Skip to main content

SpatiotEmporal ForEcasting: Coopetition to meet Current Cross-modal Challenges

Periodic Reporting for period 2 - SEE.4C (SpatiotEmporal ForEcasting: Coopetition to meet Current Cross-modal Challenges)

Reporting period: 2017-07-01 to 2018-03-31

The objective of this project was to design a means of evaluating the performance of big data analytics algorithms in the field of spatio-temporal forecasting. This is a relevant and important problem in various fields such as smart grid management, renewable energy use optimization, media, climate science, economics, finance, and has recently been empowered in big data access, computing power and algorithmic-oriented research. The objective of our work was to harness these developments, especially in the use of open-source analytics in allowing the evaluation of state-of-the-art spatio-temporal forecasting algorithms on a test problem of industrial and social relevance.
We have assembled relevant test data sets in energy and media, put together evaluation rules for algorithms which include speed and accuracy of prediction, implemented a web-based submission evaluation platform along with tools and rules for formatting of algorithms for the scope of their evaluation, assembled a Linux-container evaluation sandbox capable of executing several open-source languages and hundred of analytics toolboxes including support for GPU-powered state-of-the-art deep learning algorithms.
The competition design allows for automatic evaluation of larger data and more powerful algorithms than the current state-of-the-art in forecasting challenges. The potential impact is the fostering of a new direction in big data analytics and data science in a prize competition accessible to individual data scientists, research laboratories and companies alike.