Getting Started

Getting Started with AiDotNet

This guide will help you get up and running with AiDotNet in minutes.

Prerequisites

• .NET 8.0 SDK or later
• Visual Studio 2022, VS Code, or JetBrains Rider
• (Optional) CUDA-capable GPU for accelerated training

Installation

Via NuGet Package Manager

dotnet add package AiDotNet

Via Package Manager Console (Visual Studio)

Install-Package AiDotNet

Via PackageReference

Add to your .csproj file:

<PackageReference Include="AiDotNet" Version="*" />

For more detailed installation options, see the Installation Guide.

Architecture: The Facade Pattern

AiDotNet uses a facade pattern to provide a simple, unified API for all AI/ML tasks. You only need to learn two main classes:

Class	Purpose
AiModelBuilder<T, TInput, TOutput>	Build and train any model
AiModelResult<T, TInput, TOutput>	Access trained model and perform inference

This design:

• Hides internal complexity
• Provides a consistent API across all model types
• Makes it easy to switch between algorithms
• Ensures best practices are followed automatically

Your First Model

All AiDotNet functionality goes through the AiModelBuilder. Here's a complete example:

Example 1: Simple Classification

using AiDotNet;

// Training data: features and labels
var features = new double[,]
{
    { 5.1, 3.5, 1.4, 0.2 },  // Setosa
    { 4.9, 3.0, 1.4, 0.2 },  // Setosa
    { 7.0, 3.2, 4.7, 1.4 },  // Versicolor
    { 6.4, 3.2, 4.5, 1.5 },  // Versicolor
    { 6.3, 3.3, 6.0, 2.5 },  // Virginica
    { 5.8, 2.7, 5.1, 1.9 }   // Virginica
};

var labels = new int[] { 0, 0, 1, 1, 2, 2 };

// Build and train the model using the facade
var result = await new AiModelBuilder<double, double[], int>()
    .WithTrainingData(features, labels)
    .WithClassifier(ClassifierType.RandomForest)
    .BuildAsync();

// Make predictions through the result object
var prediction = result.Predict(new double[] { 5.5, 2.5, 4.0, 1.3 });
Console.WriteLine($"Predicted class: {prediction}");

// View metrics
Console.WriteLine($"Accuracy: {result.Metrics.Accuracy:P2}");

Example 2: Neural Network

using AiDotNet;

// Build a neural network for MNIST digit classification
var result = await new AiModelBuilder<float, Tensor<float>, int>()
    .WithTrainingData(trainImages, trainLabels)
    .WithNeuralNetwork(nn => nn
        .AddFlattenLayer()
        .AddDenseLayer(128, Activation.ReLU)
        .AddDropout(0.2)
        .AddDenseLayer(64, Activation.ReLU)
        .AddOutputLayer(10, Activation.Softmax))
    .WithOptimizer(OptimizerType.Adam, learningRate: 0.001)
    .WithLossFunction(LossType.CrossEntropy)
    .WithEpochs(10)
    .WithBatchSize(32)
    .BuildAsync();

// Inference through the result object
var prediction = result.Predict(testImage);
Console.WriteLine($"Predicted digit: {prediction}");
Console.WriteLine($"Test Accuracy: {result.Metrics.Accuracy:P2}");

Example 3: Regression

using AiDotNet;

// House price prediction
var features = new double[,]
{
    { 1400, 3, 2 },  // sqft, bedrooms, bathrooms
    { 1600, 3, 2 },
    { 1700, 3, 2 },
    { 1875, 4, 3 },
    { 2350, 4, 3 }
};

var prices = new double[] { 245000, 312000, 279000, 308000, 450000 };

var result = await new AiModelBuilder<double, double[], double>()
    .WithTrainingData(features, prices)
    .WithRegressor(RegressorType.GradientBoosting)
    .BuildAsync();

// Inference through the result object
var predictedPrice = result.Predict(new double[] { 2000, 4, 3 });
Console.WriteLine($"Predicted price: ${predictedPrice:N0}");

Core Concepts

AiModelBuilder

The AiModelBuilder<T, TInput, TOutput> is the only entry point for building models:

• T - The numeric type (usually float or double)
• TInput - Input data type (e.g., double[], Tensor<T>, string)
• TOutput - Output type (e.g., int for classification, double for regression)

Configuration Options

var result = await new AiModelBuilder<double, double[], int>()
    .WithTrainingData(features, labels)
    .WithClassifier(ClassifierType.SVM)

    // Data preprocessing
    .ConfigurePreprocessing(p => p
        .Normalize()
        .HandleMissingValues())

    // Cross-validation
    .WithCrossValidation(folds: 5)

    // Train/test split
    .WithTestSplit(0.2)  // 20% for testing

    // Telemetry and logging
    .ConfigureTelemetry(t => t.Enabled = true)

    .BuildAsync();

AiModelResult - Accessing Results

The AiModelResult object provides all inference capabilities and metrics:

// Make predictions (all inference goes through the result object)
var prediction = result.Predict(newSample);
var batchPredictions = result.Predict(samples);

// Get probabilities (for classification)
var probabilities = result.PredictProbabilities(newSample);

// Training metrics
var accuracy = result.Metrics.Accuracy;
var precision = result.Metrics.Precision;
var recall = result.Metrics.Recall;
var f1Score = result.Metrics.F1Score;

// For regression
var mse = result.Metrics.MeanSquaredError;
var mae = result.Metrics.MeanAbsoluteError;
var r2 = result.Metrics.RSquared;

// Training info
var trainingTime = result.TrainingTime;

Next Steps

Now that you have the basics, explore these topics:

1. Neural Networks - Deep learning architectures
2. Classical ML - Traditional machine learning algorithms
3. Computer Vision - Image processing and detection
4. NLP - Natural language processing
5. RAG & LLMs - Retrieval-augmented generation
6. Reinforcement Learning - RL agents and environments

Interactive Playground

Want to try AiDotNet without installing anything? Visit our Interactive Playground to write and run code directly in your browser!

Getting Help

• FAQ - Common questions and answers
• Troubleshooting - Solutions to common issues
• GitHub Issues - Report bugs or request features