CHANGES
=======

2.0.2
-----

* test: faster execution for cube and solid color tests
* fix(trace): skip adding DAF if max is 0
* test: check extremely sparse images (one or two voxels with no dust threshold)
* chore: drop py35 testing add .dockerignore

2.0.1
-----

* fix(windows): use np.uintp before casting to size\_t
* fix: appveyor needs numpy installed first
* chore: new build system for binary distribution

2.0.0
-----

* fix(intake): solid color blocks were causing errors (#56)
* perf: faster somas (#55)
* fix: python3.8 compiles cpp code (#52)
* chore: update travis to use python 3.7 and 3.8
* add python3.8 test

1.6.0
-----

* feat: avocado protection (🥑) (#43)
* chore: update ChangeLog

1.5.0
-----

* chore: add skeleton for manual testing
* feat: add fill\_holes argument (#50)

1.4.2
-----

* chore: loosen networkx requirement (#49)
* Update README.md
* docs: update memory usage diagram for version 1.4.0

1.4.1
-----

* perf: switch source and target for dijkstra

1.3.3
-----

* refactor: make type of 0L clear to std::max on Windows
* Revert "fix: don't assume vertices are uint32"
* fix: don't assume vertices are uint32
* chore: update ChangeLog

1.3.2
-----

* fix: several additional algorithms required 64-bit addressable changes

1.3.1
-----

* chore: bump dijkstra requirement
* fix: 64-bit addressable \_roll\_invalidation\_cube (#42)
* docs: shout out to fill\_voids
* fix: remove unnecessary PIL import

1.3.0
-----

* docs: describe max\_paths in the function docstring
* fix: soma center was being overriden by fix\_borders
* perf: only recompute EDT for soma if some voxels were filled
* perf: use bidirectional dijkstra on somata (increases peak memory usage)

1.2.1
-----

* docs: remove non-ascii character from README.md
* docs: link back to papers using Kimimaro

1.2.0
-----

* docs: show how to use synapses\_to\_targets
* feat: facility for converting synapse centroids into targets (#37)
* refactor+perf: use new fill-voids package

1.1.0
-----

* perf: implemented flood fill based binary\_fill\_holes (#38)

1.0.4
-----

* perf: increase postprocess speed (#35)
* perf: more judicious use of consolidate in postprocess

1.0.3
-----

* docs: update ChangeLog
* fix: preserve skeleton id during postprocessing

1.0.2
-----

* fix: allow multiple invocations of a pathos process pool
* perf: skip processing if dust\_threshold larger than image

1.0.1
-----

* fix: accept any root converable to a tuple
* fix: progress bars were disrupted in parallel feature
* docs: upload changelog

1.0.0
-----

* feat: specify extra\_targets\_before and after (#33)
* docs: fix spelling & grammar

0.7.0
-----

* docs: add parallel\_chunk\_size to README
* perf+feat: Reduce Parallel Task Starvation + Better Parallel Progress Bar (#32)
* docs: add example of join\_close\_components

0.6.0
-----

* feat: adds join\_close\_components to postprocess (#27)
* docs: link to tutorial wiki articles
* docs: add advice on tweaking parameters

0.5.4
-----

* fix: sometimes get\_mapping doesn't get everything

0.5.3
-----

* fix: object\_ids were being masked instead of mask\_excepted
* docs: show performance chart for v0.5.2

0.5.2
-----

* perf: improve performance of find\_objects 7x

0.5.1
-----

* perf: ~20x faster unique(label, return\_counts=True) (#26)
* docs: changelog update + small formatting adjustment to example

0.5.0
-----

* docs: example of how to use postprocess
* feat: import out-of-core postprocessing logic from Igneous
* docs: add object\_ids to example
* perf: improve speed of skeletontricks.get\_mapping
* fix: accept binary images of type bool
* perf: take advantage of faster segid finding if dust\_threshold == 0
* fix: compilation warning for \_roll\_invalidation\_cube
* test: add some manual visualization tests
* chore: update ChangeLog

0.4.2
-----

* release: 0.4.2
* chore: tell PyPI we're using markdown
* fix: ensure we pick max dbf close to centroid of detected somata
* chore: update ChangeLog
* docs: various corrections to the README

0.4.1
-----

* fix: add defense against setting the dust threshold lower than 1
* chore: formatting around all\_labels
* test: x and y joinability
* test: show that two 1px overlapping volumes join properly
* Update README.md
* feat: accept N-dimensional arrays with trivial dimensions above 3
* docs: add Google TEASAR run to boslster case for popularity
* fix: prevent duplicate border targets
* feat: parallel edt implementation
* fix: add support for anisotropy to distance calculations
* test: add distortion to border test
* wip: propogate anisotropy to fix\_borders calls
* fix: cuboid soma processing
* fix: bump edt to 1.2.4 to correct part of large anisotropy issue
* perf: faster masking operations with newer fastremap
* docs: encouraged the use of parallel processing in README.md
* chore: add GPLv3 classifer to setup.cfg
* chore: add ChangeLog

0.4.0
-----

* feat: parallel implementation (Cursed Seal Mode) (#10)

0.3.1
-----

* fix: INTEGER type did not include all integers

0.3.0
-----

* docs: updated credits with fix\_borders
* feat: add fix\_borders parameter & max\_paths parameter (#9)
* test+fix: remove "cd python"
* docs: add Travis CI badge
* chore: add Travis CI
* test: add basic test for skeletonizing diagonal of square and cube
* perf: improve memory consumption of object masking
* perf: introduce in\_place flag to make it safe to modify input data
* perf: use fastremap's new in\_place flag for lower memory and perf
* docs: updated credits

0.2.2
-----

* fix: accept C order arrays (#7)
* docs: reduce redundancy in example vs performance
* docs: add benchmark description
* docs: added benchmark photo
* docs: add link to citation 4
* docs: use citations 3 and 4
* docs: described "roll invalidation cube"
* docs: described algorithm in steps
* docs: describing the algorithm

0.2.1
-----

* fix: black volumes should return dict not None

0.2.0
-----

* docs: add PyPI badge
* feat: fix branching (#1)
* docs: adding sections to README

0.1.0
-----

* chore: clean up dockerfile and metadata
* docs: draft discussion of motivation and usage
* feat: export DimensionError exception (so it can be caught)
* refactor: remove path\_downsample from trace function
* docs: described parameters of skeletonize function
* chore: files required for building distributions
* wip: importing skeletonization procedure
* Initial commit
