TimeGPT<T>

Models & Types · AiDotNet.Finance.Forecasting.Foundation

TimeGPT-style time series foundation model implementation.

Inheritance: NeuralNetworkBase<T> → FinancialModelBase<T> → ForecastingModelBase<T> → TimeGPT<T>

For Beginners

TimeGPT brings GPT-like capabilities to time series:

The Key Insight: Just as GPT was trained on internet-scale text data to become a general-purpose language model, TimeGPT is trained on millions of diverse time series to become a general-purpose forecasting model.

Core Features:

  1. Large-scale Pre-training: Trained on millions of time series
  2. Zero-shot Forecasting: No training needed for new data
  3. Uncertainty Quantification: Provides prediction intervals
  4. Multi-horizon: Forecasts at any horizon

Advantages:

  • Works out-of-the-box on new time series
  • No hyperparameter tuning required
  • Handles various frequencies and domains
  • Production-ready forecasting

How It Works

TimeGPT represents a GPT-style architecture adapted for time series forecasting, featuring large-scale pre-training on diverse time series data with zero-shot and few-shot forecasting capabilities.

Reference: Garza et al., “TimeGPT-1”, 2023. https://arxiv.org/abs/2310.03589

Example

// Create a TimeGPT-style foundation model for zero-shot time series forecasting
// Pre-trained on millions of diverse time series for out-of-the-box predictions
var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.OneDimensional,
    taskType: NeuralNetworkTaskType.Regression,
    inputHeight: 512, inputWidth: 1, inputDepth: 1, outputSize: 24);
            
// Training mode with GPT-style autoregressive generation
var model = new TimeGPT<double>(architecture);
            
// ONNX inference mode with pre-trained model
var onnxModel = new TimeGPT<double>(architecture, "timegpt.onnx");

Constructors

Constructor Summary
TimeGPT(NeuralNetworkArchitecture<T> architecture, TimeGPTOptions<T> options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction) Initializes a new instance in native mode for training or fine-tuning.
TimeGPT(NeuralNetworkArchitecture<T> architecture, string onnxModelPath, TimeGPTOptions<T> options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction) Initializes a new instance using an ONNX pretrained model.

Properties

Property Summary
bool IsChannelIndependent { get; }
bool SupportsTraining { get; }
bool UseConformalPrediction { get; } Gets whether conformal prediction is used for uncertainty.
bool UseNativeMode { get; }
double ConfidenceLevel { get; } Gets the confidence level for prediction intervals.
int NumFeatures { get; }
int PatchSize { get; }
int PredictionHorizon { get; }
int SequenceLength { get; }
int Stride { get; }

Methods

Method Summary
Dictionary<string, T> Evaluate(Tensor<T> predictions, Tensor<T> actuals)
Dictionary<string, T> GetFinancialMetrics()
Dictionary<string, Tensor<T>> GetNamedLayerActivations(Tensor<T> input)
ModelMetadata<T> GetModelMetadata()
Tensor<T> ApplyInstanceNormalization(Tensor<T> input)
Tensor<T> AutoregressiveForecast(Tensor<T> input, int steps)
void CalibrateConformalPrediction(List<Tensor<T>> historicalData, List<Tensor<T>> actuals) Calibrates the model using historical data for conformal prediction.
void FineTune(List<Tensor<T>> trainingData, List<Tensor<T>> targets) Fine-tunes the model on domain-specific data.