0.75.0
 - feat: allow to set the number of bootstrapping iterations for
   differential features in `Rlme4` (`bs_iter`, default 1000) (#389)
 - fix: `import_plugin_feature_script` removed the first entry of
   `sys.path` (instead of the plugin directory) and returned the wrong
   plugin if a module with the same name was imported before (#310)
 - fix: replacing a plugin feature did not invalidate the cached
   feature data (`AncillaryFeature.hash` now includes `identifier`) (#311)
 - fix: the event count was set to the number of trace channels if
   "trace" was the first feature (`RTDCWriter.rectify_metadata`,
   `RTDC_Dict`, and `RTDCBase` without "experiment:event count") (#312)
 - fix: a corrupt file in the dataset directory with the same name as
   a file basin aborted the basin search (exception during relative
   path check) (#313)
 - fix: relative file basin paths with Windows path separators were
   not found on Unix (and vice versa); the separator check compared
   against a two-character string (#314)
 - fix: `apply_filter` raised a KeyError if a polygon filter was
   removed from a dataset and its `PolygonFilter` instance deleted (#315)
 - fix: an event manually re-included in a hierarchy child was
   excluded again when the hierarchy parent changed (if all manually
   excluded events had been re-included) (#316)
 - fix: removing the min/max keys of a box filter from the
   configuration did not remove that filter (#317)
 - fix: `Configuration.copy` did not preserve `disable_checks`, which
   removed unknown configuration keys from the copy (#318)
 - fix: exports of non-hierarchy datasets did not store the relative
   path to the original file in the basin definition (#320)
 - fix: CLI join failed if two features that are adjacent in the sorted
   feature list were missing in one of the files (feature list was
   modified while iterating over it) (#322)
 - fix: copying files with internal basins that share a basinmap
   feature failed (affects CLI repack, compress, and condense) (#324)
 - fix: copying files with zero events failed (affects CLI repack,
   compress, and condense); empty datasets are now copied as empty,
   compressed datasets instead of being dropped (#325)
 - fix: `is_properly_compressed` raised an IndexError for Zstandard
   compressed data without compression level (affects CLI repack,
   compress, condense, and `check_dataset`) (#326)
 - fix: `RTDCWriter` computed a wrong "mean" attribute for scalar
   features when appending data containing nan values (and the mean
   remained nan if the first data written were all nan) (#327)
 - fix: HDF5 files with user-defined metadata keys containing a colon
   (e.g. "user:time:start") or root attributes without a section
   could not be opened (#328)
 - fix: CLI compress with `--crop-event-images` and `--basin-input`
   failed if the input file and a basin both contained image data (#329)
 - fix: `get_volume` with `fix_orientation=True` returned wrong (also
   negative) volumes for contours in the opposite orientation, because
   the reordered contour was combined with the original x-coordinates (#330)
 - fix: `get_bright_perc` failed if `bg_off` was a numpy array (e.g.
   "bright_perc_10" for dictionary-based datasets) and for empty masks
   (now nan, like "bright_avg") (#331)
 - fix: "flX_max_ctc" of datasets with two fluorescence channels used
   the crosstalk configuration of the missing third channel, which
   gave wrong results and, since these keys were not part of the
   cache hash, stale data when they were changed (#332)
 - fix: `hashfile` with `blocksize=0` returned the hash of an empty
   file instead of the entire file and raised a TypeError when
   arguments were passed positionally (`file_monitoring_lru_cache`) (#334)
 - fix: polygon filter files (.poly) could not be loaded if the filter
   name contained "=" or if the file contained blank lines (#335)
 - fix: `downsample_rand` and `downsample_grid` reset the global random
   state of numpy to a fixed seed (downsampling results are unchanged) (#336)
 - fix: the `StoreKeeper` caching thread died with "dictionary changed
   size during iteration" when cached data were added concurrently,
   after which nothing was written to the disk store anymore and the
   memory store grew without limit (`MemoryStore` is now thread-safe
   and errors in `StoreKeeper.perform_tasks` are logged) (#337)
 - fix: cache keys of `umbrella_cache` were ambiguous, e.g. arguments
   `(1, 23)` and `(12, 3)`, `"1"` and `1`, or arrays with the same data
   but a different shape or dtype resulted in the same key (existing
   disk store cache entries are no longer used) (#338)
 - fix: functions decorated with `umbrella_cache` raised a KeyError if
   the `StoreKeeper` thread removed the cached value from the memory
   store between checking for and retrieving the value (#339)
 - fix: functions decorated with `umbrella_cache` raised a KeyError
   (instead of recomputing the value) if a disk store entry written by
   another process could not be read (#340)
 - fix: values returned by functions decorated with `umbrella_cache`
   were the cached objects themselves; modifying them in-place (e.g.
   KDE or downsampling results) modified the cached values (#341)
 - fix: image data of .tdms files differed slightly between platforms
   (e.g. x86 and ARM), because ffmpeg used optimized, inexact YUV to RGB
   conversion; dclab now requests a bitexact conversion (gray values of
   .tdms images change by up to about one gray level) (#342)
 - fix: accessing `Basin.ds` from multiple threads at the same time
   loaded the basin dataset multiple times and left all but one of the
   files open after `Basin.close` (#343)
 - fix: indexing chopped image data (internal chopped image basins)
   with a boolean array (e.g. `ds["image"][ds.filter.all]`) failed (#344)
 - fix: `RTDC_HDF5.hash` raised a TypeError for datasets opened from a
   file-like object (e.g. `io.BytesIO`), which also broke KDE methods (#345)
 - fix: the progress monitor thread of CLI compress and repack kept
   running if an error occurred (#346)
 - fix: `RTDCWriter.write_text` silently truncated appended log lines
   that were longer than the existing lines (at least 100 bytes); this
   could cut multi-byte UTF-8 characters, so the log could not be read (#347)
 - fix: the check for Shape-In issue #3 (wrong medium "CellCarrierB")
   did not work for files written by dclab, because dclab appends its
   version to "setup:software version" (#348)
 - fix: the integrity checks `check_fl_max_positive` and
   `check_fl_max_ctc_positive` failed for empty datasets and their
   result depended on the position of nan-values (#349)
 - fix: dangling HDF5 external links crashed `check_dataset` (instead of
   being reported by the external links check) (#350)
 - fix: `cli.verify_dataset` and `cli.tdms2rtdc` failed if paths were
   given as `str` instead of `pathlib.Path` (#351)
 - fix: hierarchy children inherited the "limit events" filter of the
   parent, i.e. the events were limited twice and changing the parent's
   limit did not take effect in the child (#352)
 - fix: scalar feature data of hierarchy children were writable, so
   modifying e.g. `child["deform"][:]` in-place modified the cached
   feature data of the child (the data are now read-only, like the
   data of other dataset formats) (#354)
 - fix: `export.hdf5` removed "contour" from the `features` list passed
   by the caller and failed for tuples (#355)
 - fix: `get_contour` raised an IndexError for an empty mask instead of
   `NoValidContourFoundError` (#356)
 - fix: `new_dataset` ignored `identifier` for `io.BytesIO` objects and
   additional keyword arguments for HTTP URLs (#357)
 - fix: configuration text files: values containing "#" were cut (only
   lines starting with "#" are comments now) and a key before the first
   section raised an UnboundLocalError (#358)
 - fix: the deprecated objects in `dclab.definitions` (e.g.
   `feature_names`) failed for method calls and `len()` (#359)
 - fix: `feature_deregister` removed the wrong label if several features
   had the same label (#360)
 - fix: the linear scaling functions for area_um, volume, and emodulus
   (`scale_feature`) failed for integer, scalar, or list input (#361)
 - fix: `KernelDensityEstimator.get_at` and `get_contour_lines` failed
   for ndarray `positions` and `quantiles` (#362)
 - fix: `KernelDensityEstimator.get_contour_lines` ignored `ret_levels`
   when no contour could be computed (#363)
 - fix: `find_smooth_contour_spacing` reported success for contours with
   undefined (nan) opening angles (#364)
 - fix: on a log scale, contour opening angles were computed from
   linearly normalized coordinates (resulting in nan values for points
   outside the plot range); the normalization is now done in log space (#364)
 - fix: on POSIX systems, the R subprocess calls of the lme4 module
   replaced the entire environment (e.g. no PATH) instead of extending
   it (#365)
 - fix: lme4 convergence messages were never reported, multiple lme4
   messages caused an R error, integers were formatted as floats in the
   R script, and the temporary R script was not removed (#366)
 - fix: three crash paths in the .tdms reader (image seek workaround,
   trace data with an incomplete last event, `hash` without the optional
   "_camera.ini" file) (#367)
 - fix: `HTTPFile.read()` without size returned no data (#368)
 - fix: values cached in the disk store changed their type when loaded
   (tuples became lists, non-string dictionary keys became strings);
   such values are now stored in the new "json-typed" format (#369)
 - fix: `BasinProxyFeature` did not use its cache for scalar features,
   i.e. each access after the first one copied the data event by event (#370)
 - fix: `Inverted = False` in a .poly file did not override the
   `inverted` keyword argument of `PolygonFilter` (as documented) (#371)
 - fix: unreadable disk store entries were never replaced (the value was
   recomputed on every call and the disk store size management failed) (#373)
 - fix: the identifier of plugin features created from a dictionary
   did not depend on the constants and names used in the method
   (e.g. `lambda d: d * 2` and `lambda d: d * 3` were identical) (#375)
 - fix: `get_emodulus` with `copy=False` raised a ValueError for
   lists or integer arrays with NumPy 2 (same for `convert` and
   `scale_feature` with `inplace=True`) (#376)
 - fix: a dangling external link in the "basins" group of an HDF5
   file raised a KeyError when the basins were loaded; such basin
   definitions are now ignored (`BrokenBasinDefinitionWarning`) (#377)
 - fix: `statistics.mode` returned the edge of the bin instead of its
   center (the data were rounded instead of floored before adding
   half the bin size) (#378)
 - fix: passing positional arguments (e.g. `h5kwargs`) to `RTDC_HTTP`
   or `RTDC_S3` raised a TypeError (ruff B026) (#385)
 - fix: importing the deprecated `dclab.kde_contours` raised an
   ImportError (`_find_quantile_level` was moved to the tests) (#398)
 - fix: lines with an empty key in configuration text files raised a
   TypeError (`keyval_str2typ` returned None); they are now ignored (#399)
 - fix: `emodulus.convert` raised an UnboundLocalError if `emodulus` was
   given without flow rates and viscosities (now a ValueError) (#400)
 - fix: loading a .poly file without "X Axis" or "Y Axis" raised an
   UnboundLocalError (now a `PolygonFilterError`) (#401)
 - fix: the type hints of `get_contour` and `get_contour_lazily` used
   `np.int` (removed in NumPy 1.24), so `typing.get_type_hints` failed (#402)
 - fix: `find_smooth_contour_spacing` only checked whether the first
   contour of the highest percentile was smooth, not all of them (#404)
 - fix: misleading error messages in `new_dataset` for remote data:
   S3-specific keyword arguments were passed to `RTDC_HTTP` for publicly
   available S3 URLs, and without `requests`, HTTP URLs resulted in
   "Unknown remote format" (#405)
 - fix: CLI `dclab-split` stored internal basin features (e.g.
