Simulate Random Spatial Data

[YanSong996]

Hi,

When I used ExaGeoStatCPP to simulate a set of spatial data, it always gave me a fixed pattern, although I didn't set a seed. Without a seed, we expect a random pattern/result.

Here I provide the code for generating a set of spatial data using ExaGeoStatCPP. Please repeat the code for several times, you could see the results are all the same.

%%R
library(assertthat)
library(ExaGeoStatCPP)
install.packages("fields")
install.packages("geoR")
library(fields)
library(geoR)
print("libraries assertthat, ExaGeoStatCPP, fields, and geoR loaded successfully!")

# Variables
dimension = "2D"
ncores <- 2
ngpus <- 4
dts <- 360
lts <- 0
computation <- "exact"
p <- 1
q <- 1

problem_size <- 5e3
kernel <- "univariate_matern_stationary"
initial_theta <- c(1,0.1,0.5)

# You need to provide a log_path to save the data.
log_path <- getwd()
# data path is where to read data from
# data_path <- paste(getwd(), "/synthetic_ds_Yan1/test1")
# observations file path is where to read observation file
# observations_file <- ""
# recovery file path is where to read recovery file
# recovery_file <- ""

hardware <- new(Hardware, computation, ncores, ngpus, p, q)
exageostat_data <- simulate_data(kernel=kernel, initial_theta=initial_theta, problem_size=problem_size, dts=dts, dimension=dimension, log_path=log_path)

quilt.plot(cbind(exageostat_data$x,exageostat_data$y),exageostat_data$m)

In comparison, please repeat the following code which generates spatial data using R package "geoR" to see the difference.

%%R
# set.seed(123)
geoR_data <- grf(n = 2000, grid = "irreg", cov.pars = c(1,0.1))

# Plot the simulated spatial data
quilt.plot(geoR_data$coords,geoR_data$data)

[SAbdulah]

Thanks, @YanSong996!
Please check out your branch to sabdulah/default-random-seed, reinstall the R version, and test your example.

Now:

• Without a seed → data will be randomly generated each time.
• With a seed → data will be consistently generated based on the provided seed.

Let me know how it goes, then we can merge this branch to the main branch!

[mahmoudElkarargyBS]

Hello @YanSong996!

Thanks for raising this issue regarding randomly generated data. As @SAbdulah mentioned, we now support fully random data generation each time. I tested your code snippet, and it’s working fine now.

We also support non-random but consistent generation by passing a seed value. In your example, you can update the line to:

exageostat_data <- simulate_data(
  kernel = kernel,
  initial_theta = initial_theta,
  problem_size = problem_size,
  dts = dts,
  dimension = dimension,
  log_path = log_path,
  seed = 0
)

If you remove the seed argument, data will be generated randomly each time.

I’d also like to mention that if you check out the devel branch, you’ll see these changes already reflected. We’ll close this issue once the changes are propagated to the main branch (which will happen soon). Until then, I’ll keep this issue open so that other users facing the same problem know they can switch to devel.

Thank you for using ExaGeoStatCPP! 🚀

[SAbdulah]

@YanSong996, could you please test this to close the issue?