This guide shows you how to train a model in ML.NET and create a Grasshopper component that uses said model. In this example, we predict the square root of a number using the sqrt_dataset.csv-file that contains 100 000 real numbers between 0-1000 and their square root. For installation and setup of ML.NET, follow the official Microsoft guide: https://dotnet.microsoft.com/learn/ml-dotnet/get-started-tutorial/intro

Here, Project, refers to the name of the project you want to add ML.NET to.

1. Create ML.NET model in GH project.

• Right-click the GH project you want to add a ML.NET component to → Add → Machine Learning.
• Select the value regression scenario.
• Upload the sqrt_dataset.csv and select the column titled label as your label. Use the recommended training time and Start training.
• After training is completed, two projects get added to your solution: ProjectML.ConsoleApp and ProjectML.Model.
  • Delete ConsoleApp.

2. Tweak the config to add Model-project to main project.

Add dependency to Project from ProjectML.Model

• Right-click Solution → Properties
  • Project dependencies:
    • Projects: Project → Depends on: ...Model

Add project reference from Project to ProjectML.Model

• Right-click Project → Unload project.

• Right-click unloaded Project → Edit project file.

• Above the first ItemGroup, cirka line 36, add:

  <ItemGroup>
    <ProjectReference Include="" />
  </ItemGroup>

• Right-click ProjectML.Model → Copy full path.

• Inside Include, paste the full path, e.g.:

  <ItemGroup>
    <ProjectReference Include="C:\Users\...\Project\ProjectML.Model\ProjectML.Model.csproj" />
  </ItemGroup>

• Remove everything before (and including the Project-directory):

  <ItemGroup>
    <ProjectReference Include="ProjectML.Model\ProjectML.Model.csproj" />
  </ItemGroup>

• Right-click Project → Reload project with dependencies.

3. Create GH-component

• Create GH-component as usual.
• Inside the new GH-component, add:

using ProjectML.Model;

• To use model, create new ModelInput and get the Score property of Predict-method output of ConsumeModel class like:

ModelInput modelData = new ModelInput()
{
    Number = Convert.ToSingle(numberToSquareRoot)
};
float rootedSingle = ConsumeModel.Predict(modelData).Score;

Regression input and output is by default a float. Use Convert.ToSingle to convert double to float.

Congrats, you now have a machine learning component that kinda predicts the square root of a number!