   "image_bg") as regular features in the output files instead of
   internal basins (#410)
 - fix: identical internal basin definitions in a file and in one of
   its file basins were reported as a cyclic basin dependency (#410)
 - fix: wrong environment variable name (`S3_ENDPOINT_URL` instead of
   `DCLAB_S3_ENDPOINT_URL`) in S3 error messages (#411)
 - fix: wrong output file name in `LimitingExportSizeWarning` message (#412)
 - enh: include `h5datasetchop` in `basin_priority_sorted_key`
 - fix: wrong environment variable name in `RNotFoundError` message
   (`RHOME` instead of `R_HOME`) and wrong viscosity unit in emodulus
   messages and metadata description (Pa*s instead of mPa*s) (#413)
 - fix: the deprecated `hasher_class` argument of `util.hashfile` issued
   a UserWarning instead of a DeprecationWarning (#417)
 - fix: `Isoelastics.get` and `Isoelastics.get_with_rtdcbase` returned
   zero-valued emodulus for `viscosity=0` or `viscosity=False`; a
   ValueError is now raised for viscosities that are not positive (#418)
 - fix: `find_smooth_contour_spacing` reported success ("target opening
   angle reached") when the abort event was set while iterating over
   the datasets; also pass a flat list of quantiles to
   `get_contour_lines` (nested list only worked by chance) (#419)
 - fix: rename misspelled `PolygonFilter.instace_exists` to
   `PolygonFilter.instance_exists` and
   `KernelDensityEstimationForEmtpyArrayWarning` to
   `KernelDensityEstimationForEmptyArrayWarning` (the old names are
   deprecated aliases) (#420)
 - fix: `RTDC_Dict` did not allow "file" basins, because the class
   attribute `_local_basins_allowed` was shadowed by the instance
   attribute set in `RTDCBase.__init__` (#421)
 - fix: `Export.fcs` and `Export.tsv` added the "dclab version" key to
   the `meta_data` dictionary passed by the caller (#422)
 - fix: deprecated `export.hdf5_autocomplete_config` raised an
   AttributeError when given a path (`.close()` was called on the path
   instead of on the file opened by `RTDCWriter`) (#423)
 - fix: `find_smooth_contour_spacing` returned half of the last
   evaluated contour spacing when the maximum number of iterations
   was reached (#424)
 - enh: raise TypeError instead of ValueError for arguments of the wrong
   type in `emodulus.load_lut`, `PlugInFeature`, and
   `load_plugin_feature` (#403)
 - enh: the KeyError for ancillary features (incl. plugin features)
   that cannot be computed names the requirements that are not met
   (new `AncillaryFeature.get_missing_requirements`) (#406)
 - docs: inverted polygon filters include events with nan-values (#372)
 - docs: explain fluorescence crosstalk correction (#333)
 - docs: fix and complete docstrings of the file formats (HDF5, DCOR,
   HTTP, S3, .tdms), basins, ancillary and plugin features, and caching (#411)
 - docs: document tables (`RTDCBase.tables`) (#408)
 - docs: document converting video files to .rtdc with
   `scripts/vid2dc.py` (#409)
 - docs: fix and complete docstrings of the RT-DC dataset core modules
   (`RTDCBase`, `Configuration`, `Export`, `Filter`, `RTDCWriter`,
   copier, temporary features, hierarchy) (#412)
 - docs: fix and complete docstrings of features, kde, lme4, cli,
   definitions, and utility modules (#413)
 - docs: fix typos, wrong statements, broken references, and failing
   code snippets in the user guide and examples (#414)
 - docs: add missing classes and functions to the code reference (basins,
   integrity checks, tables, hierarchy mapping, binning, and others) (#416)
 - docs: fix all broken cross-references and build the docs in nitpicky
   mode in CI (#416)
 - docs: the default branch was renamed from "master" to "main" (#321)
 - docs: fix links to renamed files in the emodulus docs (#321)
 - ref: cleanup filter.py (incorrect variable names, use of sets)
 - ref: use `super()` without arguments (ruff UP008) (#379)
 - ref: use f-strings instead of `str.format` (ruff UP032) (#380)
 - ref: remove Python 2 remnants and outdated syntax (ruff pyupgrade) (#383)
 - ref: use explicit `X | None` for optional arguments in type hints (#384)
 - ref: remove unused `features_iter` argument from `basin_definition_copy`
 - ref: fix type annotations reported by pyright (`RTDCWriter.h5file`,
   `DiskStore.path`, optional imports, wrong annotations) (#426)
 - ref: annotate shared class attributes with `ClassVar` and minor
   cleanups (ruff B004, B006, B008, PLW0120, RUF012) (#386)
 - ref: simplify code (ruff SIM, C4, PIE, PERF, FURB, FLY, PLC0206) (#387)
 - ref: add pyright stub for h5py until h5py ships type hints (#388)
 - ref: use implicit string concatenation (ruff ISC003, ISC004) (#391)
 - ref: fix remaining minor ruff findings (#393)
 - ref: use double quotes for strings (ruff Q000) (#394)
 - tests: update pinned glmer values for lme4 2.1 (#381)
 - tests: use 100 bootstrap iterations in `test_glmer_differential`
   (took about 10 minutes on Linux and longer on Windows) (#389)
 - scripts: `vid2dc.py` uses the same exact YUV to RGB conversion as
   the .tdms reader (same images on all platforms) and does not
   require PyAV anymore (#409)
 - ci: the lme4 tests were silently skipped, because lme4 could not
   be installed with R 4.1.2; use the latest R release and fail if
   lme4 is not installed (#381)
 - ci: lint with ruff (rules configured in `pyproject.toml`) (#397)
0.74.0
 - fix: CLI commands compress, condense, join, and repack deleted the
   input file if it was also given as the output path
 - fix: deadlock when filtering datasets without "experiment:event count"
   metadata (regression in 0.71.9)
 - fix: `RTDC_Hierarchy.apply_filter` rebuilt the event data without
   holding `RTDCBase.lock`, so other threads could see an empty child
 - fix: `HTTPFile` returned and cached HTTP error responses (e.g. 503
   or expired pre-signed URLs) as file data; transient errors are
   now retried and other errors raise `requests.HTTPError`
 - fix: `HTTPFile.read_range` downloaded an additional chunk (beyond
   the end of the file) when reading up to a chunk boundary
 - fix: basins with a mismatching basin identifier were still used if
   the measurement identifier of the referrer matched
 - fix: traces could not be accessed or exported via mapped basins
   (introduce `BasinProxyTrace`)
 - fix: `BasinProxyFeature.min/max/mean` returned the values of the
   entire basin for mapped basins whose basinmap did not contain 0,
   and did not ignore NaN values otherwise
 - fix: changing "emodulus viscosity" (medium "other") did not
   invalidate the cached "emodulus" feature if "emodulus temperature"
   was also set
 - fix: "qpi:scale to filter" metadata could not be loaded from .rtdc
   files (`fboolorfloat` did not accept numpy scalars)
 - fix: `find_smooth_contour_spacing` raised an IndexError for datasets
   without a contour (e.g. all events filtered)
 - fix: `KernelDensityEstimator.get_contour_lines` with automatic
   spacing failed for data containing NaN values and ignored the
   filters when determining the plot range
 - fix: `downsample_grid` raised an IndexError if all values along one
   axis were identical (e.g. scatter plot of a constant feature)
 - fix: `isoelastics.get_available_identifiers` loaded the isoelastics
   files relative to the current working directory
 - enh: never show 100% rounded progress when CLI is still working
0.73.7
 - fix: broken frame feature lead to invalid index in `write_chopped_images`
 - fix: prevent endless loop for invalid index when reading chopped images
 - fix: integer overflow and quadratic runtime when repairing an invalid
   frame feature in `write_chopped_images`
 - enh: introduce `NegativeFrameCountWarning` for `write_chopped_images`
 - tests: invalid frame feature and malformed index for chopped images
0.73.6
 - fix: memory leak in `write_chopped_images`
0.73.5
 - enh: reduce time for `write_chopped_images` via sequential reading
0.73.4
 - ref: remove unimportant print statement
0.73.3
 - fix: "setup:version info" not inherited in `cli-compress` with internal
   basins and `--crop-event-images`
0.73.2
 - fix: `dclab-compress` metadata of basin overrides input with "same" mapping
0.73.1
 - fix: integer overflow in shape determination for chopped image compression
0.73.0
 - feat: CLI `dclab-compress` can now include data from local basin files
 - fix: migrate CLI to ByteBookKeeper progress for accurate numbers
 - fix: `H5Events.iter_chunks` failed for input datasets without chunks
 - fix: `copier.h5ds_copy` did not necessarily write in correct chunk sizes
 - enh: pass-through `iter_chunks` for `BasinProxyFeature`
 - enh: `write_chopped_images` can use required features from destination file
 - tests: fix version check
 - ref: more general type hints for chop_images
 - ref: agent-based type hints and cleanup
0.72.0
 - feat: support reading and writing internal chopped image basin data
 - feat: add `H5Events.iter_chunks` for memory-friendly mask access
 - feat: add `--crop-event-images` for `dclab-compress` CLI
 - fix: correctly check for `basinmap?` dtype int64
 - enh: new feature list for images that have the ROI shape `FEATURES_IMAGE_ROI`
 - enh: remove any stale temporary files generated by CLI using atexit
 - ref: cleanup of several submodules with ruff
 - ref: implement `__array_ufunc__` for `NDArrayOperatorsMixin` subclasses
 - ref: create submodules for members of `feat_basin`
 - ref: remove `cli.common.setup_task_paths` to improve code readability
 - ref: remove deprecated `--force` kwarg for `dclab-compress` CLI
 - setup: drop support for python 3.9
 - setup: remove importlib_resources dependency
0.71.10
 - fix: `bin_width_percentile` could fail for mostly constant arrays
 - enh: remember DCOR API token by default for subsequent basin access
 - enh: don't catch DeprectaionWarning when doing dataset checks
 - enh: fallback-search basin files in referring dataset directory
 - tests: increase robustness for slow CI systems
 - setup: scipy!=1.18.0, because of regression with contiguous arrays
0.71.9
 - enh: make filtering operations thread-safe with `RTDCBase.lock`
0.71.8
 - enh: store relative basin information when exporting data (Python>=3.12)
 - tests: introduce `StoreKeeper.event_run` to avoid race conditions
 - tests: consider floating point errors in `umbrella_cache` timing test
0.71.7
 - fix: ValueError when creating hierarchy child of a zero-length dataset
0.71.6
 - fix: IndexError when attempting to compute KDE for equal-valued feature
0.71.5
 - enh: allow internal basins with `RTDCBase(.., enable_basins=False)`
0.71.4
 - fix: dataset `identifier` is session-unique for `RTDC_HTTP`
 - fix: dataset `hash` must be globally unique for `RTDC_HTTP` and `RTDC_S3`
 - fix: strip quotes from ETag in `S3File`
 - enh: internal basin features now count as `features_innate`
 - enh: add unified `logger` property for all instances of `RTDCBase`
0.71.3
 - enh: disable HDF5 file locking for basic reading operations
0.71.2
 - enh: implement min/max/mean metadata ufuncs in `BasinProxyFeature`
 - enh: pass on min/max/mean metadata to unfiltered hierarchy `ChildScalar`
 - enh: add bytes format to `DiskStore`
 - enh: add logging to `umbrella_cache`
 - enh: migrate `HTTPFile` caching to `umbrella_cache` decorator
0.71.1
 - fix: handle deleted `DiskStore` in `umbrella_cache`
 - enh: store `DiskStore` keys in POSIX format internally
 - enh: only log when entries are added to persistent cache in `StoreKeeper`
 - enh: add `evaluation_time_threshold` kwarg to `umbrella_cache`
0.71.0
 - enh: add spacing kwargs `xacc` and `yacc` to `KernelDensityEstimator.get_at`
