Neuralforecast

New features
- FEAT: TimeXer marcopeix (1267)
- All losses compatible with all types of models (e.g. univariate/multivariate, direct/recurrent) OR appropriate protection added.
- DistributionLoss now supports the use of `quantiles` in `predict`, allowing for easy quantile retrieval for all `DistributionLosses`.
- Mixture losses (GMM, PMM and NBMM) now support learned weights for weighted mixture distribution outputs.
- Mixture losses now support the use of `quantiles` in `predict`, allowing for easy quantile retrieval.
- Improved stability of `ISQF` by adding softplus protection around some parameters instead of using `.abs`.
- Unified API for any quantile or any confidence level during predict for both point and distribution losses.

Enhancements
- [DOCS] Docstrings elephaint (1279)
- FIX: Minor bug fix in TFT and a nicer error message for fitting with the wrong val_size marcopeix (1275)
- [FIX] Adds bfloat16 support elephaint (1265)
- Recurrent models can now produce forecasts recursively or directly.
- IQLoss now gives monotonic quantiles
- MASE loss now works

Breaking Changes
- [FIX] Unify API elephaint (1023)
- RMoK uses the `revin_affine` parameter instead of `revine_affine`. This was a typo in the previous version.
- All models now inherit the `BaseModel` class. This changes how we implement new models in neuralforecast.
- Recurrent models now require an `input_size` parameter.
- `TCN` and `DRNN` are now window models, not recurrent models
- We cannot load a recurrent model from a previous version to v3.0.0

Bug Fixes
- [FIX] Multivariate models give error when predicting when n_series > batch_size elephaint (1276)
- [FIX]: Insample predictions with series of varying lengths marcopeix (1246)

Documentation
- [DOCS] Update documentation elephaint (1274)
- [DOCS] Add example of modifying the default configure_optimizers() behavior (use of ReduceLROnPlateau scheduler) JQGoh (1015)

Enhancements
- FEAT: Custom RNN layers for TFT Yanam24 (1230)
- FEAT: Add the horizon weighing to the distribution losses mwamsojo (1233)

Documentation
- DOCS: Add citation note elephaint (1244)
- fix: azul AzulGarza (1245)

Breaking Change

- breaking: remove deprecated behavior jmoralez (1220)

Bug Fixes

- [FIX] Backward compatibility: missing prediction_intervals JQGoh (1224)

New Features

- [FEAT]: Support providing DataLoader arguments to optimize GPU usage jasminerienecker (1186)
- [FEAT]: Set activation function in GRN of TFT marcopeix (1175)
- [FEAT]: Conformal Predictions in NeuralForecast JQGoh (1171)

Bug Fixes

- [FIX]: Ability load models saved using versions before 1.7 tylernisonoff (1207)
- [FIX]: Conformal prediction issues elephaint (1179)
- [FIX]: Feature importance when using only hist_exog in TFT fails elephaint (1174)
- [FIX]: Remove unused output layer NBEATSx elephaint (1168)
- [FIX]: Fix Tweedie loss elephaint (1164)
- [FIX]: MLPMultivariate incorrect static_exog parsing elephaint (1170)
- [FIX]: Deprecate activation functions for GRU marcopeix (1198)

Documentation

- [DOC]: Tutorial on cross-validation marcopeix (1176)
- [DOC]: Build docs on release only elephaint (1183)

New Features
- [FEAT]: Move RevIN class to common module JQGoh (1083)
- [FEAT]: Add RMoK marcopeix (1148)
- FEAT: TimeLLM is faster and supports more LLMs ive2go (1139)
- [FEAT]: TFT-Interpretability amirouyanis (1104)
- [FEAT]: ]Add support for the local file dataloader with the Automodels jasminerienecker (1095)

Bug Fixes
- [FIX] CV Refit works with non-standard column names elephaint (1149)
- [FIX]: replace self.pred_len with self.h carusyte (1129)
- [FIX]: only define static encoder when applicable in TFT jmoralez (1114)
- [FIX]: remove cast to float in scalers jmoralez (1115)
- [FIX]: timemixer shapes mismatch and doc update carusyte marcopeix (1138)

Dependencies
- Bump pypa/gh-action-pypi-publish from 1.10.0 to 1.10.1 in the ci-dependencies group dependabot (1146)
- Bump the ci-dependencies group with 2 updates dependabot (1135)