Thread by @nixtlainc

🎉 We are very excited to release the new features of NeuralForecast! 🥳🚀

With this release time series forecasting with neural models is even more accessible and powerful. Here are some of the highlights

🧵

#timeseries #python #deeplearning #forecasting

📈 New Models:
- Temporal Convolution Network
- AutoNBEATSx
- AutoTFT (Transformers)

🧮 Recurrent models (RNN, LSTM, GRU, DilatedRNN) can now take static, historical, and future exogenous variables.

📊 Probabilistic forecasts:

- Bernoulli, Poisson, Normal, StudentT, Negative Binomial and Tweedie distributions
- Scale-decoupled optimization using Temporal Scalers to improve convergence and performance
- Predict method can return samples, quantiles, or distribution parameters

🔧 Optimization improvements:

- Added learning rate scheduler
- Added early stopping using validation loss
- Training, scheduler, validation loss computation, and early stopping are now defined in steps
- Added val loss hyperparameter to allow different train and val losses

📉 Added sCRPS loss in PyTorch to minimize errors generating prediction intervals.

📚 New tutorials and documentation:

- Probabilistic Long-horizon forecasting
- Save and Load Models to use them in different datasets.
- Temporal Fusion Transformer
- Exogenous variables
- Automatic hyperparameter tuning
- Intermittent or Sparse Time Series.

We hope these new features will help users tackle time series forecasting problems easily and flexibly. Please check out the documentation and start using the new version of NeuralForecast today! 💻👨‍💻

⭐️ github.com/Nixtla/neuralf…