class pytorch_forecasting.models.deepar.DeepAR(cell_type: str = 'LSTM', hidden_size: int = 10, rnn_layers: int = 2, dropout: float = 0.1, static_categoricals: List[str] = [], static_reals: List[str] = [], time_varying_categoricals_encoder: List[str] = [], time_varying_categoricals_decoder: List[str] = [], categorical_groups: Dict[str, List[str]] = {}, time_varying_reals_encoder: List[str] = [], time_varying_reals_decoder: List[str] = [], embedding_sizes: Dict[str, Tuple[int, int]] = {}, embedding_paddings: List[str] = [], embedding_labels: Dict[str, ndarray] = {}, x_reals: List[str] = [], x_categoricals: List[str] = [], n_validation_samples: int | None = None, n_plotting_samples: int | None = None, target: str | List[str] | None = None, target_lags: Dict[str, List[int]] = {}, loss: DistributionLoss | None = None, logging_metrics: ModuleList | None = None, **kwargs)[source]#

Bases: AutoRegressiveBaseModelWithCovariates

DeepAR Network.

The code is based on the article DeepAR: Probabilistic forecasting with autoregressive recurrent networks.

By using a Multivariate Loss such as the MultivariateNormalDistributionLoss, the network is converted into a DeepVAR network.

  • cell_type (str, optional) – Recurrent cell type [“LSTM”, “GRU”]. Defaults to “LSTM”.

  • hidden_size (int, optional) – hidden recurrent size - the most important hyperparameter along with rnn_layers. Defaults to 10.

  • rnn_layers (int, optional) – Number of RNN layers - important hyperparameter. Defaults to 2.

  • dropout (float, optional) – Dropout in RNN layers. Defaults to 0.1.

  • static_categoricals – integer of positions of static categorical variables

  • static_reals – integer of positions of static continuous variables

  • time_varying_categoricals_encoder – integer of positions of categorical variables for encoder

  • time_varying_categoricals_decoder – integer of positions of categorical variables for decoder

  • time_varying_reals_encoder – integer of positions of continuous variables for encoder

  • time_varying_reals_decoder – integer of positions of continuous variables for decoder

  • categorical_groups – dictionary where values are list of categorical variables that are forming together a new categorical variable which is the key in the dictionary

  • x_reals – order of continuous variables in tensor passed to forward function

  • x_categoricals – order of categorical variables in tensor passed to forward function

  • embedding_sizes – dictionary mapping (string) indices to tuple of number of categorical classes and embedding size

  • embedding_paddings – list of indices for embeddings which transform the zero’s embedding to a zero vector

  • embedding_labels – dictionary mapping (string) indices to list of categorical labels

  • n_validation_samples (int, optional) – Number of samples to use for calculating validation metrics. Defaults to None, i.e. no sampling at validation stage and using “mean” of distribution for logging metrics calculation.

  • n_plotting_samples (int, optional) – Number of samples to generate for plotting predictions during training. Defaults to n_validation_samples if not None or 100 otherwise.

  • target (str, optional) – Target variable or list of target variables. Defaults to None.

  • target_lags (Dict[str, Dict[str, int]]) – dictionary of target names mapped to list of time steps by which the variable should be lagged. Lags can be useful to indicate seasonality to the models. If you know the seasonalit(ies) of your data, add at least the target variables with the corresponding lags to improve performance. Defaults to no lags, i.e. an empty dictionary.

  • loss (DistributionLoss, optional) – Distribution loss function. Keep in mind that each distribution loss function might have specific requirements for target normalization. Defaults to NormalDistributionLoss.

  • logging_metrics (nn.ModuleList, optional) – Metrics to log during training. Defaults to nn.ModuleList([SMAPE(), MAE(), RMSE(), MAPE(), MASE()]).


construct_input_vector(x_cat, x_cont[, ...])

Create input vector into RNN network

create_log(x, y, out, batch_idx)

Create the log used in the training and validation step.

decode(input_vector, target_scale, ...[, ...])

Decode hidden state of RNN into prediction.

decode_all(x, hidden_state[, lengths])


Encode sequence into hidden state

forward(x[, n_samples])

Forward network

from_dataset(dataset[, ...])

