gluonts.model.wavenet package

class gluonts.model.wavenet.WaveNet(bin_values: List[float], n_residue: int, n_skip: int, dilation_depth: int, n_stacks: int, act_type: str, cardinality: List[int], embedding_dimension: int, pred_length: int, **kwargs)[source]

Bases: mxnet.gluon.block.HybridBlock

base_net(F, inputs, one_step_prediction=False, queues=None)[source]

Forward pass through the network.

Parameters
  • F

  • inputs – Inputs to the network: (batch_size, n_residue, sequence_length)

  • one_step_prediction – Flag indicating whether the network is “unrolled/propagated” one step at a time (prediction phase).

  • queues – Convolutional queues containing past computations. Should be provided if one_step_prediction is True.

Returns

Tuple – A tensor containing the unnormalized outputs of the network. Shape: (batch_size, pred_length, num_bins). A list containing the convolutional queues for each layer. The queue corresponding to layer l has shape: (batch_size, n_residue, 2^l).

Return type

(Tensor, List)

get_full_features(F, feat_static_cat: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol], past_observed_values: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol], past_time_feat: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol], future_time_feat: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol], future_observed_values: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol, None], scale: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol])[source]

Prepares the inputs for the network by repeating static feature and concatenating it with time features and observed value indicator.

Parameters
  • F

  • feat_static_cat – Static categorical features: (batch_size, num_cat_features)

  • past_observed_values – Observed value indicator for the past target: (batch_size, receptive_field)

  • past_time_feat – Past time features: (batch_size, num_time_features, receptive_field)

  • future_time_feat – Future time features: (batch_size, num_time_features, pred_length)

  • future_observed_values – Observed value indicator for the future target: (batch_size, pred_length). This will be set to all ones, if not provided (e.g., during prediction).

  • scale – scale of the time series: (batch_size, 1)

Returns

A tensor containing all the features ready to be passed through the network. Shape: (batch_size, num_features, receptive_field + pred_length)

Return type

Tensor

static get_receptive_field(dilation_depth, n_stacks)[source]

Return the length of the receptive field

is_last_layer(i)[source]
target_feature_embedding(F, target, features)[source]

Provides a joint embedding for the target and features.

Parameters
  • F

  • target ((batch_size, sequence_length)) –

  • features ((batch_size, num_features, sequence_length)) –

Returns

A tensor containing a joint embedding of target and features. Shape: (batch_size, n_residue, sequence_length)

Return type

Tensor

class gluonts.model.wavenet.WaveNetEstimator(freq: str, prediction_length: int, trainer: gluonts.mx.trainer._base.Trainer = gluonts.mx.trainer._base.Trainer(avg_strategy=gluonts.mx.trainer.model_averaging.SelectNBestMean(maximize=False, metric="score", num_models=1), batch_size=None, clip_gradient=10.0, ctx=None, epochs=200, hybridize=False, init="xavier", learning_rate=0.01, learning_rate_decay_factor=0.5, minimum_learning_rate=5e-05, num_batches_per_epoch=50, patience=10, post_initialize_cb=None, weight_decay=1e-08), cardinality: List[int] = [1], seasonality: Optional[int] = None, embedding_dimension: int = 5, num_bins: int = 1024, hybridize_prediction_net: bool = False, n_residue=24, n_skip=32, dilation_depth: Optional[int] = None, n_stacks: int = 1, train_window_length: Optional[int] = None, temperature: float = 1.0, act_type: str = 'elu', num_parallel_samples: int = 200, train_sampler: Optional[gluonts.transform.sampler.InstanceSampler] = None, validation_sampler: Optional[gluonts.transform.sampler.InstanceSampler] = None, batch_size: int = 32, negative_data: bool = False)[source]

Bases: gluonts.mx.model.estimator.GluonEstimator

Model with Wavenet architecture and quantized target.

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

  • prediction_length – Length of the prediction horizon

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

  • cardinality – Number of values of the each categorical feature (default: [1])

  • embedding_dimension – Dimension of the embeddings for categorical features (the same dimension is used for all embeddings, default: 5)

  • num_bins – Number of bins used for quantization of signal (default: 1024)

  • hybridize_prediction_net – Boolean (default: False)

  • n_residue – Number of residual channels in wavenet architecture (default: 24)

  • n_skip – Number of skip channels in wavenet architecture (default: 32)

  • dilation_depth – Number of dilation layers in wavenet architecture. If set to None (default), dialation_depth is set such that the receptive length is at least as long as typical seasonality for the frequency and at least 2 * prediction_length.

  • n_stacks – Number of dilation stacks in wavenet architecture (default: 1)

  • temperature – Temparature used for sampling from softmax distribution. For temperature = 1.0 (default) sampling is according to estimated probability.

  • act_type – Activation type used after before output layer (default: “elu”). Can be any of ‘elu’, ‘relu’, ‘sigmoid’, ‘tanh’, ‘softrelu’, ‘softsign’.

  • 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: 200)

  • 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.

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

Create and return a predictor object.

Returns

A predictor wrapping a HybridBlock used for inference.

Return type

Predictor

create_training_data_loader(data: Iterable[Dict[str, Any]], **kwargs) → gluonts.dataset.loader.DataLoader[source]
create_training_network() → gluonts.model.wavenet._network.WaveNetTraining[source]

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

Returns

The network that computes the loss given input data.

Return type

HybridBlock

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

Create and return the transformation needed for training and inference.

Returns

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

Return type

Transformation

create_validation_data_loader(data: Iterable[Dict[str, Any]], **kwargs) → gluonts.dataset.loader.DataLoader[source]
freq = None
lead_time = None
prediction_length = None
class gluonts.model.wavenet.WaveNetSampler(bin_values: List[float], num_samples: int, temperature: float = 1.0, **kwargs)[source]

Bases: gluonts.model.wavenet._network.WaveNet

Runs Wavenet generation in an auto-regressive manner using caching for speedup [PKC+16].

Same arguments as WaveNet. In addition

Parameters
  • pred_length – Length of the prediction horizon

  • num_samples – Number of sample paths to generate in parallel in the graph

  • temperature – If set to 1.0 (default), sample according to estimated probabilities, if set to 0.0 most likely sample at each step is chosen.

  • post_transform – An optional post transform that will be applied to the samples

get_initial_conv_queues(F, past_target, features)[source]

Build convolutional queues saving intermediate computations.

Parameters
  • F

  • past_target ((batch_size, receptive_field)) –

  • features ((batch_size, num_features, receptive_field)) –

Returns

A list containing the convolutional queues for each layer. The queue corresponding to layer l has shape: (batch_size, n_residue, 2^l).

Return type

List

hybrid_forward(F, feat_static_cat: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol], past_target: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol], past_observed_values: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol], past_time_feat: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol], future_time_feat: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol], scale: Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol]) → Union[mxnet.ndarray.ndarray.NDArray, mxnet.symbol.symbol.Symbol][source]

Computes the training loss for the wavenet model.

Parameters
  • F

  • feat_static_cat – Static categorical features: (batch_size, num_cat_features)

  • past_target – Past target: (batch_size, receptive_field)

  • past_observed_values – Observed value indicator for the past target: (batch_size, receptive_field)

  • past_time_feat – Past time features: (batch_size, num_time_features, receptive_field)

  • future_time_feat – Future time features: (batch_size, num_time_features, pred_length)

  • scale – scale of the time series: (batch_size, 1)

Returns

Prediction samples with shape (batch_size, num_samples, pred_length)

Return type

Tensor