PlainML.Infrastructure 0.1.192

PlainML

This library should simplify the tracking process of Machine-Learning (ML) training tasks. When you train a ML-model you can save parameters and metrics of the training attempts (named as "runs"). Also you can store artifacts and deploy it to a deploymenttarget like "dev" or "prod". On this way you can access them on an client application the deployed model.

Features

• Experimenttracking (Runs with parameters, metrics and artifacts)
• Integration of trainingprocesses
• Deployment to different targets (dev, production)
• Manage artifacts (deployment, caching)
• Visualization

Getting Started

1. Install VSCode
2. Install C# extensions for VSCode
3. Create project

   mkdir PlainMLExample
   cd PlainMLExample
   dotnet new console
   

4. Install nuget package to project

   dotnet add package PlainML
   dotnet add package PlainML.Infrastructure
   

5. Add usings on the top of program.cs-File

   using Microsoft.Extensions.DependencyInjection;
   using PlainML;
   using PlainML.Entities;
   using PlainML.Infrastructure;
   

6. Use dependency injection to configure services

   var _provider = new ServiceCollection()
       .UsePlainMLSqLite()  // other providers like SQL-Server are avaiable
       .UseArtifactStorageFilesystem() // other providers like SQL-Server are avaiable
       .AddTransient<PlainMLService>()
       .BuildServiceProvider();
   

7. Apply migration when init the database the first time or a new major version is avaiable

   var s = new PlainMLService(dbContextFactory, artifactStorage);
   await s.Migrate();
   

8. Use code

   const string experimentName = "TestExperiment";
   const string artifactsPath = "./Artifacts";
   
   // Create database
   PlainMLService s = _provider.GetRequiredService<PlainMLService>();
   await s.Migrate();
   
   // Train model
   int rundId = await s.StartRun(experimentName);
   float metricValue = await TrainModel(artifactsPath);
   await s.EndRun(
       rundId,
       parameters: new[] { new Parameter(){ Name = "Parameter1", Value = 1.123f } },
       parameters_StringType: new[] { new Parameter_StringType(){ Name = "Trainers", Value = "LightGbm, OneVersusAllTrainer" } },
       metrics: new[] { new Metric(){ Name = "MicroAccuracy", Value = metricValue } },
       artifactsPath);
   
   // Deploy model
   await s.DeployRun(rundId);
   
   // Use model
   var deployedRun = await s.GetDeployedRun(experimentName) ?? throw new NullReferenceException();
   await s.DownloadArtifacts(deployedRun.Id, "./DownloadedArtifacts");
   
   Console.WriteLine("Artifacts of run in ./DownloadedArtifacts:");
   foreach (var item in Directory.EnumerateFiles("./DownloadedArtifacts"))
   {
       Console.WriteLine(item);
   }
   
   async static Task<float> TrainModel(string artifactsPath)
   {
       Console.WriteLine("Training...");
       await Task.Delay(100);
   
       if (Directory.Exists(artifactsPath))
       {
           Console.WriteLine("Directory exists. Delete it!");
           Directory.Delete(artifactsPath, true);
       }
   
       Directory.CreateDirectory(artifactsPath);
       await File.WriteAllTextAsync(Path.Combine(artifactsPath, "TestFile.bin"), "0011010101001");
   
       return 0.1f;
   }
   

Examples

• src/PlainML.Samples/Program.cs

NuGet Packages

• https://www.nuget.org/packages/PlainML/
• https://www.nuget.org/packages/PlainML.Infrastructure/
• https://www.nuget.org/packages/PlainML.Core/

Roadmap

• Develop unstable version 0.1.*
• Create initial stable version 1.0.0
• ML.Net integration
• Create docs
• Rest-interface
• Web-interface for visualization and Manage Experiments (run-table, graphs,...)