gluonts.model.renewal package

class gluonts.model.renewal.DeepRenewalProcessEstimator(freq: str, prediction_length: int, context_length: int, num_cells: int, num_layers: int, dropout_rate: float = 0.1, interval_distr_output: =, size_distr_output: =, train_sampler: Optional[gluonts.transform.sampler.InstanceSampler] = None, validation_sampler: Optional[gluonts.transform.sampler.InstanceSampler] = None, trainer: =, batch_size=None, callbacks=None, clip_gradient=10.0, ctx=None, epochs=100, hybridize=False, init="xavier", learning_rate=0.001, learning_rate_decay_factor=0.5, minimum_learning_rate=5e-05, num_batches_per_epoch=50, patience=10, weight_decay=1e-08), batch_size: int = 32, num_parallel_samples: int = 100, **kwargs)[source]


Implements a deep renewal process estimator designed to forecast intermittent time series sampled in discrete time, as described in [TWJ19].

In short, instead of viewing sparse time series as a univariate stochastic process, this estimator transforms a sparse time series [0, 0, 0, 3, 0, 0, 7] to an interval-size format, [(4, 3), (3, 7)] where each ordered pair marks the time since the last positive time step (interval) and the value of the positive time step (size). Then, probabilistic prediction is performed on this transformed time series, as is customary in the intermittent demand literature, e.g., Croston’s method.

This construction is a self-modulated marked renewal process in discrete time as one assumes the (conditional) distribution of intervals are identical.

  • freq – Frequency of the data to train on and predict

  • prediction_length – Length of the prediction horizon

  • context_length – The number of time steps the model will condition on

  • num_cells – Number of hidden units used in the RNN cell (LSTM) and dense layer for projection to distribution arguments

  • num_layers – Number of layers in the LSTM

  • dropout_rate – Dropout regularization parameter (default: 0.1)

  • trainer – Trainer object to be used (default: Trainer())

  • interval_distr_output – Distribution output object for the intervals. Must be a distribution with support on positive integers, where the first argument corresponds to the (conditional) mean.

  • size_distr_output – Distribution output object for the demand sizes. Must be a distribution with support on positive integers, where the first argument corresponds to the (conditional) mean.

  • train_sampler – Controls the sampling of windows during training.

  • validation_sampler – Controls the sampling of windows during validation.

  • batch_size – The size of the batches to be used training and prediction.

  • num_parallel_samples – Number of evaluation samples per time series to increase parallelism during inference. This is a model optimization that does not affect the accuracy (default: 100)

create_predictor(transformation: gluonts.transform._base.Transformation, trained_network: mxnet.gluon.block.HybridBlock) → gluonts.model.predictor.Predictor[source]

Create and return a predictor object.


A predictor wrapping a HybridBlock used for inference.

Return type


create_training_data_loader(data: gluonts.dataset.common.Dataset, **kwargs) → Iterable[Dict[str, Any]][source]
create_training_network() → gluonts.model.renewal._network.DeepRenewalTrainingNetwork[source]

Create and return the network used for training (i.e., computing the loss).


The network that computes the loss given input data.

Return type


create_transformation() → gluonts.transform._base.Transformation[source]

Create and return the transformation needed for training and inference.


The transformation that will be applied entry-wise to datasets, at training and inference time.

Return type


create_validation_data_loader(data: gluonts.dataset.common.Dataset, **kwargs) → Iterable[Dict[str, Any]][source]
freq = None
lead_time = None
prediction_length = None