Nested multiresolution hierarchical simulated annealing
The {nmhsa} R package provides an R wrapper around the NMHSA Python implementation by Lemmens et al. (2019) for performing porous media reconstruction using an advanced simulated annealing approach.
You can install the latest version of {nmhsa} with the following:
if (!require(pak)) install.packages("pak")
# pak::pak("rogiersbart/nmhsa")
install.packages("nmhsa", repos = c('https://rogiersbart.r-universe.dev', 'https://cloud.r-project.org'))
To run the algorithms from the examples in the original Python code, you
at least need a 2D array as training image, and resort to the hsa(),
mhsa() or nmhsa() functions, through which the algorithms are
exposed to the R user. Hyperparameters and corresponding defaults have
been thorougly revised, so we recommend you read the help pages of these
functions in detail. For making use of the nesting available in
nmhsa(), the training image has to be processed, and one or two
modified versions of the image have to be provided as well. Here’s a
minimal hsa() example with the cement training image provided in the
package:
library(nmhsa)
#> ! {nmhsa} is still in its experimental lifecycle stage.
#> ! Use at your own risk, and submit issues here:
#> ! <https://github.com/rogiersbart/nmhsa/issues>
reconstruction <- hsa(cement)
#> v Preparing the Python backend ... done
#> v Reconstructing ... done
plot(cement, reconstruction)
For more information, refer to vignette("nmhsa").
This package depends on {reticulate} and Python, and hence a functional Python installation is required on your machine. If you don’t have anything like that, {reticulate} will normally solve that for you.