neuralforecast

Bot releases are hidden (Show)

neuralforecast - v1.6.2

Published by FedericoGarza about 1 year ago

What's Changed

[FEAT] Add horizon_weight parameter to losses and BasePointLoss in https://github.com/Nixtla/neuralforecast/pull/704
[FIX] Fix device error in horizon_weight in https://github.com/Nixtla/neuralforecast/pull/706
[FIX] Base Windows padding in https://github.com/Nixtla/neuralforecast/pull/715
[FIX] Fixed bug in validation loss scale in https://github.com/Nixtla/neuralforecast/pull/720
[FIX] Base recurrent valid loss on original scale in https://github.com/Nixtla/neuralforecast/pull/721

Full Changelog: https://github.com/Nixtla/neuralforecast/compare/v1.6.1...v1.6.2

neuralforecast - v1.5.0

Published by FedericoGarza over 1 year ago

What's Changed

Features

New models

[FEAT] VanillaTransformer, Autoformer in https://github.com/Nixtla/neuralforecast/pull/469
[FEAT] StemGNN in https://github.com/Nixtla/neuralforecast/pull/456
[FEAT] PatchTST in https://github.com/Nixtla/neuralforecast/pull/485
[FEAT] Informer, augment_calendar_df, set seeds in fit and predict in https://github.com/Nixtla/neuralforecast/pull/463
[FEAT] Hierarchical Forecasting Networks (HINT) in https://github.com/Nixtla/neuralforecast/pull/489

Misc

[FEAT] Added MSSE class to losses.pytorch notebook in https://github.com/Nixtla/neuralforecast/pull/468
[FEAT] Robustified Distribution Outputs in https://github.com/Nixtla/neuralforecast/pull/492
[FEAT] Added MS availability to augment_calendar_df function in https://github.com/Nixtla/neuralforecast/pull/506
[FEAT] Add alias argument in https://github.com/Nixtla/neuralforecast/pull/502
[FEAT] mean default distribution output in addition to quantiles in https://github.com/Nixtla/neuralforecast/pull/529
[FEAT] Predict insample in https://github.com/Nixtla/neuralforecast/pull/528

New Contributors

@VinishUchiha made their first contribution in https://github.com/Nixtla/neuralforecast/pull/517

Full Changelog: https://github.com/Nixtla/neuralforecast/compare/v1.4.0...v1.5.0

neuralforecast - v1.4.0

Published by FedericoGarza over 1 year ago

New Models

Temporal Convolution Network (TCN)
AutoNBEATSx
AutoTFT (Transformers)

New features

Recurrent models (RNN, LSTM, GRU, DilatedRNN) can now take static, historical, and future exogenous variables. These variables are combined with lags to produce "context" vectors based on MLP decoders, based on the MQ-RNN model (https://arxiv.org/pdf/1711.11053.pdf).
The new DistributionLoss class allows for producing probabilistic forecasts with all available models. By changing the loss hyperparameter to one of these losses, the model will learn and output the distribution parameters:
- Bernoulli, Poisson, Normal, StudentT, Negative Binomial, and Tweedie distributions
- Scale-decoupled optimization using Temporal Scalers to improve convergence and performance.
- The predict method can return samples, quantiles, or distribution parameters.
sCRPS loss in PyTorch to minimize errors generating prediction intervals.

Optimization improvements

We included new optimization features commonly used to train neural models:

Added learning rate scheduler, using torch.optim.lr_scheduler.StepLR scheduler. The new num_lr_decays hyperparameter controls the number of decays (evenly distributed) during training.
Added Early stopping using validation loss. The new early_stop_patience_steps controls the number of validation steps with no improvement after which training will be stopped.
New validation loss hyperparameter to allow different train and validation losses

Training, scheduler, validation loss computation, and early stopping are now defined in steps (instead of epochs) to control the training procedure better. Use max_steps to define the number of training iterations. Note: max_epochs will be deprecated in the future.