @luisquintanilla ,
how can I use this model, when I create Prediction Engine from it, throws exception

mlContext.Model.CreatePredictionEngine<HousingData, HousingDataPrediction>(trainedModel);

[luisquintanilla]

Hi @maolin-kooboo,

I suspect you're having this issue because you called your column "Feature" instead of "Features".

[maolin-kooboo]

Hi @luisquintanilla ,
I was using the sample you provided above and added one line in the TrainAndSaveModel method after Train model, and I think the column is called 'Features' on the line 'Define data preparation estimator'.

static void TrainAndSaveModel(MLContext mlContext)
{
    // ....

    // Train model
    ITransformer trainedModel = trainingPipeline.Fit(preprocessedData);

    var engine = mlContext.Model.CreatePredictionEngine<HousingData, HousingDataPrediction>(trainedModel);

    // Save data preparation pipeline
    mlContext.Model.Save(preprocessor, data.Schema, "data_preparation_pipeline.zip");

    // Save trained model
    mlContext.Model.Save(trainedModel, preprocessedData.Schema, "ogd_model.zip");
}

        class HousingDataPrediction
        {
            float PredictedLabel { get; set; }

            float Score { get; set; }
        }

[luisquintanilla]

I see what you mean @maolin-kooboo

The reason that's happening is because the CreatePredictionEngine uses the classes HousingData and HousingDataPrediction to determine the schema of the input / output columns. Since the model is trained on the preprocessed data, the input schema no longer matches HousingData. Therefore, you have to create a new class that defines this new input schema.

class PreprocessedHousingData
{
    [VectorType(4)]
    public float[] Features { get; set; }
}

This new schema contains the Features column created as part of our dataPrepPipeline which is a vector of size 4 (Size + 3 Historical Prices).

Once you have this new class defining the schema for your preprocessed data, all you have to do is replace the input schema in the CreatePredictionEngine method with PreprocessedHousingData.

var engine = mlContext.Model.CreatePredictionEngine<PreprocessedHousingData, HousingDataPrediction>(trainedModel);

[maolin-kooboo]

@luisquintanilla , Thanks, it is working.