...
- Lower memory footprint for sparse targets in multilabel classification (previously converted to dense arrays) #61
- Hubness estimation could fail when ANN does not return enough neighbors #59
- Heuristic to choose memory for Puffinn LSH.
- Switch to modern Python packaging with
pyproject.toml
andsetup.cfg
- Switch to Github Actions, dropping Travis CI and AppVeyor
0.21.2 - 2020-01-14
This is a maintenance release due to the publication in the Journal of Open Source Software.
0.21.1 - 2019-12-10
This is a bugfix release due to the recent update of scikit-learn to v0.22.
-
Require scikit-learn v0.21.3.
Until the necessary adaptions for v0.22 are completed, scikit-hubness will require scikit-learn v0.21.3.
0.21.0 - 2019-11-25
This is the first major release of scikit-hubness.
- Enable ONNG provided by NGT (optimized ANNG). Pass
optimize=True
toLegacyNNG
. - User Guide: Description of all subpackages and common usage scenarios.
- Examples: Various usage examples
- Several tests
- Classes inheriting from
SupervisedIntegerMixin
can be fit with anApproximateNearestNeighbor
orNearestNeighbors
instance, thus reuse precomputed indices.
- Use argument
algorithm='nng'
for ANNG/ONNG provided by NGT instead of'onng'
. Also setoptimize=True
in order to use ONNG.
- DisSimLocal would previously fail when invoked as
hubness='dis_sim_local'
. - Hubness reduction would previously ignore
verbose
arguments under certain circumstances. HNSW
would previously ignoren_jobs
on index creation.- Fix installation instructions for puffinn.
0.21.0a9 - 2019-10-30
- General structure for docs
- Enable NGT OpenMP support on MacOS (in addition to Linux)
- Enable Puffinn LSH also on MacOS
- Correct mutual proximity (empiric) calculation
- Better handling of optional packages (ANN libraries)
- streamlined CI builds
- several minor code improvements
- Silvan David Peter
0.21.0a8 - 2019-09-12
- Approximate nearest neighbor search
- Several minor issues
- Several documentations issues
0.21.0a7 - 2019-07-17
The first alpha release of scikit-hubness
to appear in this changelog.
It already contains the following features: