Software

Neural Prophet

Neural Prophet is a reimplementation of the popular Prophet forecasting software, that we develop together with researchers from Stanford University and Meta Platforms Inc.

Kit to Analyze Time Series Data (KATS)

KATS is developed by Meta Platforms Inc., and I contributed a LightGBM forecasting wrapper module called ML_AR.

R packages

The following are R packages that I have (co)authored and/or that I maintain. Some of them have their most recent versions on github. Most packages are also on CRAN.

Time series forecasting

  • Rlgt implements Local and Global Trend models, which is a novel class of exponential smoothing models, and fits them with Bayesian model fitting [CRAN]
  • forecast is the most popular package for forecasting in R. It is developed by Rob J. Hyndman. I re-implemented part of the ets() function in C++, achieving considerable speedups [CRAN]
  • tsDyn implements threshold autoregressive models. I implemented some patches and some (currently unpublished) extensions, mainly regarding neuro-coefficient STAR models [CRAN]
  • tsExpKit is a framework for time series experiments, aimed at the facilitation of performing well-documented, reproducible experimentation [github]
  • Rsiopred is an R package for forecasting by exponential smoothing with model selection by a fuzzy multicriteria approach (currently unpublished).
  • Mcomp contains data of the M3 forecasting competition. Its author is Rob J. Hyndman. I included the forecasts of the original competition participants into the package [CRAN]

General-purpose machine learning

  • RSNNS is a comprehensive package for neural networks in R [CRAN]
  • Rmalschains implements memetic algorithms in R [CRAN]
  • frbs is a comprehensive package for fuzzy rule-based systems in R. It is mainly developed by Lala S. Riza, and I maintain it. [CRAN]
  • RoughSets is a comprehensive package for rough set and fuzzy rough set theory in R. It is mainly developed by my collaborators, and I maintain it [CRAN]
  • ssc is a package for semi-supervised classification in R. It is mainly developed by Mabel Gonzalez, and I maintain it [CRAN]
  • opusminer provides an interface to the OPUS Miner algorithm (implemented in C++) for finding the key associations in transaction data efficiently, in the form of self-sufficient itemsets, using either leverage or lift. It is published on CRAN, I maintain it [CRAN]
  • ChoR is a wrapper for the java-implemented Chordalysis algorithm, which learns the structure of graphical models from datasets with thousands of variables [CRAN]