Forecasting Versions Save

Time Series Forecasting Best Practices & Examples

v0.2.0

3 years ago

July 2020 release

In this release, we added a new example to the R codebase, and continued to harden the quality and testing of the codebase from previous release. The added example is an introduction to forecasting with the Tidyverts framework, using monthly Australian retail turnover by state and industry code. The dataset is one of many included in the tsibbledata package of example time series datasets, which we wanted to introduce to the users through this added example. Another bigger change to the repository is a greatly improved unit test coverage for the fclib module, and we also included coverage computation in our build pipelines. Additionally, we addressed a number of bugs and issues raised by the repository users, for which we are greatly thankful. Detailed changes included in this release are outlined below.

New Features

  • Added a new data set example to the R portion of the repository that goes over the basics of time series analysis, using the tsibbledata::aus_retail dataset. It also includes a closing comment on the hazards of forecasting in the presence of COVID-19. #200

Enhancements

  • Html outputs linked in R examples table #192
  • Refactored contrib/ directory to prepare it for external contributions #199
  • Carried out a cleanup of large obsolete files to reduce the size of this repo. #206
  • Testing and refactoring fclib which added full unit testing and code coverage for fclib, as well as refactored some fclib code to improve functionality and user friendliness #214

Bugfixes

  • Fixed version conflicts caused by pmdarima package #211
  • Fixed bad definition of sMAPE function #213
  • Fixed broken link to the lightgbm example in the README #201
  • Implemented change in environment_setup.sh to stop execution if conda env is not created #194
  • Dilated CNN example improved to run more smoothly on Windows #182 #183

v0.0.1

4 years ago

TSPerf - a repository of time-series forecasting models with a comprehensive comparison of their performance over provided benchmark data sets, implemented on Azure.