 - enh: refuse to compute contour lines for identical features (#273)
 - enh: refuse to compute contour lines for monotonous-only features
 - enh: add `definitions.feat_const.FEATURES_MONOTONOUS`
 - ref: remove deprecated `KernelDensityEstimator.get_contour`
 - ref: change kwargs order in `KernelDensityEstimator` methods
0.70.0
 - feat: migrate `find_smooth_contour_spacing` from DCscope (#153)
 - enh: use `find_smooth_contour_spacing` as default spacing
   in `KernelDensityEstimator.get_contour_lines`
 - fix: array to scalar conversion in `kde.contours`
 - ref: change default contour accuracy algorithm to `bin_width_percentile`
 - ref: move KDE helper methods/decorators to `kde.helpers` submodule
 - ref: move KDE binning methods to `kde.binning` submodule
 - ref: move `kde.contours._find_quantile_level` to test module
 - ref: rename `KernelDensityEstimator.get_spacing` to `estimate_spacing`
0.69.2
 - fix: comparing two `PolygonFilter` with differing size failed
 - enh: make `PolygonFilter` operations thread-safe
0.69.1
 - fix: use `for_json` instead of `__getstate__` for custom hashing
 - tests: use multiprocessing 'spawn' for SWMR test
0.69.0
 - feat: new `cached` module with the `umbrella_cache` decorator
   which combines in-memory with optional persistent disk caching (#153)
 - fix: refresh to get matching feature sizes in SWMR mode
 - enh: KDE computation uses `umbrella_cache` in higher-level
   `KernelDensityEstimator` with custom handler for `RTDCBase`,
   instead of caching lower-level functions in `kde.methods`
 - ref: `RTDC_Dict` computes its `hash` from full dict, not only the keys
0.68.0
 - feat: print progress in CLI dclab-compress and dclab-repack
 - enh: support opening .rtdc files that are in SWMR mode
 - ref: drop support for HDF5 < 1.14.4
0.67.5
 - fix: detect invalid table values and convert to NaN for DCOR data (#287)
0.67.4
 - fix: `RTDCBase.reset_filter` did not remove box filters
0.67.3
 - fix: exporting filtered hierarchy dataset from basin-mapped datasets broken
 - fix: downsampling failed when invalid values dominated
 - ref: introduce `StoringPerishableBasinWarning`
 - tests: increase coverage for hdf5 export
0.67.2
 - fix: `KernelDensityEstimator.get_raster` returned non-identical shapes for
   `X`, `Y`, `Z` when input data size is zero
 - fix: handle all-nan/inf input arrays for KDE computation
 - ref: using `return` in `finally` block is discouraged (PEP 765)
0.67.1
 - fix: downsampling with sample size larger than dataset size failed
 - docs: type-hinted the features sub-module
 - enh: add mass density feature, deprecate some qpi features (#284)
 - enh: warn users about basin identifier mismatches
0.67.0
 - BREAKING CHANGE: make it difficult to export the "contour" feature
 - fix: `BasinProxyFeature` returned incorrect `shape` and `size`
 - enh: implement `size` for `H5MaskEvent` and `H5ScalarEvent`
 - enh: implement `BasinProxyContour`
 - enh: improve hashing of `H5ContourEvent`
 - docs: acknowledge Shape-Out is being continued as DCscope
 - tests: correct DCOR IP address
0.66.0
 - feat: introduce optional identifiers for basins
 - ref: rename `measurement_identifier` to `referrer_identifier`
   in the `feat_basin` context to clarify which dataset it belongs to
0.65.0
 - scripts: add `vid2dc.py` for converting videos to raw .rtdc files
 - enh: run identifier splits use "_" upon export and have increased length
0.64.3
 - ci: use GitHub windows-latest runner for building wheel
0.64.2
 - fix: string comparison TypeError when verifying basins, in niche cases
   when basin's run identifier was not specified or could not be computed
 - enh: raise NotImplementedError when subsetting basin contour data
 - enh: make measurement identifier computation robust against empty strings
 - ref: remove zombie code for fetching ML model label from model instance
 - ref: minor cleanup in feature definitions
 - docs: minor cleanup
0.64.1
 - enh: introduce `KernelDensityEstimator.get_at`
 - enh: add 10th, 25th, 75th, and 90th percentile statistics methods
 - enh: new `ret_dict` arg in `get_statistics` returns dict instead of tuple
 - setup: fix finding packages in pyproject.toml
0.64.0
 - feat: introduce concept of perishable (e.g. temporally expiring) basins
 - feat: add `PerishableRecord` to presigned URLs for DCOR basins (#244)
 - fix: do not export perishable basins (#254)
 - fix: allow `Basin.verify_basin` to fail
 - fix: `Basin.__repr__` failed when location did not exist
 - fix: fetching tables information from DCOR dataset could time out
 - fix: default to compression level 5 when exporting or writing (#272)
 - enh: increase timeout every time fetching data from DCOR API times out
 - docs: update docs `RTDCWriter` I/O
 - setup: drop support for Python 3.8
 - tests: add tests for fetching tables from a DCOR dataset (#229)
 - ref: add `get_contour_lines` method to `KernelDensityEstimator` class
 - ref: deprecate get_contour and use bin_width_doane_div5 in kde submodule
 - ref: cache "basins" in `RTDC_DCOR` instead of in `RTDC_DCOR.api`
0.63.1
 - docs: update documentation and code references for kde submodule
 - enh: cache S3 sessions and clients (test suite duration reduced by 30%)
 - enh: cache "basins" response for DCOR format
0.63.0
 - feat: add `progress_callback` argument to all data export methods (#271)
0.62.17
 - docs: properly employ license GPL-2.0-or-later
 - ref: add KernelDensityEstimator class
 - ref: perform avi export with PyAV instead of imageio
 - setup: bump fcswrite to 0.5.1
0.62.16
 - enh: allow editing S3 credentials via environment variables at runtime
 - docs: fix typo in S3 section (missing `DCLAB_S3_SECRET_ACCESS_KEY`)
 - docs: add more example use cases in the basin section
 - docs: add example script for rewriting paths of basins
 - docs: remove broken reference to HDF5 `linker` module
 - tests: reduce test time by 35% from 400s to 250 (smaller S3 reference data)
0.62.15
 - fix: support opening files on non-mounted network shares on Windows
0.62.14
 - fix: exporting table data broken
 - ref: introduce base class for tables `TableMeta`
0.62.13
 - ref: add custom table class for `hdf5` and `dcor` formats
0.62.12
 - fix: viscosity computation for MP-PBS at zero flow rate should be NaN
0.62.11
 - enh: support plain `numpy.array` tables
0.62.10
 - fix: `basin_definition_copy` failed when more than one basins defined
   due to redundant loop
 - fix: JSON logs during export were not sorted and indented
0.62.9
 - enh: sort basin definition before exporting to JSON; the side-effect
   is that the unique key for a basin changes, possibly resulting in
   duplicated basin definitions for existing datasets
 - enh: make sure every basin can be identified with a key created from
   its definition; important for connecting the basin dictionaries
   with the instantiated basins
0.62.8
 - fix: metadata of tables not copied by `copier.py`
 - docs: formatting typo (#270)
0.62.7
 - fix: IntegrityChecker must not load basins
 - fix: KeyError in IntegrityChecker when "basin_events" group missing (#268)
 - fix: warn user about missing internal basin features while still loading
   those that are available (#268)
 - fix: do not copy internal basin definitions for non-existent features (#269)
 - docs: clarify interpretation of [qpi]: "filter size" metadata (#267)
 - tests: avoid architecture and Python-version dependent test
0.62.6
 - enh: support passing pathlib.Path to lme4.rsetup.set_*_path methods
 - docs: correct keyword arguments for advanced section on S3
0.62.5
 - enh: make set_r_lib_path available in lme4 submodule
0.62.4
 - enh: honor R_LIBS_USER environment variable when running R
0.62.3
 - fix: get_r_version string conversion not necessary
0.62.2
 - fix: use full path to R executable when running lme4 analysis in rsetup
 - enh: properly catch CalledProcessError in run_command
 - enh: improve search for R executable path
 - ref: replace sp.check_output with run_command in lme4 submodule
0.62.1
 - fix: numpy-version-specific default for `copy` in `__array__()`
0.62.0
 - BREAKING: removed the `rtdc_dataset.linker` module, because it is not
   used by any other piece of software and because combining file-like
   objects external links are quirky and their combination finally broke
   with h5py 3.12
 - enh: added integrity check for external links
 - tests: reduce warnings
0.61.5
 - fix: integrity check failed for empty datasets
0.61.4
 - docs: update basins section
 - docs: cleanup build process
0.61.3
 - fix: TypeError when of R_HOME is not set
0.61.2
 - ref: migrate from rpy2 to custom wrapper around R and Rscript binary
 - setup: remove extra [lme4] and rpy2 requirement
0.61.1
 - fix: properly catch warnings for determining bad event image for fmt_tdms
 - fix: make sure integer features are always written as uint,
   and for that introduce `dclab.rtdc_dataset.writer.FEATURES_UINT32`
   and `dclab.rtdc_dataset.writer.FEATURES_UINT64`
 - setup: support numpy 2
0.61.0
 - BREAKING: removed the `feat_anc_ml` module and with it all functionalities
   related to loading machine learning features from .modc files (#236)
 - fix: do not attempt to load S3 data when the endpoint seems wrong
 - enh: improve availability check for S3 basin data
 - enh: improve availability check for HTTP basin data
 - enh: improve availability check for HDF5 basin data
 - enh: cache list of features available via all basins
 - enh: speed up feature retrieval for non-ancillary features
 - ref: replace `np.string_` with `np.bytes_`
0.60.9
 - build: build with macos-latest and macos-13 images
0.60.8
 - build: build with macos-12-large and macos-13-xlarge images
0.60.7
 - build: build with macos-13-large and macos-13-xlarge images
0.60.6
 - build: build with macos-latest-large and macos-latest-xlarge images
0.60.5
 - fix: return correct length for BasinProxyFeature
 - fix: export from dataset with internal basins did not work (#262)
0.60.4
 - fix: check feature list for export for duplicates (#257)
 - enh: more verbose error message when loading data
 - docs: a few clarifications on Young's modulus computation
0.60.3
 - fix: add args and kwargs to all __array__ methods
0.60.2
 - enh: add [experiment]:timestamp (unix time) to configuration
0.60.1
 - fix: subclass `BasinProxyFeature` from `NDArrayOperatorsMixin`
0.60.0
 - feat: support (mapped) internal basins
 - enh: add log entry to the output file when exporting data (#251)
0.59.1
 - fix: protection against cyclic basin dependencies
 - fix: ValueError when exporting data from basin-mapped dataset
 - fix: Check for basin availability when checking run identifier
0.59.0
 - feat: support basins with blown indices
 - enh: increase verbosity when failing to resolve basins
 - enh: implement dtype for `H5ScalarEvent`
0.58.8
 - enh: existence of "index_online" section is no longer required
0.58.7
 - enh: introduce `RTDCBase.features_local` for accessing features
   that are exclusively local and not in remote locations
 - enh: prevent `RTDCBase._get_ancillary_feature_data` from
   unnecessarily accessing feature data for hashing
 - enh: introduce new feature "bg_off", a float-valued feature for
   event-wise background offset correction
 - enh: warn user about missing endpoint URLs in S3 format and raise an
   exception when push comes to shove
 - enh: add "pipeline:dcnum mapping" metadata
 - docs: document S3 environment variables
0.58.6
 - enh: for access to private S3 data, introduce the environment
   variables `DCLAB_S3_ENDPOINT_URL`, `DCLAB_S3_ACCESS_KEY_ID`,
   and `DCLAB_S3_SECRET_ACCESS_KEY`
 - ref: reorganize the `fmt_s3` module
