NEWS

ldmppr 1.1.3.9000

Fixed gcc-UBSAN undefined behavior in C++ spatial hash key construction during thinning by replacing signed left-shift operations with unsigned-safe key packing.

Refined API exports to expose only intended user-facing functions.
Improved simulation edge-case handling and input validation consistency for helper functions.
Removed package-load side effects from OpenMP thread handling.

Updated the workflow pipeline to allow passage of S3 class objects between functions to simplify the user experience. The four main steps of the workflow (fit process model, train mark model, check model fit, simulate realizations) now each have their own S3 class and associated methods and you can pass the objects forward between functions. See the updated documentation for details.

Introduced S3 classes and associated methods for the 4 main workflow steps (fit process model/train mark model/check model fit/simulate realizations).
Replaced the estimate_parameters_sc() and estimate_parameters_sc_parallel() functions with the unified estimate_process_parameters() function. We redesigned this function to provide multiple strategies for the optimization procedure and refactored the underlying C++ code to improve efficiency.
Removed explicit dependence on the Bundle package and introduced the save_mark_model() and load_mark_model() functions to handle saving and loading trained mark models.
Updated the small example dataset and example trained mark model to reflect changes in the package functions.

updated train_mark_model() and check_model_fit() and simulate_mpp() to include scaled_rasters argument to determine if scaling needs to be performed.
added a new example dataset entitled medium_example_data and corresponding raster files.
updated the plot_mpp() function to use the operator %>% instead of the |> operator to ensure compatibility with older versions of R.