GraphAttentionPortfolioOptions<T>
Options & Configuration · AiDotNet.Models.Options
Configuration for GraphAttentionPortfolio.
Inheritance: ModelOptions → NeuralNetworkOptions → GraphAttentionPortfolioOptions<T>
For Beginners
This model builds a network of which companies move together, learns over that network which ones to hold, and is trained to maximize return per unit of risk.
How It Works
Defaults follow Kamesh Korangi, Christophe Mues and Cristián Bravo, “Large-scale Time-Varying Portfolio Optimisation using Graph Attention Networks” (arXiv:2407.15532): attention feature width 24, 8 heads, a 30-day volatility lookback, and a three-year correlation window.
Constructors
| Constructor | Summary |
|---|---|
GraphAttentionPortfolioOptions() |
Properties
| Property | Summary |
|---|---|
double DropoutRate { get; set; } |
Gets or sets the dropout rate applied after the first feed-forward block. |
double L1Regularization { get; set; } |
Gets or sets the L1 (LASSO) strength on the final scoring layer. |
double LeakyReLUSlope { get; set; } |
Gets or sets the LeakyReLU negative slope in the attention. |
double LearningRate { get; set; } |
Gets or sets the learning rate. |
int AttentionFeatureDimension { get; set; } |
Gets or sets T’, the per-head attention feature width. |
int BatchSize { get; set; } |
Gets or sets the batch size. |
int CorrelationWindow { get; set; } |
Gets or sets the correlation window in trading days. |
int MaxEpochs { get; set; } |
Gets or sets the maximum training epochs. |
int NumAssets { get; set; } |
Gets or sets the number of assets. |
int NumHeads { get; set; } |
Gets or sets K, the number of attention heads. |
int VolatilityLookback { get; set; } |
Gets or sets the volatility lookback in days. |
Methods
| Method | Summary |
|---|---|
void Validate() |
Validates the configuration. |