Create model from dataset.

predict(data[, mode, return_index, ...])

predict dataloader

construct_input_vector(x_cat: Tensor, x_cont: Tensor, one_off_target: Tensor | None = None) Tensor[source]#

Create input vector into RNN network


one_off_target – tensor to insert into first position of target. If None (default), remove first time step.

create_log(x, y, out, batch_idx)[source]#

Create the log used in the training and validation step.

  • x (Dict[str, torch.Tensor]) – x as passed to the network by the dataloader

  • y (Tuple[torch.Tensor, torch.Tensor]) – y as passed to the loss function by the dataloader

  • out (Dict[str, torch.Tensor]) – output of the network

  • batch_idx (int) – batch number

  • prediction_kwargs (Dict[str, Any], optional) – arguments to pass to to_prediction(). Defaults to {}.

  • quantiles_kwargs (Dict[str, Any], optional) – to_quantiles(). Defaults to {}.


log dictionary to be returned by training and validation steps

Return type:

Dict[str, Any]

decode(input_vector: Tensor, target_scale: Tensor, decoder_lengths: Tensor, hidden_state: Tuple[Tensor, Tensor] | Tensor, n_samples: int | None = None) Tuple[Tensor, bool][source]#

Decode hidden state of RNN into prediction. If n_smaples is given, decode not by using actual values but rather by sampling new targets from past predictions iteratively

encode(x: Dict[str, Tensor]) Tuple[Tensor, Tensor] | Tensor[source]#

Encode sequence into hidden state

forward(x: Dict[str, Tensor], n_samples: int | None = None) Dict[str, Tensor][source]#

Forward network

classmethod from_dataset(dataset: TimeSeriesDataSet, allowed_encoder_known_variable_names: List[str] | None = None, **kwargs)[source]#

Create model from dataset.

  • dataset – timeseries dataset

  • allowed_encoder_known_variable_names – List of known variables that are allowed in encoder, defaults to all

  • **kwargs – additional arguments such as hyperparameters for model (see __init__())


DeepAR network

predict(data: DataLoader | DataFrame | TimeSeriesDataSet, mode: str | Tuple[str, str] = 'prediction', return_index: bool = False, return_decoder_lengths: bool = False, batch_size: int = 64, num_workers: int = 0, fast_dev_run: bool = False, return_x: bool = False, return_y: bool = False, mode_kwargs: Dict[str, Any] | None = None, trainer_kwargs: Dict[str, Any] | None = None, write_interval: Literal['batch', 'epoch', 'batch_and_epoch'] = 'batch', output_dir: str | None = None, n_samples: int = 100, **kwargs) Prediction[source]#

predict dataloader

  • dataloader – dataloader, dataframe or dataset

  • mode – one of “prediction”, “quantiles”, “samples” or “raw”, or tuple ("raw", output_name) where output_name is a name in the dictionary returned by forward()

  • return_index – if to return the prediction index (in the same order as the output, i.e. the row of the dataframe corresponds to the first dimension of the output and the given time index is the time index of the first prediction)

  • return_decoder_lengths – if to return decoder_lengths (in the same order as the output

  • batch_size – batch size for dataloader - only used if data is not a dataloader is passed

  • num_workers – number of workers for dataloader - only used if data is not a dataloader is passed

  • fast_dev_run – if to only return results of first batch

  • show_progress_bar – if to show progress bar. Defaults to False.

  • return_x – if to return network inputs (in the same order as prediction output)

  • return_y – if to return network targets (in the same order as prediction output)

  • mode_kwargs (Dict[str, Any]) – keyword arguments for to_prediction() or to_quantiles() for modes “prediction” and “quantiles”

  • trainer_kwargs (Dict[str, Any], optional) – keyword arguments for the trainer

  • write_interval – interval to write predictions to disk

  • output_dir – directory to write predictions to. Defaults to None. If set function will return empty list

  • n_samples – number of samples to draw. Defaults to 100.


if one of the `return arguments is present,

prediction tuple with fields prediction, x, y, index and decoder_lengths

Return type: