[C#] AccessViolationException encountered using Microsoft.ML.LightGbm v4.3.0 NuGet package

[boneatjp]

Description

Having

System.AccessViolationException
HResult=0x80004003
Message=Attempted to read or write protected memory. This is often an indication that other memory is corrupt.
Source=<例外のソースを評価できません>
スタック トレース:
<例外のスタック トレースを評価できません>

Reproducible example

Installing Version 4.3.0

Environment info

Using :
Microsoft.ML Version 3.0.1
Microsoft.ML.AutoML Version 0.21.1
Microsoft.ML.CpuMath Version 3.0.1
Microsoft.ML.DataView Version 3.0.1
Microsoft.ML.FastTree Version 3.0.1
Microsoft.ML.LightGbm Version 3.0.1
Microsoft.ML.Mkl.Compreonents Version 3.0.1
Microsoft.ML.Mkl.Redist Version 3.0.1

Under OS Version:
Windows 11 Pro Version 23H2
OS Build 22631.3007

LightGBM version or commit hash:
Without Installing LightGBM Version 4.3.0, it should be Version 3.3.5

Command(s) you used to install LightGBM
After installing LightGBM Version 4.3.0,

Having problem 'System.AccessViolationException' in Microsoft.ML.LightGbm.dll

Am I suppose to not use it in this case?

Additional Comments

[jameslamb]

Thanks for using LightGBM.

Without Installing LightGBM Version 4.3.0, it should be Version 3.3.5

I don't understand this statement, could you please elaborate or explain it a different way?

Beyond that, can you tell us more about how you're using LightGBM? For example, are you able to provide a minimal code sample that results in the same error?

[boneatjp]

I'm using the packages with NuGet in Visual Studio Community 2022 mentioned for my app. If I try to use LightGBM with version 4.0.0 or later on NuGet in VS Comuunity 2022, I'm having the error I mentioned. If I don't install LightGBM Version 4.0.0 or later on NuGet, my app works fine. If I install LightGBM Version 4.0.0 or later on NuGet, my app ends up with the error.

So, I thought there might have been change between version 3.3.5 and version 4.0.0 or later. If you need minimal code sample, I'll try to write it this weekend.

[jameslamb]

If I install LightGBM Version 4.0.0 or later on NuGet, my app ends up with the error.

Yes, LightGBM 4.0.0 contains 2+ years of changes relative to v3.3.5. Without more information, we won't be able to help.

[boneatjp]

AutoMLSample.zip

This is the small sample project which gets the error I mentioned. If you uninstall LightGBM Version 4.3.0 using NuGet Package Manager, it should end without getting any error.

Can you help about this error I'm having?

[jameslamb]

I'm sorry, but I don't open untrusted zip files from the internet. Are you able to reduce this to a small code sample that demonstrates the problem, that can be shared in plaintext or as a link to a source code repository? We don't have much .NET experience here but still might be able to help.

[boneatjp]

I understand that you don't trust zip files from a stranger. I'm not sure if I could write a sample without Visual Studio. I hope you can use Visual Studio. I'm using Visual Studio Community 2022 and C#.

Make a console app project and use NuGet Package Manager to install all 8 packages I've mentioned at the first time. Edit Program.cs:

using Microsoft.ML;
using Microsoft.ML.AutoML;
using Microsoft.ML.Data;
using static Microsoft.ML.DataOperationsCatalog;

MLContext mlContext;
string sampleData = "taxi-fare-full.csv";
mlContext = new MLContext();

// Infer column information
ColumnInferenceResults columnInference = mlContext.Auto()
                                                .InferColumns(sampleData,
                                                    labelColumnName: "fare_amount",
                                                    groupColumns: false);
// Create text loader
TextLoader loader = mlContext.Data.CreateTextLoader(columnInference.TextLoaderOptions);

// Load data into IDataView
IDataView data = loader.Load(sampleData);

TrainTestData trainValidationData = mlContext.Data.TrainTestSplit(data, testFraction: 0.2);

SweepablePipeline pipeline = mlContext.Auto()
                    .Featurizer(data, columnInformation: columnInference.ColumnInformation)
                        .Append(mlContext.Auto()
                            .Regression(labelColumnName: columnInference.ColumnInformation.LabelColumnName));

AutoMLExperiment experiment = mlContext.Auto().CreateExperiment();

var regressionMetric = RegressionMetric.RootMeanSquaredError;
experiment
    .SetPipeline(pipeline)
    .SetRegressionMetric(regressionMetric,
                            labelColumn: columnInference.ColumnInformation.LabelColumnName)
    .SetTrainingTimeInSeconds(100)  // Training time in sec
    .SetDataset(trainValidationData);

// Log experiment trials
mlContext.Log += (_, e) => {
    if (e.Source.Equals("AutoMLExperiment"))
    {
        Console.WriteLine(e.RawMessage);
    }
};

TrialResult experimentResults = await experiment.RunAsync();

Download the taxi-fare-full.csv and set it to be copied.

