This repository contains R source code for the Horizon-optimized Convex Combinations (HOC2) framework, a forecast combination framework for time-series, authored by Rafael de O. Valle dos Santos, Celso F. Araujo F., Ricardo M. S. Accioly e Fernando Luiz Cyrino Oliveira.
For more information about HOC2, read/cite the paper:
- Horizon-optimized weights for forecast combination with cross-learning (To appear in "Pesquisa Operacional").
The data used in the paper can be found at the M4 Competition repo (M4 dataset) and at the MComp R package (M1 and M3 datasets).
There is a folder for each phase of the framework, 01-methods, 02-clustering, 03-training and 04-test, each one including source code and binary files.
Attention! For simplcity sake, all the source files here are set to use YEARLY data from the M4 dataset, with only 10% of the original data for the training phase.
The methods folder includes a source file (run-methods.R) to fit the individual methods to be combined (ETS, AUTOARIMA, THETA, TBATS and SNAIVE) and another (simpleAVG.R) to compute their simple average. Results are stored in binary (RData) and text (CSV) files.
The clustering folder includes a source file (run-cluster.R) to cluster time series by their ETS model form. Results are stored in a binary (RData) file.
The training folder contains one training script for each non-empty cluster (in this case, clusters 1, 2, 3, 10, 11 and 12). E.g.: train01.R trains time-series data from cluster 1.
Each data cluster may be trained in paralalel, but all resulting training files - e.g. trainLog01.RData for cluster 1 - should be joined with join.R to generate a single result binary file (trainLog.RData), that includes the calculated mean weights for each cluster. This last file shall then be used as input (alongside others) for the test phase.
Important! All folders contains scripts to report results - e.g. report-test.R in the test folder.
Enjoy!