Hey, there! This website is about the R forecasting package, onlineforecast. The team behind the package consists of researches at DTU Compute. You can learn more about us here.

You find vignettes, examples and articles on the website.

You can install the package by:

The current stable version is on cran. It can be installed by:

install.packages("onlineforecast")

The current development version can be installed directly from the git repo by:

install.packages("remotes")
library(remotes)
remotes::install_gitlab("packages/onlineforecast", ref="develop", host="lab.compute.dtu.dk")

Setting ref="master" installs the latest stable version (can be slightly ahead of the cran version).

If installing form source (on Windows or Mac) compilers are needed (follow first step for your OS here ctsm.info). We suggest when you have downloaded and installed the package to go through the vignettes and research papers such that understand the flow of the package. The data needs to be setup correctly and understand how the package optimize the parameters. The paper onlineforecasting provides an in-depth presentation.

onlineforecast computes new forecasts whenever new observations becomes available by using recursive estimation. The package makes it easy to setup models to be used for online forecasting, where the user can create their own online forecast-set-up with their own inputs. The package is created based on the authors long experience of formulating time series models which use weather forecasts as inputs. The challenge comes from the overlapping time series as the forecast needs to be updated whenever the weather forecast is updated.

The significance of the package is the online updating setup, in which the model is updated when new observations becomes available - the coefficients are recursively updated weighting down older data, and new forecasts are generated at each point in time. Also, the package allows the user to easily define transformations and thus fit non-linear time adaptive models, e.g. \[ \hat{y}_{t+k|t} = \hat{\beta}_{0,t,k} + \hat{\beta}_{1,t,k} f(u_{t+k|t}; \alpha), \] e.g. the user would like to apply a low pass filter on the outdoor temperature to model building heat dynamics - \(\alpha\) could be the filter coefficient. The package also sets up tuning of the function parameters, i.e. the \(\alpha\).

If you have a question we recommend posting it with the tag ‘onlineforecast’ on Stack Overflow. If you want report a bug then do it as an issue on gitlab.

For other inquiries you are welcome to send us an email. You find the package email in the top right corner and our emails under about.