Forecasting

Techniques used to predict future values based on historical data, where the data points are ordered by time. It involves analyzing trends, patterns, and variations in time-dependent data to make informed predictions about future behavior.
See:

• GenAI for time series
• AI/Time Series/XAI for time series
• Time Series analysis
• AI/Supervised Learning/Regression
• AI/Deep Learning/RNNs
• AI/Deep Learning/CNNs#Sequence (time series) modelling
• AI/Deep Learning/Transformers

Resources

• https://en.wikipedia.org/wiki/Forecasting
• Microsoft - Time Series Forecasting Best Practices & Examples
• Forecasting with a Time Series Model using Python:
  • https://www.bounteous.com/insights/2020/09/15/forecasting-time-series-model-using-python-part-one/
  • https://www.bounteous.com/insights/2020/09/15/forecasting-time-series-model-using-python-part-two/
• https://towardsdatascience.com/the-complete-guide-to-time-series-analysis-and-forecasting-70d476bfe775
• https://towardsdatascience.com/automl-for-time-series-advanced-approaches-with-fedot-framework-4f9d8ea3382c
• Predicting energy demand with neural networks
• Time series forecasting:
  • Part 1 of 3: Understanding the fundamentals
  • Part 2 of 3: Selecting algorithms
  • Part 3 of 3: Introducing AUTS (Adaptive Univariate Time Series forecasting algorithm)
• Python Time Series Forecasting: A Practical Approach (WANDB)
• The Reasonable Effectiveness of Deep Learning for Time Series Forecasting

Code

• #CODE Skforecast
  • https://joaquinamatrodrigo.github.io/skforecast/
  • Package towards building Explainable Forecasting and Nowcasting Models with State-of-the-art Deep Neural Networks and Dynamic Factor Model on Time Series data sets with single line of code
  • https://www.kdnuggets.com/2021/12/sota-explainable-forecasting-and-nowcasting.html
  • Skforecast: time series forecasting with Python and Scikit-learn
• #CODE Pytorch-forecasting - Time series forecasting with PyTorch
  • https://pytorch-forecasting.readthedocs.io/en/stable/index.html#
• Nixtla
  • #CODE statsforecast
    • StatsForecast offers a collection of widely used univariate time series forecasting models, including autoARIMA and ETS modeling optimized for high performance using numba
    • https://nixtla.github.io/statsforecast/
  • #CODE Neuralforecast
    • NeuralForecast is a Python library for time series forecasting with deep learning models. It includes benchmark datasets, data-loading utilities, evaluation functions, statistical tests, univariate model benchmarks and SOTA models implemented in PyTorch and PyTorchLightning
    • https://nixtla.github.io/neuralforecast
  • #CODE MLforecast
    • mlforecast is a framework to perform time series forecasting using machine learning models, with the option to scale to massive amounts of data using remote clusters.
  • #CODE Hierarchicalforecast
    • Probabilistic Hierarchical forecasting 👑 with statistical and econometric methods.
• #CODE Darts
  • https://unit8co.github.io/darts/
  • Python library for easy manipulation and forecasting of time series. It contains a variety of models, from classics such as ARIMA, Prophet, deep neural networks (NBEATS, RNNs, Transformers)
  • https://towardsdatascience.com/darts-swiss-knife-for-time-series-forecasting-in-python-f37bb74c126
• #CODE Prophet (Facebook)
  • https://facebook.github.io/prophet/
  • Prophet is a procedure for forecasting time series data based on an additive model where non-linear trends are fit with yearly, weekly, and daily seasonality, plus holiday effects. It works best with time series that have strong seasonal effects and several seasons of historical data.
  • https://research.fb.com/prophet-forecasting-at-scale/
  • http://blog.fastforwardlabs.com/2017/03/22/prophet.html
• #CODE NeuralProphet
  • https://ourownstory.github.io/neural_prophet/
  • A simple forecasting model based on Neural Networks in PyTorch
• #CODE Hcrystalball
  • Library that unifies the API for most commonly used libraries and modeling techniques for time-series forecasting in Python
• #CODE AtsPy: Automated Time Series Forecasting in Python
• #CODE Greykite (Linkedin)
  • https://linkedin.github.io/greykite/
  • A flexible, intuitive and fast forecasting library
• #CODE Scalecast
  • https://towardsdatascience.com/introducing-scalecast-a-forecasting-library-pt-1-33b556d9b019
• #CODE Deep_XF

Talks

• #TALK Feature Engineering for Time Series Forecasting | PyData London 2022

Books

• #BOOK [Forescasting: principles and practice (Hyndman, R)]
• 2nd ed.
• 3rd ed.

References

• #PAPER Engineering Extreme Event Forecasting at Uber with Recurrent Neural Networks (Laptev 2017)
• #PAPER Deep and Confident Prediction for Time Series at Uber (Zhu 2017)
  • https://eng.uber.com/neural-networks-uncertainty-estimation/
  • introduced a new end-to-end Bayesian neural network (BNN) architecture that more accurately forecasts time series predictions and uncertainty estimations at scale
• #PAPER DeepAR: Probabilistic Forecasting with Autoregressive Recurrent Networks (Salinas 2017)
• #PAPER #REVIEW Time Series Forecasting With Deep Learning: A Survey (Lim 2020)
• #PAPER N-BEATS: Neural basis expansion analysis for interpretable time series forecasting (Oreshkin 2020)
• #PAPER A flexible forecasting model for production systems (Hosseini 2021)
• #PAPER An Experimental Review on Deep Learning Architectures for Time Series Forecasting (Lara-Benitez 2021)
• #PAPER Temporal Fusion Transformers (TFT) for interpretable multi-horizon time series forecasting (Lim 2021)
  • https://ai.googleblog.com/2021/12/interpretable-deep-learning-for-time.html
  • AI/Deep Learning/Transformers
• #PAPER Understanding forecast reconciliation (Hollyman 2021)
  • Forecast Reconciliation - a process by which independently generated forecasts of a collection of linearly related time series are reconciled via the introduction of accounting aggregations that naturally apply to the data
• #PAPER A Review on Deep Sequential Models for Forecasting Time Series Data (Mahdi Ahmed 2022)
• #PAPER DeepTime: Deep Time-Index Meta-Learning for Non-Stationary Time-Series Forecasting (Woo 2022)
  • #CODE https://github.com/salesforce/DeepTime
  • DeepTime is a deep time-index based model trained via a meta-learning formulation, yielding a strong method for non-stationary time-series forecasting
  • https://pub.towardsai.net/meta-learning-for-time-series-forecasting-deeptime-in-pytorch-lightning-3d97351099c6
• #PAPER Forecast evaluation for data scientists: common pitfalls and best practices (Hewamalage 2022)
• #PAPER #REVIEW Review of ML and AutoML Solutions to Forecast Time-Series Data (Alsharef 2022)
• #PAPER TSmixer - An all-MLP architecture for time series forecasting (2023)