Stable psoho and natPsoho, transition network whitelist and C++11 relaxation

In this release, I fixed unstable behaviour of regex in C++ on both the psoho and natPsoho algorithms that broke them completely. A new parameter has been added to the DMMHC algorithm to allow forcing arcs in the transition network. There are also some additional small fixes and now the C++11 specification has been removed.

Generic functions, improved docs and generalized inverse matrix

In this release, generic functions have been added for the "dbn" and "dbn.fit" S3 classes. In addition, most of the methods exported for "bn" and "bn.fit" classes from bnlearn are now explicitly extended in dbnR so that they work for DBNs too. The documentation has been sanitized, in the sense that private functions no longer appear on calls to help(package = "dbnR") and exported functions have had their documentation improved. The calculation of the approximate inverse matrix in the exact inference mechanism now usses the function MASS::ginv() instead of a call to solve with less TOL. Several other quality of life improvements have been made, like a fix on forecasting plots to avoid having the predicted values outside the plot window or an argument that allows to plot a DBN with the naming convention reversed, so that the oldest time slice is t_0 instead of the most recent.

Visualization options, flexibility for DMMHC and quality of life improvements

In this release, the size argument no longer has to be given to the forecasting functions after a DBN is created. It has been deprecated, and will be removed in future versions. For now, it just issues a warning. The DMMHC algorithm now supports specific blacklist parameters for inter-slice arcs. Quality of life improvements such as more informative exceptions and subnetwork visualization have been added.

Smoothing and natPSOHO structure learning algorithm

New release with bug fixes and two new features: the possibility of performing smoothing with the DBNs and a new particle swarm structure learning algorithm that scales well for high-order networks. An auxiliary function to automatically and randomly generate DBN structures and synthetic datasets from them is also added.

PSOHO structure learning

New release with some bug fixes and a new structure learning algorithm: particle swarm optimization for higher order DBNs, PSOHO in short

Evidenced forecasting

Added the possibility to provide the forecasting with evidence of the known variables in each of the steps of the prediction. Useful when you know the value of some variables in the future and want to predict the rest or to test scenarios using the network as a simulator.

First release

First version of the package released on CRAN. The main DBN structure learning, forecasting and visualization tools are up and working.