0.58.5
 - fix: CLI methods returned exit status 1 since paths were returned
0.58.4
 - fix: support relative paths in basins with both nt `\\` and
   posix `/` path separators
 - enh: store original metadata in .tsv export (#255)
0.58.3
 - fix: keep only 1000 contours in LazyContourList, because keeping
   all contours causes the memory to fill up for large input files
0.58.2
 - fix: KeyError when iteration over DCOR logs or tables
 - fix: remove hack that populates feature names for basins unreliably
 - enh: lazily load logs and tables in `fmt_hierarchy`
0.58.1
 - fix: exporting with basins from a hierarchy child did not write a
   basin derived from the hierarchy root parent
 - docs: add example for `map_indices_child2parent` in `fmt_hierarchy`
 - ref: explicitly define ancillary features in `RTDC_Hierarchy`
0.58.0
 - feat: implement "mapped basins" that have a superset of events
 - feat: allow to export basin-only HDF5 files
 - fix: support input files without "events" in `rtdc_copy`
 - fix: CLI methods now store basin information in output paths (#239):
   compress, condense, and repack preserve the original input basins;
   split writes mapped basins (using export.hdf5); join does not write
   basins at all
 - fix: enable nested basins for the HDF5, S3, and HTTP basins,
   because we have the `_local_basins_allowed` property since 0.57.6
 - fix: when joining datasets, the name of the log holding the configuration
   of the source datasets should be src-#{i}_cfg instead of cfg_src-#{i}
 - enh: introduce `RTDCBase.features_ancillary`, a list of all ancillary
   features (computed on-the-fly) for a dataset
 - enh: add `include_basins` keyword argument for `rtdc_copy`
 - enh: `RTDCWriter.store_basin` now returns the basin hash/identifier
 - enh: CLI methods now return the output path
 - enh: the repack CLI method now allows to strip basins
 - docs: add an advanced usage section on basins
 - docs: update doc strings for CLI methods
 - setup: pin nptdms<1.9, because tests started to fail
 - ref: modify basin `load_dataset` methods to support mapped basins
 - ref: put input and output path in order for CLI methods
 - tests: make sure basin features are not written automatically in
   CLI methods (#246)
 - ci: install lme4 from archive
0.57.7
 - fix: raise a ValueError in HTTPFile when the server does not return
   the status code 200
 - enh: properly handle missing ETag in HTTPFile
 - enh: compute fallback identifier using hash of first chunk in RTDC_HTTP
 - ref: drop the old dcserve API version 1 and use version 2 by default
0.57.6
 - enh: allow non-existent paths in `common.get_command_log`
 - ref: factor out `cli.task_condense.condense_dataset` for condensing
   an open dataset without the knowledge of its path
 - ref: when condensing a dataset, instead of the MD5-5M, the hash of its
   configuration is used as an suffix for previous condense logs
0.57.5
 - fix: do not allow local basins for network-based subclasses of `RTDCBase`
0.57.4
 - fix: catch zero-valued contour accuracy for KDE plot (#249)
0.57.3
 - setup: the s3 extra does not require a specific version of urllib3
 - ci: use main branch of GH Actions actions to reduce maintenance burden
0.57.2
 - fix: speed-up S3 availability check (#245)
 - enh: allow to specify list of features for data copier
 - ref: migrate from s3fs to boto3 for fmt_s3
 - test: data copier with `none` and invalid features
 - docs: remove pin for sphinx
 - build: make musllinux builds for e.g. docker images
0.57.1
 - fix: RTDCWriter.rectify_metadata fails when image feature is empty
 - fix: handle empty write requests in export.hdf5 and RTDCWriter (#242)
 - tests: add test for writing all-nan data (#242)
 - docs: improve documentation of hierarchy child mapper
0.57.0
 - fix: integer overflow in downsample_grid
 - fix: removed unnecessary computation of hierarchy filter instance
 - enh: cythonize downsample_grid (~10x speed-up)
 - enh: got rid of for-loop in map_indices_parent2child (~100x speed-up)
 - enh: slight improvement of managing manual indices in hierarchy children
 - enh: added dtype properties for contour and trace events
 - enh: ensure all feature data objects have the dtype property
 - enh: globally use chunk sizes of ~1MiB when writing HDF5 data (#217)
   (minor speedup since previously a chunk size of 100 events was used
   for images and scalar features were written in one big chunk)
 - enh: favor array operations in `yield_filtered_array_stacks` (~4x speed-up)
 - enh: increase verbosity for unreachable remote basin features
 - ref: implement `__array__` methods for hierarchy event classes
 - ref: implement `__array__` method for `H5MaskEvent`
 - ref: new submodule for hierarchy format
0.56.3
 - fix: regression missing check for basin availability
0.56.2
 - enh: perform basin availability checks in daemon thread
0.56.1
 - enh: priority-based basin sorting (file over remote, http over dcor)
0.56.0
 - feat: allow nested basins
 - feat: implement DCOR basins
 - fix: make sure basins are always closed on context exit (#238)
 - fix: failed to load basin data when port was specified in location
 - enh: requests session pooling for fmt_http and fmt_dcor
 - enh: implement context manager for RTDCBase
 - enh: be forwards-compatible and ignore unsupported basin formats
 - ref: new http_utils submodule for managing HTTP connections
0.55.7
 - test: tested rtdc_dataset.config.Configuration is pickable (#190)
 - enh: add dcnum pipeline identifier metadata constants
0.55.6
 - fix: default `host` was overriding URL host in fmt_dcor
0.55.5
 - enh: register new features area_um_raw, deform_raw, eccentr_prnc,
   per_ratio, per_um_raw, sym_x, sym_y
 - ref: simplify/cleanup computation of tilt feature
0.55.4
 - enh: improve timeout and availability check for fmt_http
0.55.3
 - enh: migrate to purely requests-based HTTP file for fmt_http
0.55.2
 - fix: fmt_dcor did not properly retry failed connections
 - enh: minor optimizations for fmt_http and fmt_s3
0.55.1
 - fix: disable instance cache fsspec.HTTPFileSystem (fmt_http)
0.55.0
 - feat: implement HDF5-based HTTP file format and basin
 - fix: replace lru_cache on instance methods to fix memory leak (#214)
 - fix: remove lru_cache in DCOR format (#226)
 - fix: implement `__contains__` for DCOR logs and tables
 - enh: improve parsing of DCOR URLs (scheme in host)
 - enh: speed-up of initialization of DCOR data
 - enh: add requests timeout for DCOR data
 - enh: more caching of event size and shape for HDF5 format
 - enh: faster computation of contour length for DCOR format
 - enh: use dcserv version 2 in DCOR format (fast S3 access)
 - enh: change measurement identifier upon filtered export
 - setup: pin s3fs>=2023.10.0
 - setup: pin upper bounds of dependencies
0.54.2
 - fix: availability checks for basins
0.54.1
 - fix: properly check for basin availability before getting features
0.54.0
 - fix: partial workaround for #238
 - fix: hash alias for HDF5 Datasets in files not in the file system
 - enh: delayed computation of measurement identifiers for basins
 - enh: allow to specify which features are served by a basin
 - enh: improve speed of S3 file format by increasing chunk size
 - enh: do not immediately raise an exception when metadata are missing
 - enh: lazy initialization of .RTDCBase.filters
 - enh: lazy initialization of .RTDCBase.basins
 - enh: lazy loading of feature list for HDF5 data
 - setup: migrate from pkg_resources to importlib_resources
 - ref: remove the dclab.isoelastics.ISOFILES constant
 - ref: remove the dclab.features.emodulus.INTERNAL_LUTS constant
0.53.3
 - fix: bad hdf5 basin check
0.53.2
 - fix: catch OSError when accessing invalid hdf5 basin
 - enh: implement retrieving basins in fmt_dcor
0.53.1
 - fix: catch OSError when accessing invalid S3 URL
 - enh: implement `Configuration.as_dict`
 - enh: implement `Basin.as_dict`
0.53.0
 - feat: introduce remote type basin format s3
 - feat: implement `store_basin` in `RTDCWriter`
 - docs: add information about basins
 - ref: expand `fmt_dcor` submodule laterally
 - ref: deprecate the `feat_anc_ml` submodule
 - enh: improve URL parsing for s3 format
 - enh: extract dataset size from metadata before resorting to features
 - enh: support remote basins
0.52.6
 - maintenance release
0.52.5
 - maintenance release
0.52.4
 - ref: HDF5Basin gets its own submodule
 - setup: oldest-supported-numpy is not required anymore, since numpy
   is backwards-compatible now
0.52.3
 - maintenance release
0.52.2
 - maintenance release
0.52.1
 - docs: document the copier submodule (#218)
 - docs: simplify headings in advanced section
 - docs: refactor writing DC data into its own subsection
 - docs: add advanced section on how to work with S3 data (#228)
 - docs: add information about S3 object storage in DCOR section
 - setup: force requests >= 2.31.0 (CVE-2023-32681)
 - setup: bump scipy to 1.10.0 (CVE-2023-25399)
 - tests: add test tables and logs tests for linker
 - tests: add tests for S3 file format (#228)
0.52.0
 - feat: qpi features and related metadata added to definitions (#192)
0.51.4
 - setup: wrong package setup
0.51.3
 - setup: wrong package setup
0.51.2
 - enh: support dictionary of arrays in RTDCWriter.store_table
 - tests: added missing test files
0.51.1
 - fix: workaround for segmentation fault caused by HDF5
0.51.0
 - feat: introduce read-support for local basins (upstream features)
 - fix: minor bug lead to issue with latest rpy2 versions (invisible)
 - fix: support numpy integers and booleans for JSON serializer
 - enh: introduce optional configuration key [experiment]:"run identifier"
 - docs: restructuring and introduction of new features
 - ref: restructured fmt_hdf5 submodule
 - ref: remove deprecated `write` method for creating .rtdc files
 - ref: introduce more general `RTDCBase._finalize_init`
 - tests: tests passing on Python 3.11
0.50.4
 - setup: cleanup pyproject.toml
0.50.3
 - fix: do not allow editing reserved filter properties (#210)
 - fix: do not allow editing temporary features (#202)
 - ref: read-only features and proper subclassing in `RTDC_Dict`
 - ref: subclass from `collections.UserDict` and not from `dict`
 - docs: add more information about manual filters (#173)
0.50.2
 - enh: support reading 'tables' from DCOR datasets (#221)
0.50.1
 - enh: support reading 'logs' from DCOR datasets
 - enh: hierarchy parents inherit their parent's logs and tables
0.50.0
 - feat: support opening objects on S3-compatible storage (#213)
 - feat: support opening file-like objects as HDF5 data
 - enh: use regexp to identify valid DCOR URLs
 - enh: allow to pass keyword arguments to h5py.File for fmt_hdf5
0.49.1
 - fix: do not recompute inert_ratio_cvx and inert_ratio_raw
   computed with Shape-In >= 2.0.5 (#224)
 - tests: add additional test for inertia ratio (#223)
0.49.0
 - feat: implement `is_properly_compressed`, `rtdc_copy`, and
   `h5ds_copy` in `dclab.rtdc_dataset.copier` (#216)
 - feat: support "tables" (compound arrays) in HDF5 files
 - fix: explicitly use 64bit ints and floats when instantiating numpy
   arrays, because numpy on Windows defaults to 32bit ints
   which resulted in overflows when computing inertia ratio
 - enh: unify names of exported logs
 - enh: CLI use `rtdc_copy` in dclab-compress, dclab-condense,
   and dclab-repack (#216)
 - enh: CLI compute the command log as early as possible
 - enh: ignore "def" feature when performing an unknown-feature check
 - enh: minor improvements in error message verbosity here and there
 - ref: remove deprecated method dclab.load.check_dataset
0.48.8
