Experimental implementation of model tree dynamic Bayesian networks, mtDBN in short
This repository will serve as an initial implementation of the mtDBN model prior to being introduced into the dbnR package. Some of the objectives regarding this model are:
- Test the improvements of the mtDBN model versus a regular DBN model
- Test different kinds of splitting criteria for the tree model
- Test multivariate trees versus univariate
- Test the possibility of a forest model tree
- Implement the mtDBN model as an R6 independent module in order to be able to easily introduce it into dbnR
- Test the convergence of the PSOHO algorithm. It seems to range from pretty amazing networks to ones with abysmal forecasting performance