Abess

-------------

- R package
- Python package

- Support no-intercept model for most regressors in ``abess.linear`` with argument ``fit_intercept=False``. We assume that the data has been centered for these models.
- Use `CMake <https://cmake.org/>`__ on compiling to increase scalability.

- C++

- Support limiting beta into a range by clipping method. One application is to perform non-negative fitting.
- Support AUC criterion for Logistic and Multinomial Regression.

-------------

- R package
- Python package

- Support ``score`` function for all GLM estimators.
- Rearrange some arguments to improve legibility.
Please check `here <https://abess.readthedocs.io/en/latest/Python-package/index.html>`__ for the latest API.
- Better docstring, e.g. move important arguments to the front.
- Combine ``metrics.py`` and ``functions.py``.

- C++

- Support the base model for GLM. The Sparse GLM model can be implemented much easilier.
- Re-write logistic, poisson and gamma regression on the basis of GLM base model.

-----------------------

- R package

- Change the structure of R package such that the parameter check can be reused by different methods.
As a by-production, code coverage for R package is impressively improved.
- Support ordinal regression
- Update README.md to synchronize with the layout change of abess official website.

- Python package

- Fix bugs in sparse principal component analysis
- Support ordinal regression
- Support predicting survival function in ``CoxPHSurvivalAnalysis()``
- Modify python package to adapt to the criteria of `conda-forge <http://conda-forge.org>`__ and ``abess`` is going to appear on conda-forge.
- Spectra library is no long appear in ``python/include`` directory
- Improve pytest by suppress unnecessary come from ``scikit-learn`` and the warning about API-name change.
Moreover, some test will be skipped if some dependencies are missing.
- Add `estimator check <https://scikit-learn.org/stable/modules/generated/sklearn.utils.estimator_checks.check_estimator.html>`__
from ``scikit-learn`` into pytest
- Refine the configuration in ``setup.py`` to facilitate the source code installation
- Support ``get_params`` and ``set_params`` methods for each model

- C++

- Support ordinal regression
- Fix bugs for multiple-regressors' API
- Add more comments to improve readability, mainly in ``Algorithm.h``, ``utilities.h``, and ``workflow.h``

- Project development

- Test the package automatic submission. (It explains why the version number is quickly shifted.)
- Python maintainer changes from `Kangkang Jiang <https://github.com/Jiang-Kangkang>`__ to
`Junhao Huang <https://github.com/oooo26>`__!

-------------

- R package

- Support user-specified initial active set.

- Python package

- The API name shifts from ``abessXXX`` to ``xxxRegression`` and from ``abessXXX`` to ``SparsePCA``
- Improve the PEP8 criteria and ``scikit-learn`` criterion
- The interface between python and cpp changes from `swig <http://www.swig.org/>`__ to `pybind11 <https://pybind11.readthedocs.io/en/stable/>`__.
- On Windows, the recommended C++ compiler for ``abess`` package installation shifts from Mingw to Microsoft Visual Studio because it is suggested that `MinGW works with all Python versions up to 3.4 <https://wiki.python.org/moin/WindowsCompilers#GCC_-_MinGW-w64_.28x86.2C_x64.29>`__.
- Using `cibuildwheel <https://cibuildwheel.readthedocs.io/en/stable/>`__ and github action to build and test `wheel` files automatically
- Fix bugs in sparse principal component analysis

- Project development

- Documentation

- Add instruction for Gamma regression.
- Update the usage of ``support_size`` in PCA.
- Use Sphinx-Gallery for website layout, and update the layout of the ``Tutorial`` section.

-------------

It is the fourth stable release for ``abess``. More features and
concrete algorithms are supported now and the main Cpp code has been
refactored to improve scalability.

- Cpp

- New features:

- Support user-specified cross validation division.
- Support user-specified initial active set.
- Support flexible support size for sequentially best subset
selection for principal component analysis (PCA).

- New best subset selection tasks:

- Generalized linear model when the link function is gamma
distribution.
- Robust principal component analysis (RPCA).

- Performance improvement:

- Bug fixed

- Python

- New best subset selection features and tasks implemented in Cpp
are wrapped in Python functions.
- More comprehensive test files.
- A new release in Pypi.

- R package

- New best subset selection features and tasks implemented in Cpp
are wrapped in R functions.
- A new release in CRAN.

- Project development

- Source code

- Refactoring the Cpp source code to improve its readability and
scalability. Please check `Code
Developing <https://abess.readthedocs.io/en/latest/Contributing/CodeDeveloping.html>`__
section for more details.
- Combine all parameters (e.g. support_size and lambda) in one
list to improve expandability.
- Move the core code ``src`` directory to the root of repository.

- Documentation

- Add instruction for robust principal component analysis in
`Tutorial <https://abess.readthedocs.io/en/latest/Tutorial/RPCA.html>`__.
- Add instruction for user-specified cross validation division in
`Advanced
Features <https://abess.readthedocs.io/en/latest/Tutorial/advanced_featureshtml#User-specified-cross-validation-division>`__.
- Update development guideline according to cpp source code
change in `Code
Developing <https://abess.readthedocs.io/en/latest/Contributing/CodeDeveloping.html>`__.
- Adding more details and giving more links related to core
functions.

- Code coverage

- Add more test suites to improve coverage and stability

- Code format

- Code format is checked by
`CodeFactor <https://www.codefactor.io/repository/github/abess-team/abess>`__.
For more details, please check `Code
Format <https://abess.readthedocs.io/en/latest/Contributing/Formatting.html>`__.

-------------

It is the third stable release for ``abess``. This version improve the
runtime performance, the clarity of project’s documentation, and add
helpful continuous integration.

- Cpp

- New features:

- Support important searching to significantly improve
computational efficiency when dimensionality is large.

- Performance improvement:

- Update the version of dependencies: from Spectra 0.9.0 to 1.0.0
- Bug fixed

- R package

- Support important searching for generalized linear model in
``abess``
- A new release in CRAN.

- Python package

- Remove useless parameter to improve clarity.
- Support important searching for generalized linear model
``abessLm``, ``abessLogistic``, ``abessPoisson``, ``abessCox``,
``abessMlm``, ``abessMultinomial``
- A new release in Pypi.

- Project development

- Code coverage

- Check line covering rate for both Python and R. And the
coverage rates are summarized and report.
- Add more test suites to improve coverage and stability

- Documentation

- Add docs2search for the R package’s website
- Add a logo for the project
- Improve documentation by adding two tutorial sections:
``detail of algorithm`` and ``power of abess``.

- Improve code coverage
- Continuous integration

- Check the installation in Windows, Mac, and Linux
- Automatically generate the ``.whl`` files and publish the
Python package into Pypi when tagging the project in github.