allow converting a data.tree to and from a party class from partykit

[gluc]

as discussed with Achim Zeileis
see http://cran.r-project.org/web/packages/partykit/vignettes/partykit.pdf

[vnijs]

What about the other-way around (i.e., party to data.tree)?

[gluc]

Indeed, that's what I meant. Changed the title.

[vnijs]

Nice! Is there something similar for output from rpart perhaps? I'm interested if these these new functions in data.tree would make it easier to use Diagrammer / Mermaid to visualize trees from partykit / rpart

[gluc]

@vnijs : Not directly for now, but you can use the detour via partykit:

fit <- rpart(Kyphosis ~ Age + Number + Start, data = kyphosis)
prty <- as.party(fit)
tree <- as.Node(prty)
print(tree, 
       "splitname",
       count = function(node) nrow(node$data), 
       "splitLevel")

Yes, you can plot this. E.g.:

 SetNodeStyle(tree, 
              label = function(node) paste0(node$name, ": ", node$splitname), 
              tooltip = function(node) paste0(nrow(node$data), " observations"),
              fontname = "helvetica")

SetEdgeStyle(tree,
             arrowhead = "none",
             label = function(node) node$splitLevel,
             fontname = "helvetica",
             penwidth = function(node) 12 * nrow(node$data)/nrow(node$root$data),
             )

 plot(tree)

[gluc]

Implemented conversion from party -> data.tree (c.f. as.Node.party). data.tree -> party will not be implemented for now, as it would require a very specific structure.

[timelyportfolio]

And a very basic use with networkD3.

library(networkD3)
networkD3::diagonalNetwork(ToListExplicit(tree, unname=TRUE, childrenName="children"))
# add splitLevel as the name
tree$Do(function(node){node$name=node$splitLevel})
networkD3::diagonalNetwork(ToListExplicit(tree, unname=TRUE, childrenName="children"))