Changelog

Release notes for xyzpy.

v1.3.4 (2026-04-30)

Enhancements:

• Expose parent_dir on cultivate() and cultivate() so the on-disk crop folder can be placed somewhere other than the current working directory.

Bug fixes:

• Fix parse_into_cases() raising IndexError (or silently returning wrong cases) when both combos and cases are supplied alongside a ds whose internal dimension order differs from the case-keys-then-combo-keys insertion order. Per-variable indexers are now built in the variable’s own dim order, and dims not present in ds are now treated as new coordinate locations rather than crashing.

v1.3.3 (2026-04-30)

Bug fixes:

• Fix cultivate silently ignoring cases.

v1.3.2 (2026-04-30)

Bug fixes:

• Fix xyzpy_grow not being shipped in the built wheel/sdist after the v1.3.1 entry-point relocation, which caused xyzpy-grow to fail with ModuleNotFoundError: No module named 'xyzpy_grow' on the conda-forge feedstock CI. The hatchling build now uses force-include to ship the top-level xyzpy_grow.py alongside the xyzpy package.

v1.3.1 (2026-04-29)

Bug fixes:

• Fix xyzpy-grow --num-threads N having no effect on numpy / BLAS / OpenMP threading. The CLI entry point lived inside the xyzpy package, so importing it ran xyzpy/__init__.py (which eagerly imports xarray / numpy) before main() could set OMP_NUM_THREADS etc. — the env vars were assigned too late. The entry point is now a top-level xyzpy_grow module, so the threading env vars land before any numerical library is imported. The recursive subprocess invocation in grow_subprocess() was updated to match.

v1.3.0 (2026-03-30)

New features:

• Add cultivate() for handling the entire crop lifecycle (annotate, sow, grow, reap) in one function.

• Add grow_subprocess() for running batches in isolated subprocesses with resource control — supports gpus (GPU device pooling via CUDA_VISIBLE_DEVICES), affinities (CPU pinning via taskset), log (save stdout/stderr per batch), num_workers, num_threads, raise_errors, and custom batch_ids (GH 20)

• Add xyzpy-grow CLI entry point for driving grow_subprocess() from the command line

• Add RayExecutor and RayGPUExecutor for Ray-based parallel execution, also usable via xyzpy-grow --ray

• Add infiniplot() — a new unified plotting interface accessible via ds.xyz.plot() that auto-detects plot type (line, scatter, heatmap) from data dimensions - see Plotting for details

• Add scatter plot support (data_var vs data_var) to xyz.plot

• Add cmoke() (OKLCH-based) and cimluv() (HSLuv-based) perceptually uniform single-hue colormap generators

• Add MemoryMonitor context manager for peak memory tracking, plus get_peak_memory_usage(), report_memory(), and report_memory_gpu() utilities

• Add visualize_tensor() for visualizing arbitrarily high dimensional tensors via 2D projections

• Add format_number_with_error() for nicely formatting numbers with known errors

Enhancements:

• Add parse_into_cases(), find_missing_cases(), and is_case_missing() for case-aware dataset filtering, with vectorized internals for large speedups on big parameter spaces

• Add load_batch(), load_result(), save_result() methods for direct batch data access

• Add delete_all() to cleanly remove a crop directory and reset object state

• Add missing_only option to harvest_combos()

• Allow functions to return plain dict, including for mixed cases+combos

• Add background_color, label, xticks, yticks, xticklabels, yticklabels options to xyz.plot

• Allow hlines/vlines to be strings referencing data_vars so spans can vary by row and column

• Legend and colorbar improvements in xyz.plot

• Export neutral_style() and get_neutral_style() for matplotlib styling

• Make various SLURM header options optional in grow_cluster()

• Warn instead of erroring on non-picklable constants

• Catch confusing bug when combos involve duplicated values

• Move build to pyproject.toml with hatchling + hatch-vcs

Bug fixes:

• Fix saving of complex datasets with engine='netcdf' (GH 15)

• Fix reap() with incomplete crops when the final batch has fewer items than batchsize

• Fix for recent joblib (cachedir → location kwarg)

• Fix Ellipsis handling bug in crop preparation

v1.2.1 (12th August 2021)

Bug fixes:

• fix a few bugs related to crops and batchsizes

v1.2.0 (12th August 2021)

Enhancements

• unified interface for mixing both cases and combos

• add random shuffling of growing order to load balance on batch systems etc

• add expand_dims() and drop_sel()

v1.1.0 (25th July 2021)

Enhancements

• Defer Crop.reap clean up until after dataset sync (useful if you forget to set overwrite=True)

• Capture and print Crop.grow_cluster output

• add visualize_matrix tool

• add cimple colormap generator

• allow @xyz.label decorator to specify harvester=

• spruce docs

Bug fixes:

• Fix sowing, reaping and merging multiple sets of cases (GH 13)

• Fix incomplete crop reaping when there is a non-zero batch remainder size

• Fix for futures that raise attribute errors themselves

v1.0.0 (24th October 2020)

Breaking changes

• Remove all the data processing functionality which can now pretty much all be found in xarray. This also removes the numba, scipy and cytoolz dependencies completely.

Enhancements

• Generalize (and deprecate) xyzpy.Crop.qsub_grow() to xyzpy.Crop.grow_cluster() (PR 10)

• Add SLURM support to xyzpy.Crop.grow_cluster() (PR 10)

• Add PBS support to xyzpy.Crop.grow_cluster()

• Fix PBS crop submission for job arrays of size 1

• Add xyzpy.save_merge_ds() for manually aggregating datasets to disk

• Add allow_incomplete=True option to xyzpy.Crop.reap() for gathering data even if the crop is not fully grown (GH 7)

• Make new Crop instances by default automatically load information from disk if they have been already prepared/sown (GH 7)

• Automatically load Crops in the current (or specified) directory with xyzpy.load_crops().

• Add 'joblib' and 'zarr' as possible engines for saving and loading datasets

• Add utility xyzpy.getsizeof() to quite accurately get a python objects size

• Keep a running track of covariance using RunningCovariance.

v0.3.1 (25th January 2019)

Bug fixes:

• Make sure license is included in sdist/wheel distributions (PR 6)

v0.3.0 (21st January 2019)

Breaking changes

• Changed plot option markersize -> marker_size to match other keywords.

Enhancements

• New Sampler object - sparsely sample combos into a pandas.DataFrame

• Decorate functions directly into Runner instances using label()

v0.2.5 (3rd December 2018)

Breaking changes

• (GH 5) combo_runner key argument pool renamed to executor

Enhancements

• (GH 5) Support multiprocessing.pool in combo_runner

• Document timing and estimation utilities

• Use loky as the default parallel executor

• plotting: add xjitter and yjitter

Bug fixes:

• make sure Crop._batch_remainder synced with disk.

• update pytest marking parametrizations to xfail for recent pytest

• compatibility updates for dask and numba

• fix farming example which wasn’t appearing

v0.2.4 (1st November 2018)

Bug fixes:

• Various compatibility fixes for plotting functionality

v0.2.3 (4th October 2018)

Enhancements:

• add Timer

• add benchmark()

• add Benchmarker

• add RunningStatistics

• add estimate_from_repeats()

Bug fixes:

• various fixes to batch growing and Crop

• various fixes to plotting

v0.2.2 (7th June 2018)

Enhancements:

• allow case_runner to return Dataset

Bug fixes:

• (GH 1) make numba an optional dependency

v0.2.1 (27th May 2018)

Bug fixes:

• docs updates

• distribute crop batches more evenly