0

I have R code that gets one tree out of the random forest algorithm as follows:

testp.rf <- randomForest(new.df, group5, ntree = numoftree, 
        importance = TRUE, proximity = TRUE)
tree <- getTree(testp.rf, k=1, labelVar=TRUE)

It returns a data frame that looks like the below table screenshot. I would like to turn this into a JSON format with the correct tree hierarchy similar to the one used for this D3 example. What is the quickest way to parse the R data frame into this JSON structure.

enter image description here

UPDATE: I added a "parents" column to the data frame to use it as a factor, wrote the data frame to a flat JSON that looks like: 

 [{
"name": 1,
"parent": 0,
"left": 2,
"right": 3,
"split_var": "200862_at",
"split_point": 10.3106,
"isLeaf": 1,
"prediction": "NA"
}, /*...*/ 
]

then used the solution in this answer to parse it in my JS to the hierarchical form that can create a d3.layout.partition. The resulting tree structure looks something like this:

 {
"name": "1",
"parent": "0",
"left": "2",
"right": "3",
"split_var": "200862_at",
"split_point": "10.3106",
"isLeaf": "1",
"prediction": "NA",
"children": [{
    "name": "2",
    "parent": "1",
    "left": "4",
    "right": "5",
    "split_var": "207708_at",
    "split_point": "7.212135",
    "isLeaf": "1",
    "prediction": "NA",
    "children": [{ /*...*/ 
     /*...*/
 }

However,this is still missing important decision tree information like the rules and size of each node. The desired output should look something like the following JSON which I parsed from rpart output:

{
"name": "1",
"rule": "",
"children": [{
    "name": "2",
    "rule": "200862_at < 10.3106",
    "children": [{
        "name": "3",
        "rule": "200862_at < 10.3106 & 207708_at < 7.212135",
        "children": [{
            "name": "4",
              /*...*/
 }]

I am doing the tree generation in R and parsing in JS but I'm open to solutions in any of the two.

Community
  • 1
  • 1
MaYa
  • 180
  • 7
  • something like http://stackoverflow.com/questions/12818864/how-to-write-to-json-with-children-from-r ? – hrbrmstr Jan 17 '17 at 13:21
  • Thanks for commenting. No the example you pointed out considers one column as a factor for another. I found a few ways to build the JSON hierarchy from grouping rows in a data frame based on factors like this, but none that take the structure that is obtained from getTree which has no such factors. – MaYa Jan 17 '17 at 13:24
  • Can you add to your question what your expected output would be? Also, are you looking for an answer in JavaScript or R? – Mark Jan 18 '17 at 00:05
  • @Mark I edited the question to show the desired output as opposed to what I currently have. I am now able to draw an icicle plot in D3 from the hierarchical JSON I copied above but it's missing important information like the list of rules and the size of each node which renders an inaccurate icicle plot. I need a way to add this information to the nodes as I traverse the tree. – MaYa Jan 20 '17 at 08:45

0 Answers0