ExaGeoStat-R is an R-Wrapper for ExaGeoStat framework, a parallel high performance unified framework for geostatistics on manycore systems.
- Portable Hardware Locality (hwloc).
- NLopt.
- GNU Scientific Library (GSL).
- StarPU.
- Chameleon.
- Hicma.
- Stars-H. An easy installation of the above packages is available by using build-deps.sh
install.packages("devtools")
library(devtools)
install_git(url="https://github.com/ecrc/exageostatR")
library(exageostat)- Download exageostat_1.0.0.tar.gz from release
- Use R to install exageostat_1.0.0.tar.gz
install.packages(repos=NULL, "exageostat_1.0.0.tar.gz")
library(exageostat)Operations:
- Generate synthetic spatial datasets (i.e., locations & environmental measurements).
- Maximum likelihood evaluation using dense matrices.
- Maximum likelihood evaluation using compressed matrices based on Tile Low-Rank(TLR).
- Maximum likelihood evaluation using matrices based on Diagonal Super-Tile(DST).
A more detailed description of the underlying ExaGeoStat software package can be found. here
- Test Generating Z vector using random (x, y) locations with exact MLE computation.
library("exageostat") #Load ExaGeoStat-R lib.
seed = 0 #Initial seed to generate XY locs.
theta1 = 1 #Initial variance.
theta2 = 0.1 #Initial range.
theta3 = 0.5 #Initial smoothness.
dmetric = 0 #0 --> Euclidean distance, 1--> great circle distance.
n = 1600 #n*n locations grid.
ncores = 2 #Number of underlying CPUs.
gpus = 0 #Number of underlying GPUs.
ts = 320 #Tile_size: changing it can improve the performance. No fixed value can be given.
p_grid = 1 #More than 1 in the case of distributed systems
q_grid = 1 #More than 1 in the case of distributed systems ( usually equals to p_grid)
clb = vector(mode="double",length = 3) #Optimization function lower bounds values.
cub = vector(mode="double",length = 3) #Optimization function upper bounds values.
theta_out = vector(mode="double",length = 3) #Parameter vector output.
globalveclen = 3*n
vecs_out = vector(mode="double",length = globalveclen) #Z measurements of n locations
clb = as.double(c("0.01", "0.01", "0.01")) #Optimization lower bounds.
cub = as.double(c("5.00", "5.00", "5.00")) #Optimization upper bounds.
vecs_out[1:globalveclen] = -1.99
theta_out[1:3] = -1.99
exageostat_initR(ncores, gpus, ts)#Initiate exageostat instance
#Generate Z observation vector
vecs_out = exageostat_egenzR(n, ncores, gpus, ts, p_grid, q_grid, theta1, theta2, theta3, dmetric, seed, globalveclen) #Generate Z observation vector
#Estimate MLE parameters (Exact)
theta_out = exageostat_emleR(n, ncores, gpus, ts, p_grid, q_grid, vecs_out[1:n], vecs_out[n+1:(2*n)], vecs_out[(2*n+1):(3*n)], clb, cub, dmetric, 0.0001, 20)
#Finalize exageostat instance
exageostat_finalizeR()- Test Generating Z vector using random (x, y) locations with TLR MLE computation.
library("exageostat") #Load ExaGeoStat-R lib.
seed = 0 #Initial seed to generate XY locs.
theta1 = 1 #Initial variance.
theta2 = 0.03 #Initial range.
theta3 = 0.5 #Initial smoothness.
dmetric = 0 #0 --> Euclidean distance, 1--> great circle distance.
n = 900 #n*n locations grid.
ncores = 4 #Number of underlying CPUs.
gpus = 0 #Number of underlying GPUs.
dts = 320 #Tile_size: changing it can improve the performance. No fixed value can be given.
lts = 600 #TLR_Tile_size: changing it can improve the performance. No fixed value can be given.
p_grid = 1 #More than 1 in the case of distributed systems.
q_grid = 1 #More than 1 in the case of distributed systems ( usually equals to p_grid).
clb = vector(mode="double", length = 3) #Optimization function lower bounds values.
cub = vector(mode="double", length = 3) #Optimization function upper bounds values.
theta_out = vector(mode="double", length = 3) #Parameter vector output.
globalveclen = 3*n
vecs_out = vector(mode="double", length = globalveclen) #Z measurements of n locations.
clb = as.double(c("0.01", "0.01", "0.01")) #Optimization lower bounds.
cub = as.double(c("5.00", "5.00", "5.00")) #Optimization upper bounds.
