Time Series Module Features & Roadmap

[ngupta23]

Release (Available on pip)

pip install --pre pycaret

Feature	Supported
Univariate Forecasting without Exogenous	✅
Univariate Forecasting with Exogenous	✅
Interactive Exploratory Visualizations using plotly & streamlit	✅
Statistical Testing	✅
Models - Classical Statistical	✅
Models - Regression based forecasting	✅
Models Hyperparameter Tuning	✅
Model Fit Assessment Aids using Residuals	✅
Online Learning	✅
Experiment Logging using MLFlow	✅
Deployment on Cloud	✅
Customization for Experienced Modelers	✅
AutoML for Scaling	✅

• Sample Notebook showing capabilities can be found here

We are looking for core contributors to help take the time series roadmap forward. If you are interested in building out the following modules together, please do get in touch with us.

• Time Series Hierarchical Forecasting Module
• Time Series Global Forecasting Module
• Time Series Clustering Module
• Time Series Anomaly Detection Module

Future Releases

Releases after 3.0.0

Visualization Enhancements

New Additions

• Seasonality Box Plots & Heatmap #1750
• Time Series - Single call to plot all plots... #1819
• Time Series | Add insample prediction intervals for Anomaly Detection #1836
• Display multiple CI at plot_model() when plot='forecast' for ts module #1889
• Allow ability to run ACF and PACF on the seasonal components of decomposition - Useful to determine P and Q #2245
• Replicate TSstudio plots from R #2264

New Additions requiring format change

• Visualize Error Distribution & backtesting results across folds  #1405

Statistical Testing Enhancements

• White noise test lags in time series #1775
• Allow ability to run tests such as ADF on the seasonal components of decomposition #2246

Customization Enhancements

• [Time Series] compare_models top N by category #1436
• Add lower limit and upper limit for time series forecasting #1766
• [Forecasting] Add setup argument for lags that is used in the rest of the methods #2147

Predictions

• Time Series Predictions | Add Alpha value in dataframe #1833

Models

• Time Series | Move Auto-ARIMA to turbo=False bucket #1868

Exogenous Variables

• (Suggestion) Feature Engineering: Use tsfresh to create features for time-series data #382
• Incorporate FeatureSelector in time series #2230

Bug Fixes

• [BUG] tune_model bug in Voting Ensemble Model #1703

Misc

• [ENH] Support for non integer seasonal periods #2165

Benchmarking

• Benchmark Time Series Results against Darts #2265

Delta Release

AutoML

• AutoML for Time Series #1670
• Backtesting Time Series Strategies #1669

Models & Tuning

• Incorporate window_length as a tuning parameter in tune_model #1373
• Cleanup up default search space for all time_series models #1374
• [FEAT] Set logistic trend for prophet container #1280
• Determine max_p and max_q for AutoARIMA models #2247
• Allow ability to run tests such as ADF on the seasonal components of decomposition #2246

Backlog

• Support neural network based timeseries forecasting #1215
• Add Support for Multivariate Time Series #2096
• Add Global Forecasting #1721
• Add neuralprophet to time_series module #1724
• Add support for VAR model in time series module #1731