Changelog • isdmtools

isdmtools 0.4.0

compute_ppc_stats(): A wrapper to calculate Pearson Chi-squared statistics and Bayesian $p$ -values to assess model fit.
Added S3 methods: print.ppc_stats and plot.ppc_stats for immediate visual and console-based model diagnostics.
simulate_replicates(): A new vectorized simulation engine to generate $y_{rep}$ data from posterior samples. Supports poisson, nbinomial, and quasi-poisson families.
Posterior Predictive Checks (PPC): Introduced a robust Bayesian diagnostic suite for count data models.
plot.std_matern_corr: An S3 method for quick plot of the standardized outputs and pairwise correlation function.
std_matern_corr: A wrapper to standardize Matern covariance parameters across estimation engines and compute the pairwise correlation from LGCP models.

format_predictions(): Updated to ensure extracted coordinates are consistently lowercase for internal processing while preserving existing X/Y data columns.
Moved solve_practical_range() from utils.R to matern_helpers.R and optimized uniroot calls by removing the redundant sigma2 argument.
Added internal utility functions matern_cov() and get_variance() to standardize spatial covariance across engines and variance calculations for Poisson extensions.

Code styling: Established the lintr workflow for the automated detection of syntax errors and code styling issues.
GitHub Actions Integration: Established a suite of automated workflows including R-CMD-check for cross-platform stability and Codecov for monitoring unit tests coverage.
Automated Documentation: Implemented an automated pkgdown deployment pipeline to ensure the package website and vignettes are updated upon every push to the main branch.
Website Launch: Deployed an official pkgdown site with a custom navigation bar and categorized tutorials.

Tutorials: Added a “Get Started” guide and an advanced “ISDM Evaluation Workflow” vignette with conditional evaluation for external dependencies.
New S3 Class Architecture: Introduced GeoDiagnostic, EnvDiagnostic, and FoldsSummary classes to structure the spatial folds’ diagnostics results.
Documentation Refactor: Implemented a grouped reference system to consolidate S3 methods for improving the package Reference index.

R Compatibility: Full support for the native pipe |> across the package.
Codecov coverage: Set up additional units tests to boost the Codecov test coverage.
BugFixes: Refactoring ISDMmetrics methods to handle efficiently datasets’ names with underscores to avoid names collision during string splitting.
Blocks Diagnostics: Refined check_folds() and check_env_balance() constructors for evaluating spatial independence and environmental balance of spatial folds.

Added folds diagnostics tools: check_folds and check_env_balance are key methods operating on DataFolds objects to check the independence and representativeness of folds.
Buffer-Aware Extraction: Updated extract_fold() to automatically handle exclusion zones. Points falling within spatial buffers are now correctly identified as NA and excluded from both training and testing sets to prevent spatial autocorrelation bias.
New CV Methods: Added support for four new cv_method options for advanced spatial cross-validation:
- "block": Grid-based blocking via “spatialsample” (similar to “spatial” in blockCV but with exclusion buffer).
- "nndm": Nearest Neighbor Distance Matching for matching prediction and validation environments.
- "buffer": Distance-based exclusion zones (Leave-One-Out with buffer).
- "location": Leave-location-out/Leave-group-out CV, enabling spatiotemporal validation (e.g., by year) and source-specific validation (e.g., by observers, sites or regions).
Expanded Cross-Validation Engines: Integrated ‘spatialsample’ as a backend, providing a unified interface for advanced spatial resampling methods.
Added summary.DataFolds method for providing clean aggregated statistics on data partitioning.
Added relevant tests for DataFolds and BackgroundPoints classes and their constructors/methods.

Refactoring: Remove unnecessary dependencies (reshape2, purrr and ggspatial) and ensure consistent argument naming (snake_case).
Standardized S3 Outputs: The DataFolds object now standardizes internal indexing across different blocking engines, ensuring “folds_ids” are consistent regardless of the underlying package used.
Improved Print Method: The print.DataFolds method now explicitly labels “Excluded” points, providing a clear summary of how many observations were buffered out of the validation process.
Added robust unit tests for spatial buffer logic and multisource data fusion integrity.
Unified CV Constructor: create_folds() now acts as a high-level bridge between blockCV and spatialsample.

Integrated unit testing framework using testthat.
Standardized coordinates column names to lowercase x and y across all functions.
Updated internal logic to use .data pronoun to resolve global variable notes in R-CMD-check.

Implemented S3 methods for ISDMmetrics: print(), summary(), plot(), as.data.frame(), and [.
Added ISDMmetrics S3 class to handle outputs from compute_metrics().
Implemented get_background helper function for selected background points.
Enhanced suitability_index(): Added "linear" to output.format options.
Updated DESCRIPTION with new minimum version requirements for package dependencies (sf, terra, purrr) to ensure compatibility with recent environment updates.

Initial release supporting “Integrating Presence-only and Abundance Data to Predict Baobab (Adansonia digitata L.) Distribution: A Bayesian Data Fusion Framework”. DOI: 10.21203/rs.3.rs-7871875/v1