 - enh: CLI new --no-ancillary-features keyword argument for
   dclab-condense to skip the computation of expensive features
0.48.7
 - maintenance release
0.48.6
 - maintenance release
0.48.5
 - fix: CLI dclab-verify-datasets now complains about missing input data
 - enh: feed kwargs to sp.check_output to avoid pop-up command prompts
 - ref: migrate from "tensorflow-SavedModel" to "tf" in ml submodule
 - build: migrate to pyproject.toml (except for cython extensions)
 - build: add manylinux wheels using cibuildwheel
0.48.4
 - fix: do not cache all feature names, only defective feature names,
   because some features can be (de)registered dynamically
0.48.3
 - fix: always treat ancillary time as float64
 - enh: add defective feature check for inertia ratio features
   inert_ratio_cvx, inert_ratio_prnc, inert_ratio_raw, and tilt (#212)
 - enh: cache available feature names in the HDF5 format
0.48.2
 - fix: invalid computation of inertia ratio features (inert_ratio_cvx,
   inert_ratio_prnc, inert_ratio_raw, tilt) for events with large
   horizontal extent due to an integer overflow (#212)
 - enh: many-fold increase in HDF5 dataset read speed via caching (#189)
 - enh: detect all-zero temperature in dclab-verify-dataset (#183)
0.48.1
 - fix: emodulus values were not correctly cached
0.48.0
 - BREAKING CHANGE: remove deprecated "emodulus model" key from
   dataset configuration
 - feat: new 'buyukurganci-2022' model to compute MC-PBS viscosity (#197)
 - enh: use the 'buyukurganci-2022' model for isoelastics by default
 - enh: increase verbosity when opening an .rtdc file failed
 - enh: support shorthand for known media for viscosity computation
 - docs: update part about DCOR (#182)
0.47.8
 - fix: implement cleaner solution for time computation (#207)
0.47.7
 - fix: include time offset when computing defective time feature (#207)
 - ref: make the h5py.File object public in RTDC_HDF5
0.47.6
 - fix: register float32 time from ShapeIn as defective feature
 - docs: make "edit on GitHub" link lead to actual source file again
0.47.5
 - enh: allow temporary features for hierarchy children (#123)
 - enh: more type annotations (partially #198)
 - build: add Python 3.11 CI pipeline
0.47.4
 - enh: allow pathlib.Path objects in cli.get_command_log custom dicts
0.47.3
 - fix: np.histrogram2d does not expect broken normed argument
 - docs: fixed GH Actions badge
 - enh: implement `RTDC_HDF5.__len__`
0.47.2
 - enh: add --version flag to CLI
0.47.1
 - fix: matplotlib.plot hierarchy child scalar data not working
0.47.0
 - feat: lazy-loading and improved caching strategy for hierarchy
   child datasets via `RTDC_Hierarchy` (the call to `apply_filter`
   is now mandatory for all changes to propagate to the children)
 - feat: introduce `RTDC_Hierarchy.rejuvenate`, an alias for
   `apply_filter` which more accurately describes what happens there
 - feat: added file_monitoring_lru_cache convenience decorator
 - enh: new feature bg_med: Median frame background brightness
 - enh: add min/max/mean values as scalar dataset attributes (#175)
0.46.6
 - fix: tdms format config parser not working when para.ini missing
0.46.5
 - enh: new staticmethod RTDC_TDMS.extract_tdms_config for
   extracting metadata from tdms-based datasets; This method
   optionally returns a list of paths used for extracting metadata
 - ref: cleanup in RTDC_TDMS with dict.setdefault
0.46.4
 - enh: allow to export logs alongside HDF5 data (#180)
 - enh: add features flow_rate and pressure
0.46.3
 - docs: make compilation run on Windows
 - docs: update information on plugin features (#151)
 - ref: remove Zellmechanik Dresden branding
0.46.2
 - fix: support passing instances of `Configuration` and
   `ConfigurationDict` to `dclab.util.obj2bytes`
 - enh: return sorted json representation in `Configuration.tojson`
 - ref: use `functools.update_wrapper` in `Cache` wrapper
 - docs: make sure `dclab.downsampling.downsample_grid` is in code ref
0.46.1
 - fix: ConfigurationDict should not subclass from dict (#174)
 - ref: use f-strings and print correct feature for temporary feature
   error message (#193)
0.46.0
 - BREAKING CHANGE: The LUT data for LE-2D-FEM-19 had to be revised,
   because the simulation support was sparse at high deformations (#191)
 - feat: added new LUT and isoelastics for HE-2D-FEM-22
   and HE-3D-FEM-22 (#188)
 - docs: new LUT data version 10.6084/m9.figshare.12155064.v4
0.45.0
 - feat: introduce 10th and 90th percentile brightness features
 - enh: added definitions for the Haralick texture features
   (computed using the mahotas package from bg-corrected event mask)
0.44.0
 - feat: introduce background-corrected brightness features
 - enh: detect confusion in plugin feature names (#179)
 - enh: cache innate feature data for hierarchy children
 - enh: raise more verbose error message when the user tries to access
   non-existent features in hierarchy children (#92)
0.43.1
 - fix: script fem2rtdc.py did not work with latest openCV (#176)
 - fix: properly implement random pixel offsets in fem2rtdc.py and
   and pixelation_correction.py (#178)
 - enh: improved test for compression (require Zstd with level >= 5)
 - enh: added `close` method for `RTDCWriter`
 - ref: added tests and cleaned up fem2rtdc.py
0.43.0
 - feat: introduce Zstandard compression via hdf5plugin
 - enh: speed-up HDF5 data export if `filltarr` are all-True
 - ref: deprecate "compression" keyword argument for HDF5 export
0.42.3
 - enh: support hierarhcy feature dtype property for image and mask event
 - docs: fix regression in builds due to new H5ScalarEvent
0.42.2
 - fix: lazy-load scalar feature data for HDF5 data
 - fix: fix repr string of AncillaryFeature when identifier is None
 - ref: compute identifiers from HDF5 data using file system path
   and HDF5 path
0.42.1
 - fix: support "image_bg" feature for DCOR data
 - ref: replace formats with f-strings in check.py
0.42.0
 - enh: reduce computation time of file hashes in the cli submodule
   (and thus all other software downstream) by switching from a full
   SHA256 hash to a 5MB md5 hash; this speeds up operations on
   slow network shares
 - ref: minor change when installing R packages
   (rlme tests are broken on Windows currently)
 - ref: use np.float32 for principal inertia ratio computation
   (compatibility with OpenCV)
0.41.0
 - feat: pull DCOR access token management from downstream DCOR-Aid
 - feat: allow to set alternate DCOR server certificate bundles by
   appending paths `dclab.rtdc_dataset.fmt_dcor.DCOR_CERTS_SEARCH_PATHS`
 - enh: allow to skip checks and by default use the innate features
   during export of an RTDCBase
 - setup: remove appveyor build pipeline
 - ref: cleanup fmt_dcor
0.40.0
 - setup: bump numpy from 1.17.0 to 1.21.0
 - setup: bump scipy from 0.14.0 to 1.8.0
 - setup: drop support for Python 3.7
0.39.18
 - fix: regression since 0.39.0; contours with wrong index where not
   repaired
 - ref: Python distutils library is deprecated
 - ref: some scipy namespaces are deprecated
0.39.17
 - enh: allow to specify unique_id of PolygonFilter when loading
   from file
0.39.16
 - fix: not all statistics exported if no feature was specified
0.39.15
 - fix: fix accessing inherited non-scalar temporary or plugin features
   for RTDC_Hierarchy class (#165)
 - fix: properly handle user-defined (temporary or plugin) features
   when exporting to HDF5 (#166)
 - enh: add "shape" keyword argument in RTDCWrite.store_feature to
   allow the user to pass the correct data shape; this reduces warning
   messages when using the RTDCBase.export.hdf5 functionality
 - ref: simplify data export to HDF5 with generator function
 - tests: use correct tdms-dataset for tdms tests
0.39.14
 - fix: enable loading of non-scalar plugin data as h5py.Dataset (#162)
 - build: minor fixes in CI and testing pipeline
0.39.13
 - fix: recompute the "ml_class" feature if the ml_score features
   change (either the underlying model or the number of features)
0.39.12
 - fix: support nan values for the computation of the "ml_class"
   feature (#161)
 - fix: cache of util.hashfile used access times of files (and thus
   might not have been of any use in some cases)
0.39.11
 - fix: misleading error message when loading plugin features
   with Python modules that are not available (#160)
 - docs: add information on how to call CLI functions from within
   Python (#145)
0.39.10
 - fix: make absolutely sure HDF5 files are closed if RTDCWriter is
   used as a context manager
 - docs: properly document the shapes option of plugin features (#146)
0.39.9
 - fix: spelling of Young's modulus (#159)
 - ref: define minimal R version 3.6.0 in lme4 submodule and
   propagate any errors with possible solutions to the user
 - ref: extract exact version string for R
0.39.8
 - enh: cache computed file hashes with additional os.stat info
 - tests: fix tests on Windows
0.39.7
 - fix: handle empty trace feature in RTDCWriter.rectify_metadata
0.39.6
 - fix: do not close H5File object passed to RTDCWriter.__init__
0.39.5
 - fix: write FeatureSetNotIdenticalJoinWarnings to logs in dclab-join
0.39.4
 - fix: KeyError object 'index_online' doesn't exist when joining
   two datasets with dclab-join (#158)
 - fix: generalize online_filter configuration keys (#156)
 - fix: relax validation of online_filter configuration key types
 - fix: make sure that only common features of the input files for
   dclab-join are written to the output file (#157)
 - enh: add `with_unit` keyword to `get_feature_label`
 - ref: slightly modified online_filter configuration descriptions
0.39.3
 - ref: deprecate these lists/dicts in dclab.definitions:
   config_descr, config_funcs, config_types, feature_names,
   feature_labels, feature_name2label (#135)
 - ref: use more intuitive trace feature in fmt_dcor
0.39.2
 - ref: renamed mask and contour labels
 - ref: increase verbosity in error messages
 - tests: add .modc test model and data
0.39.1
 - fix: don't create __pycache__ folders for plugin features
 - fix: add "date" to machine-learning feature info dict
0.39.0
 - enh: implement .shape and .__len__ for non-scalar features
   and all file formats (#117)
 - ref: adjust ancillary machine-learning feature API to match
   that of plugin features (this is not a breaking change,
   because nobody is using machine-learning features yet)
 - ref: minor refactoring of tdms file format code might lead to
   faster loading of trace data
0.38.2
 - fix: reduce memory usage when writing "mask" feature data
0.38.1
 - maintenance release
0.38.0
 - enh: add identifier to AncillaryFeature class
 - enh: introduce MachineLearningFeature to simplify the use of
   machine learning models in ancillary features
 - fix: load_modc should return a list of models
 - ref: rename `rtdc_dataset.ancillary` submodule to `anc_feat_core`
 - ref: rename `rtdc_dataset.plugins` submodule to `anc_feat_plugin`
 - ref: move the `ml` submodule to `rtdc_dataset.anc_feat_ml`
 - ref: move tensorflow-related code into its own submodule
 - setup: bump minimum Python version to 3.7
0.37.3
 - fix: ignore defective feature "volume" in pre 0.37.0 .rtdc files
 - fix: always perform version branding when using RTDCWriter context
   manager
 - ref: silence numpy deprecation warning in RTDCWriter
0.37.2
 - enh: speed-up contour computation by a factor of three
   (volume is computed by factor of two faster)
0.37.1
 - enh: make IntegrityChecker class available on top level module
0.37.0
 - BREAKING CHANGE: The volume feature in all previous versions was
   overestimated by about 2µm³. Please re-run your analysis pipeline.
