Dppy

New version following 62

danielecc contributed with an implementation of a fast exact Finite DPP sampling algorithm which does not require looking at all items and applies to both DPPs and k-DPPs.

*[Sampling from a k-DPP without looking at all items](https://papers.nips.cc/paper/2020/hash/4d410063822cd9be28f86701c0bc3a31-Abstract.html)*
Daniele Calandriello, Michal Derezinski, Michal Valko, NeurIPS, 2020.

danielecc simplified installation instructions to work with minimal dependencies: `numpy`, `scipy`, `matplotlib`.
See also [`README`](https://github.com/guilgautier/DPPy) for more details.
Additional dependencies:
- `zonotope` for the zonotope MCMC based sampler using `cvxopt`,
- `trees` for uniform spanning tree samplers using `networkx`,
- `docs` for the documentation using `sphinxcontrib-bibtex`and `sphinx_rtd_theme`,

can be installed locally after cloning the repo.

guilgautier contributed with (see also [`/notebooks`](https://github.com/guilgautier/DPPy/tree/master/notebooks)):
- an exact sampler for multivariate Jacobi ensembles used to do Monte Carlo integration
[*On two ways to use determinantal point processes for Monte Carlo integration*](https://papers.nips.cc/paper/8992-on-two-ways-to-use-determinantal-point-processes-for-monte-carlo-integration)
G. Gautier, R. Bardenet, M.Valko, NeurIPS, 2019.

- a Markov chain based sampler for beta-ensembles with polynomial potential
[*Fast sampling from beta-ensembles*](http://arxiv.org/abs/2003.02344)
G. Gautier, R. Bardenet, M.Valko, arXiv preprint, 2020.

danielecc contributed with an implementation of the `vfx` sampler, the associated documentation and tests.

In practice
python
DPP = FiniteDPP('likelihood', **{'L_eval_X_data': (eval_L, X_data)})
DPP.sample_exact(mode='vfx')


See the corresponding NeurIPS 2019 paper of Derezinski, Calandriello, and Valko [Exact sampling of determinantal point processes with sublinear time preprocessing](https://papers.nips.cc/paper/9330-exact-sampling-of-determinantal-point-processes-with-sublinear-time-preprocessing)

Since the last submission, we have put efforts on:
- the coverage rate 14% -> 90%
- [Travis build + 90% coverage](https://travis-ci.com/guilgautier/DPPy/builds/123017755)
- [Coveralls](https://coveralls.io/github/guilgautier/DPPy?branch=master)
- the documentation with better explanations, illustrations, docstrings
- [Documentation build](https://readthedocs.org/projects/dppy/builds/9510168/)
- the implementation of the [multivariate Jacobi Ensemble](https://dppy.readthedocs.io/en/latest/continuous_dpps/multivariate_jacobi_ope.html), used for Monte Carlo integration

Companion paper:
- [arXiv v2](https://arxiv.org/abs/1809.07258)
- [GitHub](https://github.com/guilgautier/DPPy_paper)

We feel the project mature enough to be released on [PyPI](https://pypi.org/project/DPPyPI/) and to be submitted to the special [MLOSS](http://jmlr.csail.mit.edu/mloss/) track of JMLR. The last version of the corresponding companion paper can be found at [DPPy_paper](https://github.com/guilgautier/DPPy_paper).