InformerOptions<T>

Options & Configuration · AiDotNet.Models.Options

Configuration options for the Informer model (Informer: Beyond Efficient Transformer for Long Sequence Time-Series Forecasting).

Inheritance: ModelOptions → RegressionOptions<T> → TimeSeriesRegressionOptions<T> → InformerOptions<T>

For Beginners

Informer is an efficient version of the Transformer architecture designed specifically for long time series. Traditional transformers become very slow with long sequences, but Informer uses smart tricks to be much faster while maintaining accuracy. It’s particularly good for forecasting that requires looking far back in history (like predicting next month based on the past year).

How It Works

Informer addresses the computational complexity challenges of vanilla Transformers for long-sequence forecasting. Key innovations include:

  • ProbSparse self-attention mechanism (O(L log L) complexity instead of O(L²))
  • Self-attention distilling for efficient stacking
  • Generative style decoder for one-forward prediction

Constructors

Constructor Summary
InformerOptions() Creates options with the default values.
InformerOptions(InformerOptions<T> other) Creates a copy of another options instance.

Properties

Property Summary
InformerAttention Attention { get; set; } Gets or sets the self-attention of the encoder and decoder (attn, ProbSparse).
bool Distil { get; set; } Gets or sets whether a distilling layer (conv, batch norm, ELU, max-pool halving the length) follows every encoder layer but the last (distil, true).
bool Mix { get; set; } Gets or sets whether the decoder self-attention mixes its heads as the reference does (mix, true).
bool UseGelu { get; set; } Gets or sets whether the feed-forward activation is GELU (activation ‘gelu’; false: ReLU).
double DropoutRate { get; set; } Gets or sets the dropout rate (dropout, 0.05).
double LearningRate { get; set; } Gets or sets the initial learning rate of Adam (learning_rate, 1e-4), halved after every epoch (lradj ‘type1’).
int BatchSize { get; set; } Gets or sets the batch size (batch_size, 32).
int EmbeddingDim { get; set; } Gets or sets the model width (d_model, 512).
int Epochs { get; set; } Gets or sets the number of training epochs (train_epochs, 6).
int FeedForwardDim { get; set; } Gets or sets the feed-forward width (d_ff, 2048).
int ForecastHorizon { get; set; } Gets or sets the forecast horizon, the number of predicted steps (pred_len, 24).
int LabelLength { get; set; } Gets or sets the start-token length: the last steps of the lookback window that open the decoder input (label_len).
int LookbackWindow { get; set; } Gets or sets the lookback window, the encoder input length (seq_len; the reference’s default 96).
int NumAttentionHeads { get; set; } Gets or sets the number of attention heads (n_heads, 8).
int NumDecoderLayers { get; set; } Gets or sets the number of decoder layers (d_layers, 1).
int NumEncoderLayers { get; set; } Gets or sets the number of encoder layers (e_layers, 2).
int SamplingFactor { get; set; } Gets or sets the ProbSparse sampling factor c (factor, 5): c·⌈ln L⌉ keys are sampled per query and as many queries attend.