Adding N-HiTS network (N-BEATS successor)

Added

• Added new N-HiTS network that has consistently beaten N-BEATS (#890)
• Allow using torchmetrics as loss metrics (#776)
• Enable fitting EncoderNormalizer() with limited data history using max_length argument (#782)
• More flexible MultiEmbedding() with convenience output_size and input_size properties (#829)
• Fix concatentation of attention (#902)

Fixed

• Fix pip install via github (#798)

Contributors

• jdb78
• christy
• lukemerrick
• Seon82

Maintenance Release

Added

• Added support for running pytorch_lightning.trainer.test (#759)

Fixed

• Fix inattention mutation to x_cont (#732).
• Compatability with pytorch-lightning 1.5 (#758)

Contributors

• eavae
• danielgafni
• jdb78

Maintenance Release (26/09/2021)

Added

• Use target name instead of target number for logging metrics (#588)
• Optimizer can be initialized by passing string, class or function (#602)
• Add support for multiple outputs in Baseline model (#603)
• Added Optuna pruner as optional parameter in TemporalFusionTransformer.optimize_hyperparameters (#619)
• Dropping support for Python 3.6 and starting support for Python 3.9 (#639)

Fixed

• Initialization of TemporalFusionTransformer with multiple targets but loss for only one target (#550)
• Added missing transformation of prediction for MLP (#602)
• Fixed logging hyperparameters (#688)
• Ensure MultiNormalizer fit state is detected (#681)
• Fix infinite loop in TimeDistributedEmbeddingBag (#672)

Contributors

• jdb78
• TKlerx
• chefPony
• eavae
• L0Z1K

Simplified API

Breaking changes

• Removed dropout_categoricals parameter from TimeSeriesDataSet.
  Use categorical_encoders=dict(<variable_name>=NaNLabelEncoder(add_nan=True)) instead (#518)

• Rename parameter allow_missings for TimeSeriesDataSet to allow_missing_timesteps (#518)

• Transparent handling of transformations. Forward methods should now call two new methods (#518):

  • transform_output to explicitly rescale the network outputs into the de-normalized space
  • to_network_output to create a dict-like named tuple. This allows tracing the modules with PyTorch's JIT. Only prediction is still required which is the main network output.

  Example:

  def forward(self, x):
      normalized_prediction = self.module(x)
      prediction = self.transform_output(prediction=normalized_prediction, target_scale=x["target_scale"])
      return self.to_network_output(prediction=prediction)

Added

• Improved validation of input parameters of TimeSeriesDataSet (#518)

Fixed

• Fix quantile prediction for tensors on GPUs for distribution losses (#491)
• Fix hyperparameter update for RecurrentNetwork.from_dataset method (#497)

Generic distribution loss(es)

Added

• Allow lists for multiple losses and normalizers (#405)
• Warn if normalization is with scale < 1e-7 (#429)
• Allow usage of distribution losses in all settings (#434)

Fixed

• Fix issue when predicting and data is on different devices (#402)
• Fix non-iterable output (#404)
• Fix problem with moving data to CPU for multiple targets (#434)

Contributors

• jdb78
• domplexity

Simple models

Added

• Adding a filter functionality to the timeseries datasset (#329)
• Add simple models such as LSTM, GRU and a MLP on the decoder (#380)
• Allow usage of any torch optimizer such as SGD (#380)

Fixed

• Moving predictions to CPU to avoid running out of memory (#329)
• Correct determination of output_size for multi-target forecasting with the TemporalFusionTransformer (#328)
• Tqdm autonotebook fix to work outside of Jupyter (#338)
• Fix issue with yaml serialization for TensorboardLogger (#379)

Contributors

• jdb78
• JakeForsey
• vakker

Bugfix release

Added

• Make tuning trainer kwargs overwritable (#300)
• Allow adding categories to NaNEncoder (#303)

Fixed

• Underlying data is copied if modified. Original data is not modified inplace (#263)
• Allow plotting of interpretation on passed figure for NBEATS (#280)
• Fix memory leak for plotting and logging interpretation (#311)
• Correct shape of predict() method output for multi-targets (#268)
• Remove cloudpickle to allow GPU trained models to be loaded on CPU devices from checkpoints (#314)

Contributors

• jdb78
• kigawas
• snumumrik

Fix for output transformer

• Added missing output transformation which was switched off by default (#260)

Adding support for lag variables

Added

• Add "Release Notes" section to docs (#237)
• Enable usage of lag variables for any model (#252)

Changed

• Require PyTorch>=1.7 (#245)

Fixed

• Fix issue for multi-target forecasting when decoder length varies in single batch (#249)
• Enable longer subsequences for min_prediction_idx that were previously wrongfully excluded (#250)

Contributors

• jdb78

Adding multi-target support

Added

• Adding support for multiple targets in the TimeSeriesDataSet (#199) and amended tutorials.
• Temporal fusion transformer and DeepAR with support for multiple tagets (#199)
• Check for non-finite values in TimeSeriesDataSet and better validate scaler argument (#220)
• LSTM and GRU implementations that can handle zero-length sequences (#235)
• Helpers for implementing auto-regressive models (#236)

Changed

• TimeSeriesDataSet's y of the dataloader is a tuple of (target(s), weight) - potentially breaking for model or metrics implementation
  Most implementations will not be affected as hooks in BaseModel and MultiHorizonMetric were modified.

Fixed

• Fixed autocorrelation for pytorch 1.7 (#220)
• Ensure reproducibility by replacing python set() with dict.fromkeys() (mostly TimeSeriesDataSet) (#221)
• Ensures BetaDistributionLoss does not lead to infinite loss if actuals are 0 or 1 (#233)
• Fix for GroupNormalizer if scaling by group (#223)
• Fix for TimeSeriesDataSet when using min_prediction_idx (#226)

Contributors

• jdb78
• JustinNeumann
• reumar
• rustyconover