PymaBandits contains **python and matlab implementations of algorithms for multi-armed bandit problems**. The code has been written with Aurélien Garivier and Emilie Kaufmann and was used to perfom the simulations in

- E. Kaufmann, O. Cappé, and A Garivier.
On Bayesian Upper Confidence Bounds for Bandit
Problems.
In
*Proc. AISTATS, JMLR W&CP*, volume 22, pages 592-600, La Palma, Canary Islands, April 21-23 2012. - A. Garivier and O. Cappé.
The KL-UCB Algorithm for Bounded Stochastic
Bandits and Beyond.
In
*COLT*, Budapest, Hungary, July 2011.

PymaBandits can be downloaded here.

Wapiti is a **very fast CRF toolkit** (written in
C++) for segmenting and labeling items and sequence data written and maintained by Thomas Lavergne; Wapiti was developed within the CroTAL project.

Wapiti's implementation is described in

- T. Lavergne, O. Cappé, and F. Yvon. Practical Very Large Scale CRFs. In Proc. 48th Annual Meeting Association for Computational Linguistics (ACL), pages 504-513, Uppsala, Sweden, July 2010.

and is based, partly, on

- N. Sokolovska, T. Lavergne, O. Cappé, and F. Yvon. Efficient Learning of Sparse Conditional Random Fields for Supervised Sequence Labelling. IEEE J. Sel. Topics Signal Process., 4(6):953-964, December 2010.

Wapiti is also available from the public evaluation site MLcomp and its performance appears to be very good on the *SequenceTagging* tasks (best program from the repository on 5 of the 16 datasets as of August 2010).

OnlineHMM contains the source code of MATLAB routines implementing the **online Expectation-Maximization algorithm for the 'Markov chain in noise' HMM**. It is available on the *Journal of Computational and Graphical Statistics* web site as the supplementary material of

- O. Cappé. Online EM Algorithm for Hidden Markov Models. J. Comput. Graph. Statist., 20(3):728-749, September 2011.

OnlineEM contains the MATLAB/OCTAVE routines implementing the **online Expectation-Maximization algorithm** used for the numerical simulations presented in

- O. Cappé. Online Expectation-Maximisation. In K. Mengersen, M. Titterington, and C. P. Robert, editors, Mixtures: Estimation and Applications. Wiley, 2011.
- O. Cappé and E. Moulines. On-line Expectation-Maximization Algorithm for Latent Data Models. J. Royal Statist. Soc. B, 71(3):593-613, 2009.

CosmoPMC is a C-written environment for
running **adaptive population Monte Carlo** for cosmology applications. CosmoPMC supports parallelism using MPI. It was written mostly by Martin Kilbinger, Karim Benabed and Simon Prunet for the ECOSSTAT project. Related papers are

- M. Kilbinger, D. Wraith, C. P. Robert, K. Benabed, O. Cappé, J-F. Cardoso, G. Fort, S. Prunet, and F. R. Bouchet. Bayesian model comparison in cosmology with Population Monte Carlo. MNRAS, 45:2381-2390, 2010.
- D. Wraith, M. Kilbinger, K. Benabed, O. Cappé, J-F. Cardoso, G. Fort, S. Prunet, and C. P. Robert. Estimation of cosmological parameters using adaptive importance sampling. Phys. Rev. D, 80(2), 2009.

Ihmm are the MATLAB/OCTAVE functions that where
used for some of the simulations featured in the book *Inference in Hidden Markov Models*
(Cappé, Rydén and Moulines - Springer, 2005), in particular for Kalman smoothing
and (particle) resampling.

Currently available code includes functions for implementing

- Kalman filtering and smoothing (with all three forms of smoothing: RTS, disturbance and Backward Information Recursion) for general (non-stationary) linear state-space models.
- Resampling in a matlab efficient way (for multinomial, stratified, systematic and residual resampling).

CT/RJ-mix implements
**transdimensional Markov Chain Monte Carlo (MCMC) for inference in (scalar)
Gaussian mixture models**, with unknown number of components. Two methods are
implemented: Reversible Jump (RJ) MCMC for `rj_mix`

and Continuous
Time (CT) for `ct_mix`

This is the C code that was used to produce the figures in Section 4 of

O. Cappé, C. Robert and T. Rydén.Reversible jump, birth-and-death and more general continuous time Markov chain Monte Carlo samplers. Journal of the Royal Statistical Society: Series B (Statistical Methodology), Volume 65, Issue 3, pages 679-700, 2003.

Please see this page for more information.

H2m is a set of MATLAB/OCTAVE functions for the
EM **estimation of mixture and hidden Markov models**. h2m includes functions for
Poisson and negative binomial models in addition to the multivariate Gaussian
ones.

These functions (together with the corresponding documentation) are available as gz-compressed unix archive and PC zip file.

For more information please take a look at HTML documentation for H2m or the Pdf documentation

Env contains a set
of MATLAB functions that implement the **spectral envelope estimation methods**
described in

*M. Campedel-Oudot, O. Cappé, and E. Moulines. Estimation of the spectral envelope of voiced sounds using a penalized likelihood approach. IEEE Trans. Speech Audio Process., 9(5):469-481, July 2001.*

The gz-compressed unix archive also contains the signal and results shown in section IV.D of the paper.

Dcv is a set of MATLAB functions for testing various **maximum likelihood
and Bayesian blind deconvolution/estimation procedures in the case of
discrete input signals**. These functions (together with the corresponding
documentation) are available as a compressed unix archive. For more
information please see:

- the HTML documentation, or
- the PDF documentation for dcv.