July 2023

Released on 2023-07-14

Uploaded to PyPI : pip install pyaf

Changelog :

1. Python 3.11 Support :
   Python 3.11 support #227
2. New Hardware Support :
   RISC-V Hardware Platform Validation #208
3. New Performance Measures :
   Outlier-resistant forecasting Performance Measures #209,
   Add Differentiable Variant of SMAPE Performance Measure #221
4. Model Selection Improvement :
   Investigate Model Esthetics for PyAF #212,
   Investigate Large Horizon Models #213 ,
   Revisit Model Complexity Definition #223,
   Use MASE by default for PyAF Model Selection #229
5. Signal Transformation Improvements :
   Use MaxAbsScaler for some Multiplicative Signal Transformations #235,
   Pyaf 5.0 Final Touch 8 : Use an Optimal Choice Rule for the Quantization Signal transform #239
6. Generic Modeling :
   PyAF 5.0 Final Touch 1 : discard some non-significant components #230,
   PyAF 5.0 Final Touch 2: Disable alpha in ridge regressions #231,
   Pyaf 5.0 Final Touch 5 : Add more info about Exogenous Data Used in ARX Models #236,
   Pyaf 5.0 Final Touch 7 : Improve the Guess of Window Length for Moving Average Trends #238
7. Plotting Functions Improvements and Bug Fixes :
   Bad plot for shaded area around prediction intervals in hourly data #216,
   Forecast Quantiles Plots Improved #225,
   Pyaf 5.0 Final Touch 3 : report plot filenames in the logs #232
8. New Docs :
   Provide some UML docs for PyAF integrators #233
9. Bug Fixes :
   Failure to build a multiplicative ozone model with Lag1 trend #220
10. PyAF "Forecast Tasks" :
    Use PyTorch as the reference deep learning framework/architecture for future projects #211,
    Automate Prototyping Activities - R-based Models #217
11. Recurrent Tasks :
    Re-run the Benchmarking process for PyAF 5.0 #222,
    Run some Sanity Checks for PyAF 5.0 #224,
    Pyaf 5.0 Final Touch 4 : Add More Tests #234,
    Pyaf 5.0 Final Touch 6 : Disable Timing Loggers by default #237

July 2022

RELEASE 4.0 ( 2022-07-14 )

1. Python 3.10 support #186
2. Add Multiplicative Models/Seasonals #178
3. Speed Performance Improvements : #190 , #191
4. Exogenous data support improvements : #193, #197, #198
5. PyAF support for ARM64 Architecture #187
6. PyTorch support : #199
7. Improved Logging : #185
8. Bug Fixes : #156, #179, #182, #184
9. Release Process : Pre-release Benchmarks #194
10. Release Process : Profiling and Warning Hunts #195
11. Release Process : Review Existing Docs #196, #35

July 2021

RELEASE 3.0 ( 2021-07-14 )

1. Python 3.9 support #149
2. Probabilistic Forecasting : Forecast quantiles (#140), CRPS (#74), Plots and Docs (#158).
3. Add LightGBM based models #143
4. Add more Performance Measures : MedAE (#144) , LnQ ( #43 )
5. PyAF Powerpc support (IBM S822xx) #160
6. More Parallelization Efforts (#145)
7. Add Missing Data Imputation Methods (#146 )
8. Improved long signals modeling (#167)
9. Warning Hunts (#153)
10. Some Bug Fixes (#163, #142, #168).
11. Switched to Circle-CI (#164)
12. Plot Functions Improvement #169
13. Model Complexity Improvement (#171)
14. Documentation review/corrections (#174)

July 2020

RELEASE 2.0 (2020-07-14)

1. Time column is normalized frequently leading to a performance issue. Profiling. Significant speedup. Issue #121
2. Corrected PyPi packaging. Issue #123
3. Allow using exogenous data in hierarchical forecasting models. Issue #124
4. Properly handle very large signals. Add Sampling. Issue #126
5. Add temporal hierarchical forecasting. Issue #127
6. Analyze Business Seasonals (HourOfWeek and derivatives) . Issue #131
7. Improved logs (More model details). Issue #133, #134, #135
8. More robust cycles (use target median instead of target mean encoding). Issue #132
9. Analyze Business Seasonals (WeekOfMonth and derivatives). Issue #137
10. Improved JSON output (added Model Options). Issue #136
11. Improved CPU usage (parallelization) for hierarchical models. Issue #115
12. Speedups in multiple places : forecasts generation, plotting, AR Modelling (feature selection).

April 2020 - Fixes - 3

A few fixes, Added long description for PyPI

April 2020 - Fixes - Bis

Minor release following Travis-ci tests.

April 2020 - Fixes

PyAF now has a pypi installer. You can now use :

pip install pyaf

to install it.

Addiitonal tweaks ... double check PyPI / twine / demo scripts.

July 2019

Some bug fixes
Some optional Models : XGBoost (#98) + Croston(#97 )
Some optional Modeling features : Cross Validation (#105 )

Jan 2018

Forecast dates are shifting #86

use pandas.DateOffset instead of numpy.timedelta

Updatd these logs

First Benchmarked Release

Put in place a "serious" benchmark process #45

Benchmark data under https://github.com/antoinecarme/PyAF_Benchmarks