h5features-2.0.0

• Breaking change: Complete rewrite of the codebase in C++ using HighFive, with Python bindings using nanobind and built with scikit-build-core.
• The Python API is made of four classes: Item, Reader, Writer and Version. There is no top-level functions like h5features.write or h5features.read anymore.
• The Python package is distributed with wheels on PyPI only, the conda channel will not be updated anymore.
• The documentation is available at: https://docs.cognitive-ml.fr/h5features.

h5features-1.4.1

• Fixed a bug introduced in cersion 1.4 with pip install h5features (missing requirements.txt).
• This minor release is the first to be compatible with h5py>=3.0.
• No more tested on python2.

h5features-1.3.3

ChangeLog

Improvements in automatic deployment to pypi and conda.

h5features-1.3.2

Changelog

• Bugfix: no more size limitation on properties.

• properties are now exposed at the top-level (legacy) h5features.write and h5features.read functions.

h5features-1.3.1

• Bugfix: properties are now appendable to an existing group.
• Bugfix: fixed issue #5.

h5features-1.3

• Add an optional properties field in the dataset to store a
  dictionary of various entries (basically features parameters).

• chunk_size is automatically computed by default in the writer.

• New optional compression in h5features.Writer.

h5features-1.2.2

• bugfix: broken test on python-3.6.3.

• bugfix: missing files in MANIFEST.in for installation with pip install h5features.

h5features-1.2.1

• The script convert2h5features is now installed by the setup
  script.

• Removed deprecated tests since 1.2

• Added a VERSION file for tracking version number in the whole
  project.

h5features-1.2

What’s new in h5features 1.2

• Breaking change Labels associated with features data must be sorted in increasing order. This is
  convenient to use with timestamps and improve reading huge datasets with long labels.

• Breaking change Appending new data to an exisiting item is no more allowed. Suppose a h5f file with 3 items ['a', 'b', 'c'], in 1.1 it was possible to append 3 items ['c', 'd', 'e'], giving a file with the 5 items ['a', 'b', 'c', 'd', 'e'], where the item 'c' being the concatenation of original and appended data. That facility was messy and is removed in 1.2.

• Bugfix when writing unidimensional features

• Bugfix when reading from time/to time in Reader

• Safely overwrite existing groups in h5features files with mode=’w’

• Now more than 100 test cases

h5features-1.1

Major package refactoring with backward compatibility