`fieldsMAGMA': A MAGMA-accelerated extension to the `fields' spatial statistics R package / by John Paige, Douglas Nychka, and Dorit Hammerling

By: Contributor(s): Series: | NCAR Technical NotesBoulder, CO : National Center for Atmospheric Research (NCAR), 2015Content type:
  • text
Media type:
  • unmediated
Carrier type:
  • volume
ISSN:
  • 2153-2397
  • 2153-2400
Subject(s): Online resources: Abstract: This report introduces the `fieldsMAGMA' R package, an extension to the `fields' package for spatial data analysis that is available on github. fieldsMAGMA uses the Cholesky decomposition functionality of the MAGMA multi-GPU, multi-CPU computing library and eliminates some unnecessary distance and covariance calculations to create accelerated versions of spatial statistics methods in fields. We demonstrate the performance of fieldsMAGMA's accelerated functions when applied to simulated datasets and the CO2 dataset available in fields. We show that using the single precision Cholesky decomposition in particular has the potential for vast improvements in the Cholesky decomposition and in spatial likelihood computation time, yet the accuracy of the likelihood maximization is not signicantly reduced. We gather some of our timing results on a 2014 MacBook Pro with a stock graphics processing unit (GPU), an NVIDIA GeForce GT 750M with 2048 MB GDDR5 RAM.
Holdings
Item type Current library Call number Copy number Status Date due Barcode Item holds
REPORT REPORT NCAR Library Mesa Lab 03722 1 Available 50583020003897
Total holds: 0

2015-08

Technical Report

This report introduces the `fieldsMAGMA' R package, an extension to the `fields' package for spatial data analysis that is available on github. fieldsMAGMA uses the Cholesky decomposition functionality of the MAGMA multi-GPU, multi-CPU computing library and eliminates some unnecessary distance and covariance calculations to create accelerated versions of spatial statistics methods in fields. We demonstrate the performance of fieldsMAGMA's accelerated functions when applied to simulated datasets and the CO2 dataset available in fields. We show that using the single precision Cholesky decomposition in particular has the potential for vast improvements in the Cholesky decomposition and in spatial likelihood computation time, yet the accuracy of the likelihood maximization is not signicantly reduced. We gather some of our timing results on a 2014 MacBook Pro with a stock graphics processing unit (GPU), an NVIDIA GeForce GT 750M with 2048 MB GDDR5 RAM.

Questions? Email library@ucar.edu.

Not finding what you are looking for? InterLibrary Loan.