You should be able to run the sample without any error. Then Install LightGBM by NuGet Package Manager and try to run the sample. You should see the error.

[boneatjp]

I'm not sure if I gave enough information to check the problem.

[jameslamb]

I'm not sure if I gave enough information to check the problem.

I don't know, sorry. It does look more likely to help someone find the root cause than your original post.

But I personally have never compiled or run a C# program and don't have easy access to a Windows development environment. I'll try to help with this when I can, otherwise hopefully someone else will be able to help you.

---

I've also changed the title of this issue to make it more specific, please do modify it if I've misunderstood and misstated the problem.

[boneatjp]

Thank you so much for your time to check my problem. I think the new title is fine.

[winest]

@jameslamb
I can help to provide a sample code to reproduce the issue

using Microsoft.ML;
using Microsoft.ML.Data;
using Microsoft.ML.Trainers.LightGbm;

namespace LambdaRankExample
{
    class Program
    {
        static void Main(string[] args)
        {
            // Create MLContext
            var mlContext = new MLContext();

            // Define training data
            var trainData = new List<RankingData>
            {
                // First query results
                new RankingData { Features = new float[] { 1.0f, 0.1f, 0.2f }, Label = 4, GroupId = 0 },
                new RankingData { Features = new float[] { 0.7f, 0.4f, 0.5f }, Label = 3, GroupId = 0 },
                new RankingData { Features = new float[] { 0.6f, 0.8f, 0.9f }, Label = 1, GroupId = 0 },
                // Second query results
                new RankingData { Features = new float[] { 0.8f, 0.2f, 0.3f }, Label = 3, GroupId = 1 },
                new RankingData { Features = new float[] { 0.5f, 0.6f, 0.7f }, Label = 2, GroupId = 1 },
                new RankingData { Features = new float[] { 0.3f, 0.4f, 0.5f }, Label = 1, GroupId = 1 },
                new RankingData { Features = new float[] { 0.2f, 0.1f, 0.4f }, Label = 1, GroupId = 1 },
                new RankingData { Features = new float[] { 0.9f, 0.7f, 0.6f }, Label = 0, GroupId = 1 }
            };

            // Load training data
            var trainDataView = mlContext.Data.LoadFromEnumerable(trainData);

            // Define LightGBM ranking trainer options
            var options = new LightGbmRankingTrainer.Options
            {
                LabelColumnName = nameof(RankingData.Label),
                FeatureColumnName = nameof(RankingData.Features),
                RowGroupColumnName = nameof(RankingData.GroupId),
                NumberOfLeaves = 7,
                LearningRate = 0.1,
                NumberOfIterations = 100,
                //MinimumExampleCountPerGroup = 1,
                //MinimumExampleCountPerLeaf = 20,
                //MaximumBinCountPerFeature = 1000
            };

            // Define the training pipeline
            var pipeline = mlContext.Ranking.Trainers.LightGbm(options);

            // Train the model
            var model = pipeline.Fit(trainDataView);

            // Define test data
            var testData = new List<RankingData>
            {
                new RankingData { Features = new float[] { 0.8f, 0.2f, 0.3f } },
                new RankingData { Features = new float[] { 0.5f, 0.6f, 0.7f } }
            };

            // Load test data
            var testDataView = mlContext.Data.LoadFromEnumerable(testData);

            // Perform predictions
            var predictions = model.Transform(testDataView);

            // Extract scores
            var scores = mlContext.Data.CreateEnumerable<RankingPrediction>(predictions, reuseRowObject: false);

            // Output predictions
            foreach (var score in scores)
            {
                Console.WriteLine($"Prediction: {score.Score}");
            }
        }
    }

    public class RankingData
    {
        [VectorType(3)]
        public float[] Features { get; set; }
        public float Label { get; set; }
        public uint GroupId { get; set; }
    }

    public class RankingPrediction
    {
        public float Score { get; set; }
    }
}

Which will return at runtime

Fatal error. System.AccessViolationException: Attempted to read or write protected memory. This is often an indication that other memory is corrupt.
Repeat 2 times:
--------------------------------
   at Microsoft.ML.Trainers.LightGbm.WrappedLightGbmInterface.DatasetSetField(SafeDataSetHandle, System.String, IntPtr, Int32, CApiDType)
--------------------------------
   at Microsoft.ML.Trainers.LightGbm.Dataset.SetLabel(Single[])
   at Microsoft.ML.Trainers.LightGbm.Dataset..ctor

I am using Visual Studio 2022 Enterprise 64bit version 17.12.3

microsoft.ml\4.0.0
lightgbm\4.5.0
microsoft.ml.onnxruntime\1.20.1
microsoft.ml.lightgbm\4.0.0
microsoft.ml.tokenizers\1.0.0
microsoft.ml.onnxtransformer\4.0.0\