Mamba<T>

Models & Types · AiDotNet.Finance.Forecasting.StateSpace

Mamba (Selective State Space Model) implementation for time series forecasting.

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

For Beginners

Mamba is a breakthrough in efficient sequence modeling:

The Key Insight: Transformers have O(n^2) complexity due to attention, which is slow for long sequences. State space models (SSMs) have O(n) complexity but are less expressive. Mamba makes SSM parameters input-dependent (selective), combining the best of both.

How It Works:

  1. State Space Model: Maintains a hidden state updated recurrently
  2. Selective Mechanism: Parameters (A, B, C, delta) vary with input
  3. Hardware-aware Algorithm: Efficient implementation via parallel scan
  4. Linear Complexity: O(n) time and memory for sequence length n

Advantages:

  • Linear time complexity (vs O(n^2) for attention)
  • Handles very long sequences efficiently
  • Strong performance on language, audio, and time series
  • Hardware-efficient implementation

How It Works

Mamba is a selective state space model that achieves linear-time complexity for sequence modeling while maintaining the expressiveness of transformers through input-dependent (selective) state space parameters.

Reference: Gu and Dao, “Mamba: Linear-Time Sequence Modeling with Selective State Spaces”, 2024. https://arxiv.org/abs/2312.00752

Example

var architecture = new NeuralNetworkArchitecture<double>(
    inputType: InputType.OneDimensional,
    taskType: NeuralNetworkTaskType.Regression,
    inputHeight: 512, inputWidth: 1, inputDepth: 1, outputSize: 24);
var model = new Mamba<double>(architecture);
var onnxModel = new Mamba<double>(architecture, "mamba.onnx");

Constructors

Constructor Summary
Mamba(NeuralNetworkArchitecture<T> architecture, MambaOptions<T> options, IGradientBasedOptimizer<T, Tensor<T>, Tensor<T>> optimizer, ILossFunction<T> lossFunction) Initializes a new instance in native mode for training.
Mamba(NeuralNetworkArchitecture<T> architecture, string onnxModelPath, MambaOptions<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 UseBidirectional { get; } Gets whether bidirectional processing is used.
bool UseNativeMode { get; }
int ExpandFactor { get; } Gets the expansion factor.
int NumFeatures { get; }
int PatchSize { get; }
int PredictionHorizon { get; }
int SequenceLength { get; }
int StateDimension { get; } Gets the state dimension of the SSM.
int Stride { get; }

Methods

Method Summary
Dictionary<string, T> Evaluate(Tensor<T> predictions, Tensor<T> actuals)
Dictionary<string, T> GetFinancialMetrics()
ModelMetadata<T> GetModelMetadata()
Tensor<T> ApplyInstanceNormalization(Tensor<T> input)
Tensor<T> AutoregressiveForecast(Tensor<T> input, int steps)