v0.12.0

This release fixes a large backlog of bugs found in a review of the whole codebase. It also makes balancing faster, and more correct in the trans-only case. It also removes features that didn't work well, and drops several dependencies.

Fixes

• cooler cload assigned a position at the end of a chromosome to the next chromosome's first bin, turning a cis contact into a trans one. Such positions are now rejected.
• cooler cp/mv/ln -w` could empty the destination file when the operation failed.
• cooler zoomify destroyed its input when the default output name was the input file, and --base-uri kept only one base cooler.
• Custom field aggregators: cload pairs --field ...:agg=max (or min) combined chunks by summing them, and fields given out of column order swapped their values.
• Aggregating multiple value columns: coarsen and zoomify dropped every value column except count.
• Extra long chromosomes: chromosomes longer than 2^31 bp overflowed, and pixel values too large for their column were clipped. Long chromosomes are now stored correctly, and values that don't fit raise an error.
• Guarding against unsorted input: Unsorted input could produce coolers with a corrupt index or duplicate pixels.
• A failed write no longer leaves a partial cooler behind or destroys an existing one.

Balancing

• Trans-only balancing balances to exact targets and filters bins based on trans data only. Trans-only weights change, mostly by a few percent.
• Cis-only balancing is now as fast as genome-wide balancing on assemblies with many contigs: 4.6 s instead of 316 s on 2,010 contigs.
• --force keeps the existing weights until new ones are stored, and storing weights works on assemblies with many thousands of contigs (#226).
• New options: cooler balance --rescale-marginals/--no-rescale-marginals (#448) and balance_cooler(..., ignore_dist=...). --ignore-dist works with variable-size bins (#464).

New features

• cload pairs --field ...:agg=sum_squares. With a sum and a count, it can be used to give a column's mean and variance, and all three stay exact when coarsening and merging.
• Range queries are faster: matrix queries on HDF5 files take about two thirds of the time.
• 2D range queries can return several value columns at once.
• cooler.log_to_stderr() shows cooler's log messages in scripts and notebooks. Importing cooler no longer configures logging.

Removed

• cooler csort is gone. cooler cload pairs can ingest unsorted pairs directly. To sort and index pairs files, use pairtools sort instead.
• mean and other field aggregations that depend on how pixels are grouped. To get a mean, store a sum and a count, which combine exactly.
• cooler.sandbox, cooler zoomify --legacy, and the deprecated append argument of cooler.create.create.

Dependencies

• cooler no longer depends on cytoolz, multiprocess or dill. Parallel commands use Python's multiprocessing, which made parallel balancing about 2.2x faster. Functions passed to a process pool must now be picklable, so lambdas no longer work.
• pyfaidx is optional and included in cooler[all].
• Tested on Python 3.9 to 3.14, NumPy 1.26 and 2.x, and pandas 1.5 to 3.x.

Full Changelog: v0.11.0...v0.12.0