Introduction
With the development of observing techniques and computing devices, it has become easier and more common to obtain large datasets. Statistical inference in spatial statistics becomes computationally challenging. For decades, various approximation methods have been proposed to model and analyze large-scale spatial data when the exact computation is infeasible. However, in the literature, the performance of the statistical inference using those proposed approximation methods was usually assessed with small and medium datasets only, for which the exact solution can be obtained. Then, for real-world large datasets, the exact computation was no longer feasible. The inference with approximation methods was often validated empirically or via prediction accuracy with the fitted model.
In this competition, the goal is to reassess existing approximation methods on large spatial datasets in a uniform way that guarantees a fair comparison. The results will be compared to the exact solution provided by the ExaGeoStat software. We generated a collection of synthetic datasets on a large scale from a set of selected true models. We aim at validating the statistical performance of the state-of-the-art approximation methods in terms of modeling, inference, and prediction. The selected true models cover disparate spatial properties to ensure a fair comparison among all the competitors' methods.
Getting Started
Participants should register for the competition by filling in this Registration Google Form by December 15, 2020. The datasets links will be sent to each registered team via e-mail after the registration (but no earlier than November 23, 2020).
Timeline
Result submission is now closed.
Registration is open until December 15, 2020. All submissions should be received by 11:59 pm (UTC±00:00) on January 15, 2021 (extended to February 1, 2021). You can use this Results Google Form to complete your submission.
Instructions
More information about the competition can be found here: competition_description.pdf.
More information about the competition and the true models can be found here (added on February 11, 2021): competition_description_and_true_models.pdf.
Final Rankings
(February 11, 2021)
The winners of each of the four sub-competitions are as follows:
The winner of Sub-competition 1a:
SpatStat-Fans
Ganggang Xu, University of Miami
Bohai Zhang, Nankai University
Jiahao Cao, Renmin University of China
Shiyuan He, Renmin University of China
The winner of Sub-competition 1b:
RESSTE (submission: CL/krig/nonpara-trans-GPGP)
Denis Allard, INRAE
Thomas Opitz, INRAE
Lucia Clarotto, Mines ParisTech
Thomas Romary, Mines ParisTech
The winner of Sub-competition 2a:
RESSTE (submission: Tukey-g-h-trans-GPGP)
The winners of Sub-competition 2b (ties from three submissions):
Tohoku-University
Yasumasa Matsuda, Tohoku University
Takaki Sato, Tohoku University
Toshihiro Hirano, Kanto Gakuin University
RESSTE (submission: CL/krig/nonpara-trans-GPGP)
RESSTE (submission: Tukey-g-h-trans-GPGP)
The full assessment for each team is summarized in the following Google Sheet http://bit.ly/2MWrTeL
Datasets
Contact
If you have any questions about this competition, you can contact us at kaustcompspat@gmail.com.