Forum: help

RE: define tree from list [ Reply ] By: Markus Loecher on 2016-08-07 12:28	[forum:43501]
Thanks a lot for this very useful perspective. I think I will simply create a "fake" rpart object then, great pointer. Best Markus

RE: define tree from list [ Reply ]
By: Achim Zeileis on 2016-08-06 13:06

[forum:43410]

Markus,

there are coercion methods from "partynode" to "list" and vice versa. With the WeatherPlay example you can do

R> (pl <- as.list(py$node))
[[1]]
[[1]]$id
[1] 1

[[1]]$split
$varid
[1] 1

$breaks
NULL

$index
[1] 1 2 3

$right
[1] TRUE

$prob
NULL

$info
NULL

attr(,"class")
[1] "partysplit"

[[1]]$kids
[1] 2 5 6

[[2]]
[[2]]$id
[1] 2

...

R> as.partynode(pl)
[1] root
| [2] V1 in (-Inf,1]
| | [3] V3 <= 75 *
| | [4] V3 > 75 *
| [5] V1 in (1,2] *
| [6] V1 in (2, Inf]
| | [7] V4 <= 1 *
| | [8] V4 > 1 *

HOWEVER: This requires that you know in advance the IDs of kid nodes. So I think for the format you have this isn't so simple.

Therefore, I would probably recommend to simply grow the tree recursively. See for example partykit:::as.party.rpart on how to do that. The core you will be interested in is the rpart_node() function that is called with rpart_node(1) and from there called recursively until the whole tree has been grown.

define tree from list [ Reply ]
By: Markus Loecher on 2016-08-06 12:15

[forum:43409]

I really like the WeatherPlay example on pp.2-5 of the partykit vignette.
The explicit code to construct the entire tree on p.4 is also great but maybe a bit too "manual" or static to be generalized easily.
So I wonder how one would construct a tree "dynamically" from a file which contains the split informations in a recursive fashion.
For example, I am using software that records its generated trees to a file in the following format:

split 0 5 1 9
split 1 10 6 14
split 0 3 1 5
leaf 1 0
leaf 1 2
split 1 12 10 14
leaf 1 1

For a split the first number is the variabe index (starting at 0 in the C convention), the second number the split value (e.g. left and right from var1==5 ) and the last two numbers represent the current bounding box of the respective variable.

Are there any existing functions that might help reading such a tree into a party structure?

Thanks!
Markus