tlr_acc = 7 #Approximation accuracy 10^-(acc)
tlr_maxrank = 450 #Max Rank
vecs_out[1:globalveclen] = -1.99
theta_out[1:3] = -1.99
#Initiate exageostat instance
exageostat_initR(ncores, gpus, dts)
#Generate Z observation vector
vecs_out = exageostat_egenzR(n, ncores, gpus, dts, p_grid, q_grid, theta1, theta2, theta3, dmetric, seed, globalveclen)
#Estimate MLE parameters (TLR approximation)
theta_out = exageostat_tlrmleR(n, ncores, gpus, lts, p_grid, q_grid, vecs_out[1:n], vecs_out[n+1:(2*n)], vecs_out[(2*n+1):(3*n)], clb, cub, tlr_acc, tlr_maxrank, dmetric, 0.0001, 20)
#Finalize exageostat instance
exageostat_finalizeR()- Test Generating Z vector using random (x, y) locations with DST MLE computation.
library("exageostat") #Load ExaGeoStat-R lib.
seed = 0 #Initial seed to generate XY locs.
theta1 = 1 #Initial variance.
theta2 = 0.03 #Initial range.
theta3 = 0.5 #Initial smoothness.
dmetric = 0 #0 --> Euclidean distance, 1--> great circle distance.
n = 900 #n*n locations grid.
ncores = 4 #Number of underlying CPUs.
gpus = 0 #Number of underlying GPUs.
ts = 320 #Tile_size: changing it can improve the performance. No fixed value can be given.
p_grid = 1 #More than 1 in the case of distributed systems.
q_grid = 1 #More than 1 in the case of distributed systems ( usually equals to p_grid).
clb = vector(mode="double", length = 3) #Optimization function lower bounds values.
cub = vector(mode="double", length = 3) #Optimization function upper bounds values.
theta_out = vector(mode="double", length = 3) #Parameter vector output.
globalveclen = 3*n
vecs_out = vector(mode="double", length = globalveclen) #Z measurements of n locations.
clb = as.double(c("0.01", "0.01", "0.01")) #Optimization lower bounds.
cub = as.double(c("5.00", "5.00", "5.00")) #Optimization upper bounds.
dst_thick = 3 #Number of used Diagonal Super Tile (DST).
vecs_out[1:globalveclen] = -1.99
theta_out[1:3] = -1.99
#Initiate exageostat instance
exageostat_initR(ncores, gpus, ts)
#Generate Z observation vector
vecs_out = exageostat_egenzR(n, ncores, gpus, ts, p_grid, q_grid, theta1, theta2, theta3, dmetric, seed, globalveclen)
#Estimate MLE parameters (DST approximation)
theta_out = exageostat_dstmleR(n, ncores, gpus, ts, p_grid, q_grid, vecs_out[1:n], vecs_out[n+1:(2*n)], vecs_out[(2*n+1):(3*n)], clb, cub, dst_thick, dmetric, 0.0001, 20)
#Finalize exageostat instance
exageostat_finalizeR()- Test Generating Z vector using given (x, y) locations with exact MLE computation.
library("exageostat") #Load ExaGeoStat-R lib.
theta1 = 1 #Initial variance.
theta2 = 0.1 #Initial range.
theta3 = 0.5 #Initial smoothness.
dmetric = 0 #0 --> Euclidean distance, 1--> great circle distance.
n = 1600 #n*n locations grid.
ncores = 2 #Number of underlying CPUs.
gpus = 0 #Number of underlying GPUs.
ts = 320 #Tile_size: changing it can improve the performance. No fixed value can be given.
p_grid = 1 #More than 1 in the case of distributed systems
q_grid = 1 #More than 1 in the case of distributed systems ( usually equals to p_grid)
clb = vector(mode="double",length = 3) #Optimization function lower bounds values.
cub = vector(mode="double",length = 3) #Optimization function upper bounds values.
theta_out = vector(mode="double",length = 3) #Parameter vector output.
globalveclen = n
vecs_out = vector(mode="double",length = globalveclen) #Z measurements of n locations.
x = rnorm(n = globalveclen, mean = 39.74, sd = 25.09) #x measurements of n locations.
y = rnorm(n = globalveclen, mean = 80.45, sd = 100.19) #y measurements of n locations.
clb = as.double(c("0.01", "0.01", "0.01")) #Optimization lower bounds.
cub = as.double(c("5.00", "5.00", "5.00")) #Optimization upper bounds.
vecs_out[1:globalveclen] = -1.99
theta_out[1:3] = -1.99
#Initiate exageostat instance
exageostat_initR(ncores, gpus, ts)
#Generate Z observation vector based on given locations
vecs_out = exageostat_egenz_glR(n, ncores, gpus, ts, p_grid, q_grid, x, y, theta1, theta2, theta3, dmetric, globalveclen)
#Estimate MLE parameters (Exact)
theta_out = exageostat_emleR(n, ncores, gpus, ts, p_grid, q_grid, x, y, vecs_out, clb, cub, dmetric, 0.0001, 20)
#Finalize exageostat instance
exageostat_finalizeR()#!/bin/bash
#SBATCH --job-name=job_name
#SBATCH --output=output_file.txt
#SBATCH --partition=XXXX
#SBATCH --nodes=4
#SBATCH --ntasks=4
#SBATCH --ntasks-per-node=1
#SBATCH --cpus-per-task=31
#SBATCH --time 00:30:00
srun Rscript Rtest.r
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent