v2.1.4

• Support interpolation from a pre-computed trend fit via interpolate().

v2.1.3

• Updated required version of subpar.
• Added dependency on sanisizer to protect against integer overflows.

v2.1.2

• Improve handling of y with no variance after robustness weighting. This allows early termination if all non-outlier points are the same, and it ensures that outliers do not forcibly inflate MAD estimates during iterations.

v2.1.1

• Relaxed SortBy methods to accept workspace pointers of different types.
• Minor fixes for includes of standard library headers.
• Use the subpar library for a centralized parallelization scheme. This sets the WEIGHTEDLOWESS_CUSTOM_PARALLEL macro by default to subpar::parallelize(), with all the associated improvements.
• Sprinkle in OpenMP SIMD pragmas for potential efficiency improvements.

v2.1.0

• Enable precomputation of window locations based on x-coordinates.
• Enable more flexible construction and reuse of SortBy instances.
• Specialize the single-array case for SortBy::(un)permute for efficiency.

v2.0.0

• Refactor to use regular functions that accept an Options class instance.
• Optionally support parallelization for all loops across data or anchors.
• Removed in-place functionality in favor of sorted input, which is now facilitated by the SortBy class.

v1.0.1

• Create a dedicated Defaults class for default settings.
• Added more options for controlling the span and window size.
• Constified all functions, at the cost of memory usage.
• Small bugfix for span calculation near start of range.
• Added CMake installation commands for find_package().
• Fix all compiler warnings.
• Minor correction to the handling of delta = 0.

v1.0.0

Now that we've got some code coverage, let's make a new release 🚀 🎉