NEWS

---

LBBNN 0.1.6 (2026-06-30)

• Initial tests added; cleanup for R CMD check portability.
• Initial CRAN submission.

• Added more vignettes, built pkgdown website.
• More robust testing.
• Bug fix in conv2d layer.
• Small changes to make training on gpu more efficient.
• Added possibility to turn off printing to console during training.
• Utility functions work with standard LBBNN architecture without input-skip.
• predict() and validate_lbbnn() no longer change the model object.
• fixed bug that compute_paths() was not being called for LBBNN models.
• added initialization keywords such as 'dense' or 'balanced' for inclusion probabilities.
• added possibility for different weight initializations.

LBBNN 0.1.5 (2026-04-23)

• All torch-dependent examples are now guarded by torch_available() to ensure safe execution when torch or libtorch is not available.

LBBNN 0.1.4 (2026-01-12)

• Added relu to input-skip layers. Updated experiments.

LBBNN 0.1.3 (2026-01-07)

• Changed some function names to be consistent with CRAN guidelines.
• Removed export of function that is only used internally.
• Changed assignment operator to be consistent everywhere.
• Removed datasets not used in RJ article submission.

LBBNN 0.1.2 (2025-12-10)

• Fix bug in LBBNN_Linear so layers work properly in convolutional architecture.
• Updated experiment on convolutional architecture for R journal submission.
• Added default values to mpm and draws in predict() function.