 - fix: volume was computed incorrectly (#141)
0.36.1
 - setup: bump h5py version to 3.0 (issue with integer chunk argument)
0.36.0
 - feat: introduce new RTDCWriter class (#144)
 - fix: in some rare cases, an empty .rtdc file could be generated
   when splitting an .rtdc file via the CLI dclab-split
 - fix: user-defined metadata were not copied with CLI repack
 - fix: support exporting non-scalar plugin features to HDF5 (#143)
 - enh: implement consistent `.shape` property for contour data
   (partially #117)
 - enh: improve export speed of contour data
 - enh: improve export speed if data filtering is disabled
 - ref: disable `time_offset` of internal `export.hdf5_append` function
 - ref: deprecate write_hdf5.write function
 - ref: deprecate export.hdf5_autocomplete_config function
 - ref: deprecate export.hdf5_append function
0.35.8
 - fix: write logs to fixed-length strings in HDF5 files (HDF5 never
   supported variable length string datasets with fletcher32 or compression
   filters)
 - fix: non-scalar features that are not defined in dclab were silently
   ignored upon export; temporarily raise a ValueError instead (#143)
 - setup: release pinning of h5py
0.35.7
 - fix: more Shape-In versions are affected by the medium-write bug
   addressed in 0.33.1
 - enh: improve error handling when loading rpy2
 - enh: add checks for negative or zero-valued metadata
 - tests: fix broken example dataset with zero-valued metadata (#137)
0.35.6
 - fix: config converters were not applied to raw metadata (#139)
0.35.5
 - enh: add sanity_check method to IntegrityChecker
0.35.4
 - enh: implement function for obtaining the description of a
   configuration key get_config_value_descr
0.35.3
 - enh: implement dumping of Configuration to json
0.35.2
 - enh: allow to compress, condense, and repack without checking for
   the input file suffix
 - docs: fixed example data (sphinx encountered unexpected warnings)
0.35.1
 - fix: ignore empty logs (raised IndexError before)
 - tests: increase coverage (#136)
0.35.0
 - enh: support new configuration keys for online polygon filtering
   in the form of "[online_filter]: area_um,deform polygon points" (#134)
 - enh: new helper functions in dclab.definitions: get_config_value_type,
   get_config_value_func, and config_key_exists
 - ref: issue a warning when the data type of a configuration key is
   not correct
0.34.6
 - fix: make get_job_info and get_command_log available in cli submodule
 - tests: add documentation for test data (#119)
0.34.5
 - fix: cast data to float when filling with nan values when checking the
   size of ancillary features
 - fix: dclab-split did not work when nptdms is not installed
 - tests: skip tests that require nptdms, requests, or rpy2 if those
 - enh: add check to warn users from setting None-type config values
   packaages are not installed
 - tests: add test for non-scalar plugin feature
 - ci: run tests with different setup.py-extras installed
0.34.4
 - fix: use temporary file names during CLI operations
0.34.3
 - fix: workaround for when rpy2 finds R_HOME, but R doesn't
0.34.2
 - fix: slicing of non-scalar features for hierarchy children (#128)
 - fix: passing a directory to the CLI tdms2rtdc did not work
 - enh: support slicing of non-scalar features for DCOR data (#132)
 - enh: support slicing of the contour feature for HDF5 data (#132)
 - enh: support slicing of LazyContour for contour AncillaryFeature (#132)
 - enh: minor optimizations of the DCOR format
 - ref: raise NotImplementedError for futile attempts to slice the image,
   contour, or mask features for .tdms data (#132)
 - ref: cleanup CLI code
 - tests: increase coverage of CLI (#116)
0.34.1
 - enh: introduce user-defined "user" configuration section (#125)
 - scripts: refactor fem2lutiso_std.py and fem2iso_volume.py so that
   model-specific post-processing is done in hooks named after the FEM
   identifier, e.g. fem_hooks/LE-2D-FEM-19.py (#90)
 - docs: new LUT data version 10.6084/m9.figshare.12155064.v3
 - ref: minor code cleanup
0.34.0
 - feat: introduce user-defined plugin features (#105)
 - fix: dclab-verify-dataset now also prints other errors it encounters
 - fix: installing the "lmer" R package failed, because "statmod" was not
   installed
 - fix: correct data types for "fluorescence:sample rate",
   "imaging: roi position x", and "imaging: roi position y" (#124)
 - enh: support new .rtdc attribute "online_contour:bg empty" which is
   True when the online background image is computed only from frames
   that do not contain any events (#124)
 - enh: `AncillaryFeature` now populates other ancillary features when they
   share the same method (#104)
 - enh: dclab-verify-dataset now returns a non-zero exit code if there
   were errors, alerts, or violations (#120)
 - ref: streamlined dataset check function for missing meta data keys
0.33.3
 - fix: add "chip identifier" to "setup" configuration section and
   make it optional during dataset checks (#109)
 - fix: ignore empty metadata strings (partly #109); removed the
   `check_metadata_empty_string` check function because it does
   not apply anymore
0.33.2
 - fix: some datasets with unknown feature names could not be
   opened (AssertionError regression in 0.33.1)
 - fix: workaround for sporadic JSONDecodeError when accessing DCOR
 - ref: cleanup cli.py
 - ref: cleanup util.py and deprecate `hash_class` argument in
   `hashfile`
0.33.1
 - fix: add dataset check for wrong medium written to .rtdc file
   by Shape-In
 - fix: filters were ignored when exporting trace data to hdf5 (#112)
 - enh: allow to set ICue identifier for integrity checking
 - ref: code cleanup in export.py
0.33.0
 - feat: introduce user-defined temporary features (point 2 in #98)
 - fix: catch errors for integrity checks on non-raw datasets (#102)
 - fix: add check for negative fluorescence values (#101)
 - enh: add metadata keys for baseline offset (#107)
 - setup: remove deprecated setup.py test
0.32.5
 - fix: add check for zero-flow rate in dclab-verify-dataset
 - setup: added new file CREDITS for docs and only use maintainer
   in setup.py
 - docs: add autodoc to constant variables (#94)
 - docs: add "features_innate" and "features_loaded" to scripting
   goodies section
 - ref: cleanup of RTDCBase class
 - ref: int/bool deprecation warnings in numpy 1.20.0 (#93)
 - tests: test for area_cvx as a float, consistent with output from
   Shape-In (#96)
0.32.4
 - fix: TypeError when registering emodulus LUT (#91)
 - ref: minor cleanup
0.32.3
 - build: use oldest-supported-numpy in pyproject.toml
0.32.2
 - fix: export trace data in chunks to avoid out-of-memory
   errors when compressing large files
 - ref: introduce CHUNK_SIZE in write_hdf5.py and use it when
   exporting to .rtdc
0.32.1
 - enh: dclab-compress now by default does not compress any input
   files that are already fully compressed (fully compressed means
   that all HDF5 datasets are compressed somehow); to get the old
   behavior back (compress in any case, use the "force" keyword
   argument)
0.32.0
 - feat: allow to register external look-up tables for Young's
   modulus computation (#88)
 - ref: restructure look-up table file handling
 - ref: deprecated [calculation]: "emodulus model" metadata key in favor
   of the more descriptive "emodulus lut" key.
 - ref: the "method" argument in the context of isoelasticity lines
   is deprecated in favor of the "lut_identifier" argument
0.31.5
 - fix: writing "filtering" and "calculation" metadata sections
   to .rtdc files should not be allowed
0.31.4
 - ci: fix rtd builds
 - ci: fix PyPI releases
0.31.3
 - ci: migrate to GitHub Actions
0.31.2
 - enh: add soft type check (assertion) for "emodulus medium"
   key in ancillary features (#86)
 - fix: make sure that strings are not written as bytes in
   hdf5 files
0.31.1
 - enh: add boolean "model converged" key to return dictionary of
   `Rlme4.fit` (#85)
0.31.0
 - feat: implement (generalized) linear mixed-effects models
   via a wrapper around R/lme4 using rpy2 (install with extra
   "lme4")
0.30.1
 - fix: `new_dataset` attempts to load DCOR dataset when
   given a non-existent path as a string (#81)
0.30.0
 - BREAKING CHANGE: drop support for Python 2 (#34)
 - feat: new machine learning (ML) submodule `dclab.ml`
 - feat: implement ML model file format .modc (#78)
 - feat: add tensorflow helper functions for RT-DC data
 - setup: bump numpy>=1.17.0
 - ref: minor improvements of code readability
 - tests: set global temp dir and remove it after testing
0.29.1
 - enh: lift restrictions on valid options for [setup]:medium
   (can now be any arbitrary string)
0.29.0
 - feat: support the "image_bg" feature which contains the
   rolling mean background image computed by Shape-In
0.28.0
 - feat: new CLI command dclab-split to split a large dataset into
   multiple smaller datasets
0.27.11
 - fix: do not cache hierarchy child feature values; this might lead
   to intransparent situations where a child has different features
   than its parent (you cannot always rely on the user to call
   `apply_filter`)
 - fix: hierarchy child configuration section "calculation" was not
   updated with the hierarchy parent values
 - docs: add example for loading data from DCOR and computing
   the Young's modulus
0.27.10
 - fix: support unicode characters when writing HDF5 in Python2
0.27.9
 - docs: add artwork
 - fix: support unicode characters in sample names in Python2
0.27.8
 - docs: add more information on emodulus computation
 - docs: add script for visualizing emodulus LUTs
0.27.7
 - ref: replace deprecated .tostring() with .tobytes()
0.27.6
 - fix: video seek issue workaround also for the first 100 frames
 - cli: also skip the final event in tdms2rtdc if the image is empty
 - cli: renamed kwarg `--include-initial-empty-image`
   to `include-empty-boundary-images`
 - enh: improve detection and recovery of missing images for fmt_tdms
0.27.5
 - maintenance build
0.27.4
 - maintenance build
0.27.3
 - fix: ignore ResourceWarning due to unknown _io.BufferedReader
   in third-party software when converting .tdms to .rtdc
0.27.2
 - maintenance build
0.27.1
 - setup: bump imageio to 2.8.0 for Python>=3.4
 - ref: removed NoImageWarning during export (warning is already
   issued by fmt_tdms.event_image)
0.27.0
 - feat: introduce new feature names `ml_score_???` where `?`
   can be a digit or a lower-case letter of the alphabet (#77)
 - feat: introduce new functions `dclab.definitions.feature_exists`
   and `dclab.definitions.scalar_feature_exists` for checking the
   existence of features (including the `ml_score_???` features
   which are not in `dclab.definitions.feature_names`)
 - feat: introduce ancillary feature `ml_class` which is
   defined by the `ml_score_???` features
 - enh: fmt_dict automatically converts scalar features to arrays
 - ref: replace check for `dclab.definitions.feature_names`
   by `dclab.definitions.feature_exists` where applicable
 - ref: replace access of `dclab.definitions.feature_name2label`
   by `dclab.definitions.get_feature_label` where applicable
 - ref: do not automatically fill up all the box filtering ranges
   in `RTDCBase.config["filtering"]` with zeros; raise ValueError
   if user forgets to set both ranges
 - docs: major revision (promote Shape-Out 2 and DCOR)
0.26.2
 - fix: `kde_methods.bin_num_doane` now uses 5 as default if it
   ecounters nan or zero-division
 - docs: updates related to Young's modulus computation
0.26.1
 - enh: cache more online data in fmt_dcor
 - enh: add `dclab.warn.PipelineWarning` which is used as a parent
   class for warnings that a user might be interested in
 - fix: temperature warnings during emodulus computation revealed
   only the lower temperature limit of the data
0.26.0
 - feat: implement volume-deformation isoelasticity lines (#70)
 - fix: specifying an external LUT as ndarray did not work
 - scripts: finish 'fem2iso_volume.py' for extracting volume-
   deformation isoelasticity lines
 - scripts: add 'pixelation_correction.py' for visualizing
   pixelation effects on area_um, volume, and emodulus
 - ref: renamed isoelasticity line text files
0.25.0
 - fix: appending data to an hdf5 file results in a broken "index"
   feature (re-enumeration from 0), if the given dataset contains
   the "index_online" feature
 - enh: allow to set external LUT files or LUT data when computing
   the Young's modulus with the `lut_data` keyword argument in
   `dclab.features.emodulus.get_emodulus`.
