[1.1.0] - 2025-11-08

Added

• Support for zeta units. New keyword units in encoding, decoding, and plotting functions allows to
  specify if the PDF is sampled in units of redshift (z) or ζ = ln(1+z).

• Auto-optimization of the number of quantiles. Batch-encoding functions now accept a new keyword optimize that instructs ColdPress to find the optimal number of quantiles for encoding each PDF within the given packet length. It finds a compromise between the competing goals of minimizing epsilon and maximizing the quantile count.

Changed

• Reorganized CLI options. Some arguments for the encode and decode modes of the CLI have been redefined to accomodate both redshift and zeta units.

• Modified treatment of tolerance parameter. Now tolerance refers to the maximum error in ln(1+z), instead of z, for the quantiles. The condition on epsilon has been updated to epsilon < 2 * tolerance.

Fixed

• Ambiguous encoding for zero-sized last jump. In some very weird PDFs, the last two quantiles can be close enough for their difference being rounded to 0 during quantization. A final jump of 0 is imposible to distinguish from trailing zeros indicating empty bytes in the packet. This causes the last quantile to be missed in decoding. Fixed by turning a zero-sized last jump into an 1.

• Fixed minor issues in the decode module.

• Added access via API to additional functions. These were already implemented but not directly accesible from the API.

Release v1.0.1

This is a patch release that addresses two key bugs in the command-line interface, improving robustness and compatibility with FITS standards.

Fixes Included in This Release

• Case-Insensitive FITS Column Handling: All command-line interface commands now handle FITS column names in a case-insensitive manner.

• Variable-Length Column Decoding: A decoding error that occurred in the decode, measure, and plot commands has been fixed. This error affected encoded PDF columns stored on FITS files using a variable-length format.

Thanks to Tamara Civera for invaluable help in tracing down the variable-length format bug.

Coldpress 1.0.0 — Initial Stable Release

🚀 Added

• Implemented decode_to_density()

🐛 Fixed

• Added encode_from_density() to API
• Fixed insufficient tolerance in range check
• Fixed error in decode_quantiles() that resulted in unsorted quantiles
• Prevent _batch_encode() from crashing on un-encodable PDFs (rejects them instead)
• encode_from_samples() now excludes sources if >10% of their samples are NaN

🙏 Thanks

Special thanks to beta testers Jaan Laur and Zhang Liu for their feedback during the beta phase.

[1.0.0-beta] - 2025-07-19

Changed

• BREAKING CHANGE: The encoding algorithm now stores differences in log(1+z) instead of z. This improves accuracy and extends the effective redshift range. The size and structure of the header of compressed PDF packets has also changed. PDFs encoded with previous versions are no longer compatible and must be re-encoded.
• The decoding algorithm now detects and corrects the seesaw pattern caused by rounding of inter-quantile jumps to small integers in intervals with high probability density. A mechanism for
  preventing zero inter-quantile separation is also implemented.
• The definition of Z_MODE is now more robust, calculated as the center of the narrowest inter-quantile interval within the 68% Highest Posterior Density Credible Interval.

Added

• New --interactive option for the plotcommand allows interactive visualization of PDFs.
• New --clip-fraction option for the encode command to remove outliers from random samples.
• New clip_fraction keyword in the samples_to_quantiles() function.

Fixed

• Improved accuracy of density_to_quantiles() by up-sampling the PDF grid, which is especially important for narrowly peaked PDFs.
• Tiny shift in quantiles that decode to same value prevents singularities when reconstructing the PDF.
• Encoding logic restructured for readability and performance.