riemtan 0.2.5

New Features

Added comprehensive parallel processing support via the futureverse framework (future + furrr packages)
Cross-platform parallelization: Works on all platforms including Windows (replaces platform-specific parallel::mclapply)
Intelligent auto-detection: Automatically uses parallel processing when beneficial based on data size
Flexible configuration: Users can control parallelization via set_parallel_plan()
Progress reporting: Optional progress bars via progressr package integration

New module R/parallel_config.R:
- set_parallel_plan(): Configure parallel strategy (sequential, multisession, multicore, cluster)
- is_parallel_enabled(): Check parallel status
- should_parallelize(): Auto-detect when to parallelize based on threshold
- get_n_workers(): Query number of parallel workers
- reset_parallel_plan(): Reset to sequential mode
New module R/progress_utils.R:
- with_progress(): Execute expressions with progress reporting
- create_progressor(): Create progress reporters for loops
- configure_progress(): Configure progress handlers
- is_progress_available(): Check progressr availability

TangentImageHandler (R/tangent_handler.R):
- compute_tangents(): Parallel tangent space projections
- compute_vecs(): Parallel vectorization
- compute_conns(): Parallel exponential maps
- set_reference_point(): Parallel tangent relocation
- All methods support optional progress parameter
ParquetBackend (R/parquet_backend.R):
- get_all_matrices(): Parallel matrix loading from Parquet files
- get_matrices_parallel(): New method for batch parallel loading
- Automatic parallelization for I/O-bound operations
Core Algorithms (R/other_utils.R):
- relocate(): Now uses furrr::future_map() instead of parallel::mclapply (cross-platform!)
- compute_frechet_mean(): Parallel processing support with progress parameter
- Both functions support progress parameter
CSample Class (R/sample.R):
- compute_tangents(): Pass-through progress support
- compute_vecs(): Pass-through progress support
- compute_unvecs(): Parallel unvectorizati with progress
- compute_conns(): Pass-through progress support
- compute_fmean(): Pass-through progress support
- change_ref_pt(): Pass-through progress support
- load_connectomes_batched(): NEW - Batch loading with memory management for large Parquet datasets

5-10x speedup for loading large Parquet datasets (parallel I/O)
3-8x speedup for tangent computations on large samples (scales with workers)
2-5x speedup for Frechet mean computation (n > 100)
Reduced overhead for small datasets via intelligent threshold-based auto-detection

relocate() function signature changed: added optional progress parameter (default: FALSE)
- Existing code continues to work without modification
compute_frechet_mean() function signature changed: added optional progress parameter
- Existing code continues to work without modification

Removed dependency on parallel::mclapply (platform-specific, Windows incompatible)
All parallel operations now use futureverse framework for consistency
Progress reporting infrastructure integrated throughout codebase

Added support for Apache Arrow Parquet format for efficient storage and lazy loading of large connectome datasets
Introduced backend abstraction layer (DataBackend, ListBackend, ParquetBackend) to decouple storage from business logic
ListBackend: Wraps existing list-based storage for backwards compatibility
ParquetBackend: Lazy-loads matrices from Parquet files with LRU caching (default cache size: 10 matrices)
New I/O functions:
- write_connectomes_to_parquet(): Export matrices to Parquet format with metadata
- validate_parquet_directory(): Validate Parquet directory structure
- create_parquet_backend(): Convenience function to create ParquetBackend
CSample now accepts backend parameter for flexible storage options
Full backwards compatibility: existing code using lists continues to work unchanged
Metadata support: Store subject IDs, data provenance, and custom metadata with Parquet datasets
New vignette: “Using Parquet Storage for Large Datasets”

Added super_sample.R, introducing the CSuperSample class for handling and analyzing collections of CSample objects.
Added comprehensive tests for CSuperSample in test-csupersample.R.

Implements various Riemannian metrics for symmetric positive definite matrices, including:
- Affine Invariant Riemannian Metric (AIRM)
- Log-Euclidean
- Euclidean
- Log-Cholesky
- Bures-Wasserstein metrics
Provides functions for:
- Computing logarithmic and exponential maps
- Vectorization

Depends on R (>= 4.3.0)
Imports the following packages:
- Matrix
- methods
- expm
- R6
- purrr
- MASS
- furrr
Suggests the following packages for testing and documentation:
- testthat (>= 3.0.0)
- knitr
- rmarkdown