 - ref: refactored `features.emodulus`: New submodules `pxcorr` and
   `scale_linear`; `convert` is deprecated in favor of `scale_feature`
0.24.8
 - setup: include Python 3.8 builds and remove Python<=3.5 builds
 - scripts: renamed 'extract_lut_and_iso.py' to 'fem2lutiso_std.py'
0.24.7
 - fix: `ConfigurationDict.update` did not take into account
   invalid keys (everything is now done with (__setitem__`)
0.24.6
 - maintenance release
0.24.5
 - maintenance release
0.24.4
 - maintenance release
0.24.3
 - fix: `ConfigurationDict.update` did not actually perform the
   requested update (does not affect `Configuration.update`)
 - enh: also use points_in_polygon from scikit-image to determine
   contour levels
0.24.2
 - build: import new skimage submodules so that PyInstaller
   will find and use them
0.24.1
 - enh: improve polygon filter speed by roughly two orders of
   magnitude with a cython version taken from scikit-image; there
   are only minor differences to the old implementation (top right
   point included vs. lower left point included), so this is not
   a breaking change (#23)
0.24.0
 - data: refurbished LUT for linear elastic spheres provided by
   Dominic Mokbel and Lucas Wittwer (based on the FEM simulation
   results from https://doi.org/10.6084/m9.figshare.12155064.v2);
   compared to the old LUT, there is a relative error in Young's
   modulus below 0.1 %, which should not cause any breaking changes
 - data: updated isoelasticity lines (better spacing): analytical
   data was made available by Christoph Herold, numerical data
   was interpolated from the new LUT
 - scripts: added 'scripts/extract_lut_and_iso.py' for extracting
   Young's modulus LUT and isoelastics from FEM simulation data
   provided by Lucas Wittwer; this is now the default method for
   extracting new LUTs and isoelastics
 - scripts: added 'scripts/fem2rtdc.py' for generating in-silico
   .rtdc datasets from FEM simulation data provided by Lucas Wittwer
 - fix: dclab-verify-dataset failed when the "logs" group was not
   present in HDF5 files
 - fix: use predefined chunks when writing HDF5 data to avoid
   exploding file sizes when writing one event at a time
 - fix: create a deep copy of the metadata dictionary when writing
   HDF5 data because it leaked to subsequent calls
 - ref: changed the way isoelasticity lines and emodulus LUTs
   are stored and loaded (e.g. json metadata and a few more sanity
   checks)
0.23.0
 - feat: enable emodulus extrapolation for `area_um`/`deform` values
   outside of the given LUT.
0.22.7
 - enh: dclab-verify-dataset now also checks whether the sheath and
   sample flow rates add up to the channel flow rate
 - ref: Configuration does not anymore load unknown meta data
   keyword arguments, but ignores them. This implies that
   dclab-verify-dataset will not anymore check them actively.
   Instead, any warning issued when opening a file is added
   to the list of cues.
 - setup: bump nptdms to 0.23.0
0.22.6
 - fix: data export to HDF5 did not work when the "frame rate"
   is not given in the configuration
0.22.5
 - enh: add checks for valid keys in the Configuration dictionary
   of a dataset `RTDCBase().config`; unknown keys will issue
   an UnknownConfigurationKeyWarning (#58)
 - ref: moved `rtdc_dataset.fmt_hdf5.UnknownKeyWarning` to
   `rtdc_dataset.config.UnknownConfigurationKeyWarning`
 - ref: renamed `rtdc_dataset.config.CaseInsensitiveDict`
   to `rtdc_dataset.config.ConfigurationDict` and added option
   to check new keys
0.22.4
 - fix: disable computation of Young's modulus for reservoir
   measurements (#75)
 - enh: new keyword argument `req_func` for `AncillaryFeature` to
   define additional logic for checking whether a feature is
   available for a given instance of `RTDCBase`.
0.22.3
 - enh: add `data` property to ICues (and use it when checking for
   compression)
0.22.2
 - fix: when computing the contour from the mask image, always use
   the longest contour - critical when the mask image contains
   artefacts
 - fix: minor issue with dclab-verify-dataset when nptdms was not
   installed and an exception occured
 - enh: dclab-verify-dataset shows some info on data compression
0.22.1
 - enh: remember working API Key
 - docs: document DCOR format
0.22.0
 - feat: implement DCOR client
 - enh: improved .rtdc file format detection (with wrong extension)
0.21.2
 - enh: dclab-verify-dataset now also checks HDF5 "mask" feature
   attributes
 - setup: bump h5py to 2.10.0 (need `<object>.attrs.get_id`)
0.21.1
 - fix: correct type of HDF5 image attributes for "mask" feature
0.21.0
 - feat: implement new CLI dclab-repack
 - fix: don't write "logs" group to HDF5 files if there aren't any
 - fix: support HDF5 files that have no "logs" group
 - docs: fix docstring of dclab-join
0.20.8
 - fix: regression where old .tmds data could not be opened if
   they did not contain the "area_msd" feature
 - fix: convert bytes logs to string in fmt_hdf5
 - enh: support len(ds.logs) for fmt_hdf5
 - enh: replace "info" by "build" in CLI job info
0.20.7
 - fix: ensure file extension is .rtdc in dclab-join
 - fix: correct "frame" and "index_online" features when exporting
   to hdf5
 - enh: allow to set metadata dictionary in dclab.cli.join
0.20.6
 - fix: typo in contour check resulted in small tolerance
0.20.5
 - fix: be more trustful when it comes to contour data in the tdms
   file format; instead of raising errors, issue warnings (#72)
0.20.4
 - ref: move integrity checks to new class check.IntegrityChecker
 - docs: document remaining dictionaries in dclab.dfn
0.20.3
 - docs: fix bad anchors
0.20.2
 - ref: using temperature values outside the range for viscosity
   computation now issues a warning instead of raising an error;
   warnings were added for the CellCarrier buffers
 - fix: handle number detection correctly in get_emodulus
0.20.1
 - fix: always return an array when computing the KDE
 - ref: make accessible static function RTDCBase.get_kde_spacing
0.20.0
 - feat: compute elastic modulus from "temp" feature (#51)
 - enh: computing isoelastics from datasets can use
   [setup]: "temperature" to compute the viscosity/emodulus (#51)
 - enh: define new meta data key [setup]: "temperature"
 - docs: add an advanced section on Young's modulus computation (#51)
0.19.1
 - fix: hierarchy children did not pass `force` argument to
   hierarchy parent when `apply_filter` is called
 - fix: revert histogram2d "density" argument to "normed" to support
   numpy 1.10 (Shape-Out 1)
 - fix: implement unambiguous `RTDCBase.__repr__`
0.19.0
 - feat: added better contour spacing computation based on percentiles
   (dclab.kde_methods.bin_width_percentile)
 - feat: add feature "index_online" which may be missing values (#71)
 - feat: implement __getstate__ and __setstate__ for polygon filters
 - fix: write UTF-8 BOM when exporting to .tsv
 - enh: add check whether unique_id exists in PolygonFilter
0.18.0
 - fix: correctly handle filtering when features are removed from a
   dataset
 - ref: move dclab.rtdc_dataset.util to dclab.util
 - ref: minor cleanup in computation of viscosity (support lower-case
   `medum` values, add `dclab.features.emodulus_viscosity.KNOWN_MEDIA`)
 - ref: cleanup dclab.rtdc_dataset.filter (use logical operators,
   correctly display nan-warning messages, keep track of polygon
   filters, add consistency checks, improve readability)
0.17.1
 - maintenance release
0.17.0
 - feat: add command line script for compressing HDF5 (.rtdc)
   data "dclab-compress"
 - enh: record warnings under "/log" for all command line scripts
 - enh: set gzip data compression for all command line scripts
0.16.1
 - fix: circumvent UnicodeDecodeErrors which occured in frozen macOS
   binaries that use dclab
 - enh: support subsecond accuracy in the the configuration key
   [experiment] time (e.g. "HH:MM:SS.SSS" instead of "HH:MM:SS")
 - enh: store the correct, relative measurement time in dclab-join (#63)
0.16.0
 - fix: RTDCBase.downsample_scatter with ret_mask=True did not return
   boolean array of len(RTDCBase) as indicated in the docs
 - ref: remove RTDCBase._plot_filter, which was confusing anyway
 - ref: deprecate usage of RTDCBase._filter
0.15.0
 - feat: add method RTDCBase.reset_filter
 - feat: implement RTDCBase.features_loaded
 - feat: allow to instantiate RTDC_Hierarchy without initially
   applying the filter
 - fix: non-scalar columns of RTDC_Hierarchy did not implement len()
 - docs: add an example script dedicated to data plotting
 - ref: remove circular references between Filter and RTDCBase
0.14.8
 - fix: Ignore feature "trace" when the trace folder exists but is
   empty (HDF5 format)
 - fix: If no contour can be found, raise an error before other
   ancillary features produce cryptic errors
0.14.7
 - enh: allow to add meta data when exporting to .fcs or .tsv
   (dclab version is saved by default)
 - setup: bump fcswrite from 0.4.1 to 0.5.0
0.14.6
 - fix: improved handling of tdms trace data (split trace with fixed
   samples per event to avoid ValueError when exporting to hdf5)
 - fix: transposed roi size x/y config value when exporting to hdf5
0.14.5
 - cli: write warning messages to logs in tdms2rtdc
 - ref: increase verbosity of warning messages
0.14.4
 - fix: discard trace data when "samples per event" has multiple values
   for tdms data
 - fix: prefer image shape over config keywords when determining the
   shape of the event mask and check the shape in dclab-verify-dataset
 - fix: avoid ContourIndexingError by also searching the neighboring
   (+2/-2) events when the contour frame number does not match (#67)
0.14.3
 - enh: explicitly check contour data when testing whether to include
   the first event in tdms2rtdc
0.14.2
 - ref: convert said ValueError to ContourIndexingError
0.14.1
 - fix: ValueError when verifying contour frame index due to comparison
   of float with int
0.14.0
 - feat: new command line script for creating a scalar-feature-only
   dataset with all available ancillary features "dclab-condense"
 - enh: enable scalar feature compression for hdf5 export
 - docs: fix doc string for dclab-tdms2rtdc
   (`--include-initial-empty-image` falsely shown as "enabled by default")
0.13.0
 - feat: allow to obtain a mask representing the filtered data with
   the `ret_mask` kwarg in `RTDCBase.get_downsampled_scatter`
 - feat: allow to force-exclude invalid (inf/nan) events when downsampling
   using the `remove_invalid` keyword argument
 - feat: exclude empty initial images in dclab-tdms2rtdc;
   they may optionally be included with "--include-initial-empty-image"
 - feat: new property `RTDCBase.features_innate` (measuerd feature)
 - enh: log which ancillary features were computed in dclab-tdms2rtdc (#65)
 - enh: improved tdms meta data import (also affects dclab-tdms2rtdc)
 - enh: update channel count and samples per event when writing hdf5 data
 - enh: dclab-verify-dataset now recognizes invalid tdms data
 - enh: several other improvements when reading tdms data
 - enh: group meta data in log files (dclab-tdms2rtdc and dclab-join)
 - fix: correctly handle hdf5 export when the image or contour columns
   have incorrect sizes (affects dclab-tdms2rtdc)
 - fix: ignore empty configuration values when loading tdms data
 - fix: image/contour files were searched recursively instead of
   only in the directory of the tdms file
 - fix: check for precence of "time" feature before using it to
   correct measurement date and time
 - fix: ancillary feature computation for brightness had wrong
   dependency coded (contour instead of mask)
 - fix: ancillary feature computation when contour data is involved lead to
   error, because `LazyContourList` did not implement `identifier` (see #61)
 - ref: remove NoContourDataWarning for tdms file format
 - tests: improve dataset checks (#64)
0.12.0
 - feat: add command line script for joining measurements "dclab-join" (#57)
 - feat: make log files available as `RTDCBase.logs`
 - feat: include log data in "dclab-join" and "dclab-tdms2rtdc"
 - fix: `features` property for tdms file format falsely contains
   the keys "contour", "image", "mask", and "trace" when they are
   actually not available.
