latent module

utils.latent.encode_diresa(model: diresa_torch.arch.models.Diresa, data_loader: torch.utils.data.DataLoader, input_filter: callable = <function <lambda>>) → torch.Tensor

Encodes a dataset into a latent representation. Prerequisite: latent components are already ordered.

Parameters:

  • model – model to use

  • data_loader – data to be encoded

  • input_filter – function used to filter input data, default is no filtering

Returns:

latent representation of the dataset

utils.latent.latent_r2_per_variable(model: diresa_torch.arch.models.Diresa, data_loader: torch.utils.data.DataLoader, incr: bool = False, input_filter: callable = <function <lambda>>, target_filter: callable = <function <lambda>>, verbose: bool = False) → list

Computes R2 scores of latent components per variable. Variables should be on axis 1 (first is 0). If this is not the case, input_filter and target_filter can be used to swap axes. Prerequisite: latent components are already ordered.

Parameters:

  • model – Diresa model

  • data_loader – DataLoader

  • incr – If True incremental R2 score are calculated, default is False

  • input_filter – Function used to filter input data, default is no filtering

  • target_filter – Function used to filter target data, default is no filtering

  • verbose – If True, prints most important component per variable, default is False

Returns:

R2 scores of latent components per variable, shape (latent_dim, nbr of variables)

utils.latent.latent_vectors(model: diresa_torch.arch.models.Diresa, data_loader: torch.utils.data.DataLoader, factor: float = 0.5, incr: bool = False, input_filter: callable = <function <lambda>>) → torch.Tensor

Calculates decoded latent vectors. Prerequisite: model must be ordered. See: https://journals.ametsoc.org/view/journals/aies/4/3/AIES-D-24-0034.1.xml Appendix D: c.Latent variable interpretation.

Parameters:

  • model – Diresa model

  • data_loader – DataLoader

  • factor – Multiplication factor for standard deviation

  • incr – If True incremental vectors are calculated, default is False

  • input_filter – Function used to filter input data, default is no filtering

Returns:

Decoded latent vectors