CHANGES
=======

1.0.3
-----

* CI: appveyor do not use wheelhouse_uploader
* CI: appveyor fix
* DOC: add reminder to update whats_new.rst
* CI: appveyor adjustments
* CI: appveyor do not build Python 2.7.0
* CI: appveyor fixes for Python 3.5
* DOC: Add release notes
* CI: appveyor, attempt to build msi
* CI: appveyor updates for Python 3.5
* Release the GIL during sampling
* DOC: add link to scikit-learn's online LDA
* corrected range in example
* Add comment
* Split computation of PZS_new into multiple lines for clarity
* Don't need to copy PZS_new
* Don't need to initialize PZS_new with vectorization
* Skip initialization (rather, initialize to 0)
* Vectorize a loop
* Compute PZS_sum in fewer steps
* Compute theta_doc in one line
* Rename variable for clarity
* Break out loop of transform() method into a new method
* Rely on fewer local variables (use object members instead)
* Add build instructions
* Fix size of ndz_ in docstring
* Allow any version of pbr
* Remove errant comment
* Transform now handles sparse data
* Make dtm2ldac errors more informative
* Raise error if user passes sparse matrix of floats
* DOC: Update pypi downloads shield
* Allow use of pbr 0.7
* DOC: setup.py build_ext must be run before sdist
* Revert "Removes pbr pre_hook pending investigation"
* NumPy /= is integer division even in Python 3
* Remove waffle.io badge
* DOC: fix RST formatting

1.0.2
-----

* Release 1.0.2

1.0.1
-----

* Release lda 1.0.1

1.0.0
-----

* Removes pbr pre_hook pending investigation
* Ignore .testrepository
* Make lda.datasets available after import lda
* Remove errant comment in CI script
* DOC: Update docs to cover new features
* Set default n_iter to be 2000
* TST: add test for loglikelihoods_
* Track of loglikelihoods for convergence monitoring
* Minor speedup for transform
* TST: Improve and simplify transform tests
* Add transform() via iterated-pseudocount method
* Configure logging following Python guidelines
* DOC: List supported versions of Python
* Reduce number of pre-allocated random variates
* Trim unused code
* Work on copies of random variates during sampling
* Increase number of random variates allocated in advance
* Add Waffle.io badge
* CI: Update Miniconda used by Travis
* TST: add additional tests
* Additional speedup in matrix to list conversion
* DOC: improve docstring and comment in utils
* Minor optimizations
* TST: Add additional tests for matrix to lists
* DOC: add zenodo link
* DOC: Fix link to Pritchard et al. (2000)
* CI: Only build and test Windows wheels on releases, not every commit

0.3.2
-----

* TST: Fix TypeError arising only when using a sparse matrix in Python 3.3
* DOC: update release instructions to reflect change in c code generation
* CI: Fix appveyor, fetch Windows scipy wheels needed for tests
* PEP8
* CI: Generate .c files using Cython in CI setups
* DOC: Mention lda-users mailing list
* Support sparse document-term matrices
* TST: Add tests covering case when user provides sparse matrix
* Remove _lda.c from repository, should be generated using cython
* CI: Add script to help download wheels
* DOC: add more details about release steps
* DOC: Improve documentation index
* DOC: Add PPA install instructions for Ubuntu

0.3.1
-----

* CI: Test against Python 2.7 on Travis
* Reference additional files in MANIFEST.in for Debian packaging
* Flake8 is needed for running tests with tox
* Use wheelhouse-uploader to upload Windows wheels
* DOC: update release howto
* Minor C style adjustments
* Raises error if zero or negative value passed to lgamma
* Check that malloc succeeds during sampling
* DOC: Improve documentation and correct numpydoc errors

0.3.0
-----

* CI: build wheels for Python 3.3 on Windows
* Avoid use of numpy keepdims=True for compatibility with older numpy
* PEP8, order of imports
* Remove dependency on scipy
* DOC: fix typo
* DOC: use apidoc to generate sphinx files
* Update pypi classifiers
* DOC: expand documentation
* DOC: Add note about OS support

0.2.0
-----

* Remove unused import
* Improve getting started example, add supporting data
* Add Windows CI configuration
* Namespace C functions gamma and lgamma
* DOC: Fix typo in README
* Minimally validate eta and alpha
* Remove unused import
* Improve formatting in setup.py
* DOC: minor edits to README
* Generate code with new version of Cython
* Improve Makefile
* Add benchmarking scripts
* Use John Cook's lgamma rather than scipy's gammaln to reduce dependencies
* Rewrite loglikelihood calculation in Cython
* DOC: minor README edits
* Reorder LDA's methods
* Comply with PEP8
* DOC: Add release instructions

0.1.0
-----

* Use oslotest package to simplify tests
* Initial commit
