Nixtla/neuralforecast v1.6.1

on GitHub

New Models

• DeepAR
• FEDformer

New features

• Available Mask to specify missing data in input data frame.
• Improve fit and cross_validation methods with use_init_models parameter to restore models to initial parameters.
• Added robust losses: HuberLoss, TukeyLoss, HuberQLoss, and HuberMQLoss.
• Added Bernoulli DistributionLoss to build temporal classifiers.
• New exclude_insample_y parameter to all models to build models only based on exogenous regressors.
• Added dropout to NBEATSx and NHITS models.
• Improved predict method of windows-based models to create batches to control memory usage. Can be controlled with the new inference_windows_batch_size parameter.
• Improvements to the HINT family of hierarchical models: identity reconciliation, AutoHINT, and reconciliation methods in hyperparameter selection.
• Added inference_input_sizehyperparameter to recurrent-based methods to control historic length during inference to better control memory usage and inference times.

New tutorials and documentation

• Neuralforecast map and How-to add new models
• Transformers for time-series
• Predict insample tutorial
• Interpretable Decomposition
• Outlier Robust Forecasting
• Temporal Classification
• Predictive Maintenance
• Statistical, Machine Learning, and Neural Forecasting methods

Fixed bugs and new protections

• Fixed bug on MinMax scalers that returned NaN values when the mask had 0 values.
• Fixed bug on y_loc and y_scale being in different devices.
• Added early_stopping_steps to the HINT method.
• Added protection in the fit method of all models to stop training when training or validation loss becomes NaN. Print input and output tensors for debugging.
• Added protection to prevent the case val_check_step > max_steps from causing an error when early stopping is enabled.
• Added PatchTST to save and load methods dictionaries.
• Added AutoNBEATSx to core's MODEL_DICT.
• Added protection to the NBEATSx-i model where horizon=1 causes an error due to collapsing trend and seasonality basis.