I-Transformer Model.

class flood_forecast.transformer_xl.itransformer.ITransformer(forecast_history, forecast_length, d_model, embed, dropout, n_heads=8, use_norm=True, e_layers=3, d_ff=512, freq='h', activation='gelu', factor=1, output_attention=True, targs=1)[source]

Paper link: https://arxiv.org/abs/2310.06625.

__init__(forecast_history, forecast_length, d_model, embed, dropout, n_heads=8, use_norm=True, e_layers=3, d_ff=512, freq='h', activation='gelu', factor=1, output_attention=True, targs=1)[source]

The complete iTransformer model.

Parameters:

  • forecast_history (int) – The number of historical steps to use for forecasting

  • forecast_length (int) – The length of the forecast the model outputs.

  • d_model (int) – The embedding dimension of the model. For the paper the authors used 512.

  • embed (str) – THe embedding type to use. For the paper the authors used ‘fixed’.

  • dropout (float) – The dropout for the model.

  • n_heads (int, optional) – Number of heads for the attention, defaults to 8

  • use_norm (bool, optional) – Whether to use normalization, defaults to True

  • e_layers (int, optional) – The number of embedding layers, defaults to 3

  • d_ff (int, optional) – _description_, defaults to 512

  • freq (str, optional) – The frequency of the time series data, defaults to ‘h’ for hourly

  • activation (str, optional) – The activation, defaults to ‘gelu’

  • factor (int, optional) – =n_, defaults to 1

  • output_attention (bool, optional) – Whether to output the scores, defaults to True

forecast(x_enc, x_mark_enc, x_dec, x_mark_dec)[source]

_summary_

Parameters:

  • x_enc (_type_) – _description_

  • x_mark_enc (_type_) – _description_

  • x_dec (_type_) – _description_

  • x_mark_dec (_type_) – _description_

Returns:

_description_

Return type:

_type_

forward(x_enc, x_mark_enc, x_dec, x_mark_dec, mask=None)[source]

_summary_

Parameters:

  • x_enc (_type_) – _description_

  • x_mark_enc (_type_) – _description_

  • x_dec (_type_) – _description_

  • x_mark_dec (_type_) – _description_

  • mask (_type_, optional) – _description_, defaults to None

Returns:

_description_

Return type:

_type_