 Wishlist for the gRain package
====================================

* The .Rd files need re-organization

* For a node which is a discretization of a continuous distribution or
  similar, it could make sense to calculate a posterior mean and
  variance based on the conditional probabilities and the class
  labels.

* Implement IPS on triangulated graph using something like Jirouseks 
  "On the effective implementation of the iterative proportional
  fitting procedure". Use this for fitting graphical models
  where the graph is not decomposable.

* If the network is compiled at the time of a query, then in some
  cases only a smaller network needs to be compiled. 

* When evidence is entered into a potential, the dimension of the
  potential can be reduced. 

* Implement junction tree propagation as alternative to LS

* Implement max-propagation

* Create update function

* Describe the control-list

* Graphs: 
  hg(): hypergraph
  separates: Distinguish between LWF/AMP
  as.matrix(ug); as.matrix(dag)
  as.nel(ug); as.nel(dag)....
  Embed mcs in queryg
  Display dag edges with directions
  vpav should display a|bc...
  Markov blanket = parents + children + parents of children = boundary in moral graph
  Ancestral graph for undirected

- There is a fragile construction in cpt2. 

- demo() ends in a strange way

- Look into discretization of variables; calculate mean/var when a variable is ordinal.

- Create cpt from a dataframe/table...

- Postpone triangulation until evidence E=e* and a set U of nodes of interest are specified.
  Form ancestral graph G' for U \cup E
  Enter evidence
  Remove all edges from evidence nodes
  Update potentials (potentials with evidence have their dimension reduced)
  Triangulate etc.
  
- Optimal triangulation: Lauritzen Sheehan, Statistical science, graphical models in genetics



