slise - Version 2.2.3

SLISE now works with Python 3.11 (and probably soon 3.12, when numba gets updated) .

- Python
Published by Aggrathon about 2 years ago

slise - Version 2.2.2

What's Changed

• Numba tweaks by @Aggrathon in https://github.com/edahelsinki/pyslise/pull/9
  • Default to num_threads = 1 to avoid horrible performance on some CPUs
  • Add signatures to jitted functions (for faster and more reusable jitting)

- Python
Published by Aggrathon about 2 years ago

slise - Version 2.2.1

What's Changed

• Automatically ravel Y (in case of a matrix)
• Since (the dependency) PyLBFGS does not work with cython >= 3.0, the requirements has been updated

- Python
Published by Aggrathon about 2 years ago

slise - Version 2.2.0

Rename impact to terms and add a citation for the newly published paper on using SLISE for local explanations.

- Python
Published by Aggrathon over 2 years ago

slise - Version 2.1.3

Improve the documentation (and build an online documentation webpage)

- Python
Published by Aggrathon over 3 years ago

slise - Numba tweaks

Add nogil=Trueto the jitted functions and handle NUMBA_DISABLE_JIT=1.
Mention optional Numba dependencies in the README.md.

- Python
Published by Aggrathon almost 4 years ago

slise - Threading

• Added option for limiting the number of threads used.
• Added new helpful warnings (such as warning about potentially bad numba.threading_layer()).
• Deprecate (only warning for now) get_params() in favour of coefficients.

- Python
Published by Aggrathon almost 4 years ago

slise - Weights

Changes: - Add optional weights to the algorithm. - Do not normalise logits. - Make some fields "private". - Increase version to match the R variant.

- Python
Published by Aggrathon about 4 years ago

slise - SLISE on PyPI

SLISE is now available on PyPI and can be installed with: sh pip install slise

- Python
Published by Aggrathon over 4 years ago

slise - Consider Normalisation

This release fixes a bug when using normalise=True in SLISE-regression.

Furthermore, the impact calculated from normalised and unnormalised values tells different stories. Unnormalised impact lets you reconstruct the original prediction, while normalised impact is more of a comparison to the rest of the data. Thus, they are given separate rows when using the built in print and plot_dist functions.

- Python
Published by Aggrathon over 4 years ago

slise - Improved Examples

Some of the built-in plotting functions have been improved. Warnings from intermediate optimisation steps are now hidden by default (warnings from the last optimisation step are still shown). Finally, the examples have been extended and improved, especially in regards to how to interpret the results.

- Python
Published by Aggrathon almost 5 years ago

slise - The Python version of SLISE

First public release of the Python version of the SLISE algorithm.

- Python
Published by Aggrathon almost 5 years ago