Skip to content

Latest commit

 

History

History
48 lines (34 loc) · 2.22 KB

README.md

File metadata and controls

48 lines (34 loc) · 2.22 KB

R scripts for forecasting in Q-Rapids

R script file including a set of forecasting methods. This script acts as a wrapper of some state of the art forecasting methods, implemented in packages like the well-known forecast. Apart from that, it also implements some other useful aspects, like wrapping the elastic package to allow easy metrics, factors, and strategic indicators gathering from an elasticsearch DB. The script also implements saving and reusing the fitted forecasting models, in order to save time and computation resources.

The script is intended to be used along with the q-rapids-forecast library, but is prepared to be used alone as well.

Available forecasting methods

The available methods and the corresponding R package implementing them are as follow:

Method Name R Package
Arima forecast
Arima (force seasonality) forecast
Theta forecast, forecastHybrid
ETS forecast
ETS (force damped) forecast
Bagged ETS forecast
STL forecast
Neural Network (nnetar) forecast
Hybrid forecastHybrid
Prophet prohpet

Dependencies

The script depends on the following packages:

  • forecast
  • devtools
  • elastic
  • forecastHybrid
  • prophet

These required packages should be installed beforehand using the following commands:

 install.packages("devtools")
 library(devtools)
 install_version("forecast", version="8.7")
 install_version("prophet", version="0.4")
 install_version("Rserve", version="1.7-3.1")
 install_version("elastic", version="0.8.4")
 install_version("forecastHybrid", version="4.2.17")

Licensing

Software licensed under GNU Lesser General Public License v3.0. Permissions of this copyleft license are conditioned on making available complete source code of licensed works and modifications under the same license or the GNU GPLv3. Copyright and license notices must be preserved. Contributors provide an express grant of patent rights. However, a larger work using the licensed work through interfaces provided by the licensed work may be distributed under different terms and without source code for the larger work.