 - enh: support loading TDMS data using the 'with' statement
 - docs: add example for joining measurements
 - docs: other minor improvements
 - setup: add Python 3.7 wheels for Windows (#62)
 - setup: remove Python 2 wheels for macOS
0.11.1
 - docs: add example for fluorescence trace visualization
 - docs: restructure advanced usage section
 - ref: make dclab in principle compatible with imageio>=2.5.0;
   Dependencies are pinned due to segfaults during testing
 - setup: make tdms format support and data export dependency optional;
   To get the previous behavior, use `pip install dclab[all]`
0.11.0
 - feat: compute contours lazily (#61)
0.10.5
 - setup: migrate to PEP 517 (pyproject.toml)
0.10.4
 - enh: ignore defective feature "aspect" from Shape-In 2.0.6 and 2.0.7
 - enh: support loading HDF5 data using the 'with' statement
   (e.g. `with dclab.new_dataset(rtdc_path) as ds:`)
0.10.3
 - fix: add numpy build dependency (setup_requires)
0.10.2
 - fix: HDF5-export did not re-enumerate "index" feature
0.10.1
 - fix: support nan-valued events when computing quantile levels
   in submodule `kde_contours`
0.10.0
 - BREAKING CHANGE: Change np.meshgrid indexing in
   `RTDCBase.get_kde_contour` from "xy" to "ij"
 - feat: new submodule `kde_contours` for computing kernel density
   contour lines at specific data percentiles (#60)
 - fix: range for contour KDE computation did not always contain end
   value (`RTDCBase.get_kde_contour`)
 - fix: `positions` keyword argument in `RTDCBase.get_kde_scatter`
   was not correctly scaled in the logarithmic case
 - ref: cleanup and document PolygonFilter.point_in_poly
 - ref: move skimage code to separate submodule "external"
 - ref: drop dependency on statsmodels and move relevant code
   to submodule "external"
0.9.1
 - fix: all-zero features were treated as non-existent due to relic
   from pre-0.3.3 era
 - fix: correct extraction of start time from tdms format
   (1h offset from local time and measurement duration offset)
 - fix: correct extraction of module composition from tdms format
   (replace "+" with ",")
 - enh: add configuration key mapping for tdms format to simplify
   conversion to hdf5 format (see ``fmt_tdms.naming``)
 - enh: do not add laser info for unused lasers for tdms format
 - enh: dclab-verify-dataset checks for image attribute dtype
 - enh: include original software version when exporting to rtdc format
0.9.0
 - feat: add new feature: gravitational force, temperature,
   and ambient temperature
 - ref: remove unused `has_key` function in
   `rtdc_dataset.config.CaseInsensitiveDict`
 - setup: require numpy>=1.10.0 because of `equal_nan` argument in `allclose`
0.8.0
 - fix: usage of "xor" (^) instead of "or" (|) in statistics
 - feat: support `remove_invalid=False` in downsampling.downsample_rand (#27)
 - feat: add keyword arguments `xscale` and `yscale` to improve data
   visualization in `RTDCBase.get_downsampled_scatter`,
   `RTDCBase.get_kde_contour`, and `RTDCBase.get_kde_scatter` (#55)
 - enh: make downsampling code more transparent
 - BREAKING CHANGE: low-level downsampling methods refactored
   - downsampling.downsample_grid: removed keyword argument
     `remove_invalid`, because setting it to `False` makes no sense
     in this context
   - downsampling.downsample_rand: changed default value of `remove_invalid`
     to `False`, because this is more objective
   - rename keyword argument `retidx` to `ret_idx`
   - these changes do not affect any other higher level functionalities
     in `dclab.rtdc_dataset` or in Shape-Out
0.7.0
 - feat: add new ancillary feature: principal inertia ratio (#46)
 - feat: add new ancillary feature: absolute tilt (#53)
 - feat: add computation of viscosity for water (#52)
0.6.3
 - fix: channel width not correctly identified for old tdms files
0.6.2
 - ci: automate release to PyPI with appveyor and travis-ci
0.6.0
 - fix: image export as .avi did not have option to use unfiltered data
 - fix: avoid a few unicode gotchas
 - feat: use Doane's formula for kernel density estimator defaults (#42)
 - docs: usage examples, advanced scripting, and code reference update (#49)
0.5.2
 - Migrate from os.path to pathlib (#50)
 - fmt_hdf5: Add run index to title
0.5.1
 - Setup: add dependencies for statsmodels
 - Tests: filter known warnings
 - fmt_hdf5: import unknown keys such that "dclab-verify-dataset"
   can complain about them
0.5.0
 - BREAKING CHANGES:
   - definitions.feature_names now contains non-scalar
     features (inlcuding "image", "contour", "mask", and "trace"). To
     test for scalar features, use definitions.scalar_feature_names.
   - features bright_* are computed from mask instead of from contour
 - Bugfixes:
   - write correct event count in exported hdf5 data files
   - improve implementation of video file handling in fmt_tdms
 - add new non-scalar feature "mask" (#48)
 - removed configuration key [online_contour]: "bin margin" (#47)
 - minor improvements for the tdms file format
0.4.0
 - Bugfix: CLI "dclab-tdms2rtdc" did not work for single tdms files (#45)
 - update configuration keys:
   - added new keys for [fluorescence]
   - added [setup]: "identifier"
   - removed [imaging]: "exposure time", "flash current"
   - removed [setup]: "temperature", "viscosity"
 - renamed feature "ncells" to "nevents"
0.3.3
 - ref: do not import missing features as zeros in fmt_tdms
 - CLI:
   - add tdms-to-rtdc converter "dclab-tdms2rtdc" (#36)
   - improve "dclab-verify-dataset" user experience
 - Bugfixes:
   - "limit events" filtering must be integer not boolean (#41)
   - Support opening tdms files with capitalized "userDef" column names
   - OSError when trying to open files from repository root
0.3.2
 - CLI: add rudimentary dataset checker "dclab-verify-dataset" (#37)
 - Add logic to compute parent/root/child event indices of RTDC_Hierarchy
   - Hierarchy children now support contour, image, and traces
   - Hierarchy children now support and remember manual filters (#22)
 - Update emodulus look-up table with larger values for deformation
 - Implement pixel size correction for emodulus computation
 - Allow to add pixelation error to isoelastics (`add_px_err=True`) (#28)
 - Bugfixes:
   - Pixel size not read from tdms-based measurements
   - Young's modulus computation wrong due to faulty FEM simulations (#39)
0.3.1
 - Remove all-zero dummy columns from dict format
 - Implement hdf5-based RT-DC data reader (#32)
 - Implement hdf5-based RT-DC data writer (#33)
 - Bugfixes:
   - Automatically fix inverted box filters
   - RTDC_TDMS trace data contained empty arrays when no trace
     data was present (trace key should not have been accessible)
   - Not possible to get isoelastics for circularity
0.3.0
 - New fluorescence crosstalk correction feature recipe (#35)
 - New ancillary features "fl1_max_ctc", "fl2_max_ctc", "fl3_max_ctc" (#35)
 - Add priority for multiple ancillary features with same name
 - Bugfixes:
   - Configuration key values were not hashed for ancillary features
 - Code cleanup:
   - Refactoring: Put ancillary columns into a new folder module
   - Refactoring: Use the term "feature" consistently
   - Unify trace handling in dclab (#30)
   - Add functions to convert input config data
0.2.9
 - Bugfixes:
   - Regression when loading configuration strings containing quotes
   - Parameters missing when loading ShapeIn 2.0.1 tdms data
0.2.8
 - Refactor configuration class to support new format (#26)
0.2.7
 - New submodule and classes for managing isoelastics
 - New ancillary columns "inert_ratio_raw" and "inert_ratio_cvx"
 - Bugfixes:
   - Typo when finding contour data files (tdms file format)
 - Rrefactoring:
   - "features" submodule with basic methods for ancillary columns
0.2.6
 - Return event images as gray scale (#17)
 - Bugfixes:
   - Shrink ancillary column size if it exceeds dataset size
   - Generate random RTDCBase.identifier (do not use RTDCBase.hash)
     to fix problem with identical identifiers for hierarchy children
   - Correctly determine contour data files (tdms file format)
   - Allow contour data indices larger than uint8
0.2.5
 - Add ancillary columns "bright_avg" and "bright_sd" (#18, #19)
 - Standardize attributes of RTDCBase subclasses (#12)
 - Refactoring:
   - New column names and removal of redundant column identifiers (#16)
   - Minor improvements towards PEP8 (e.g. #15)
   - New class for handling filters (#13)
 - Bugfixes:
   - Hierarchy child computed all ancillary columns of parent upon
     checking availability of a column
0.2.4
 - Replace OpenCV with imageio
 - Add (ancilliary) computation of volume (#11)
 - Add convenience methods for `Configuration`
 - Refactoring (#8):
   - Separate classes for .tdms, dict-based, and hierarchy datasets
   - Introduce "_events" attribute for stored data
   - Data columns (including image, trace, contour) are accessed via
     keys instead of attributes.
   - Make space for new hdf5-based file format
   - Introduce ancilliary columns that are computed on-the-fly
     (new "_ancilliaries" attribute and "ancilliary_columns.py")
0.2.3
 - Add look-up table for elastic modulus (#7)
 - Add filtering option "remove invalid events" to remove nan/inf
 - Support nan and inf in data analysis
 - Improve downsampling performance
 - Refactor downsampling methods (#6)
0.2.2
 - Add new histogram-based kernel density estimator (#2)
 - Refactoring:
   - Configuration fully handled by RTDC_DataSet module (#5)
   - Simplify video export function (#4)
   - Removed "Plotting" configuration key
   - Removed .cfg configuration files
0.2.1
 - Support npTDMS 0.9.0
 - Add AVI-Export function
 - Add lazy submodule for event trace data and rename `RTDC_DataSet.traces`
   to `RTDC_DataSet.trace`
 - Add "Event index" column
0.2.0
 - Compute sensible default configuration parameters for
   KDE estimation and contour plotting
 - Speed-up handling of contour text files
 - Add support for "User Defined" column in tdms files
0.1.9
 - Implement hierarchical instantiation of RTDC_DataSet
 - Bugfix: Prevent instances of PolygonFilter that have same id
 - Load InertiaRatio and InertiaRatioRaw from tdms files
0.1.8
 - Allow to instantiate RTDC_DataSet without a tdms file
 - Add statistics submodule
 - Bugfixes:
   - Faulty hashing strategy in `RTDC_DataSet.GetDownSampledScatter`
 - Code cleanup (renamed methods, cleaned structure)
 - Corrections/additions in definitions (fRT-DC)
0.1.7
 - Added channel: distance between to first fl. peaks
 - Added fluorescence channels: peak position, peak area, number of peaks
 - Allow to disable KDE computation
 - Add filter array for manual (user-defined) filtering
 - Add config parameters for log axis scaling
 - Add channels: bounding box x- and y-size
 - Bugfixes:
   - cached.py did not handle `None`
   - Limiting number of events caused integer/bool error
0.1.6
 - Added `RTDC_DataSet.ExportTSV` for data export
 - Bugfixes:
   - Correct determination of video file in RTDCDataSet
   - Fix multivariate KDE computation
   - Contour accuracy for Defo overridden by that of Circ
0.1.5
 - Fix regressions with filtering.
   https://github.com/DC-analysis/DCscope/issues/43
 - Ignore empty columns in .tdms files (#1)
 - Moved RTDC_DataSet and PolygonFilter classes to separate files
 - Introduce more transparent caching - improves speed in some cases
0.1.4
 - Added support for 3-channel fluorescence data (FL-1..3 max/width)
0.1.3
 - Fixed minor polygon filter problems.
 - Fix a couple of Shape-Out-related issues:
   - https://github.com/DC-analysis/DCscope/issues/17
   - https://github.com/DC-analysis/DCscope/issues/20
   - https://github.com/DC-analysis/DCscope/issues/37
   - https://github.com/DC-analysis/DCscope/issues/38
0.1.2
 - Add support for limiting amount of data points analyzed
   with the configuration keyword "Limit Events"
 - Comments refer to "events" instead of "points" from now on
