cforest prints empty tree

Question

I'm trying to use cforest function(R, party package).

This's what I do to construct forest:

library("party")
set.seed(42)
readingSkills.cf <- cforest(score ~ ., data = readingSkills, 
                         control = cforest_unbiased(mtry = 2, ntree = 50))

Then I want to print the first tree and I do

party:::prettytree(readingSkills.cf@ensemble[[1]],names(readingSkills.cf@data@get("input")))

The result look like this

     1) shoeSize <= 28.29018; criterion = 1, statistic = 89.711
       2) age <= 6; criterion = 1, statistic = 48.324
    3) age <= 5; criterion = 0.997, statistic = 8.917
      4)*  weights = 0 
    3) age > 5
      5)*  weights = 0 
  2) age > 6
    6) age <= 7; criterion = 1, statistic = 13.387
      7) shoeSize <= 26.66743; criterion = 0.214, statistic = 0.073
        8)*  weights = 0 
      7) shoeSize > 26.66743
        9)*  weights = 0 
    6) age > 7
      10)*  weights = 0 
1) shoeSize > 28.29018
  11) age <= 9; criterion = 1, statistic = 36.836
    12) nativeSpeaker == {}; criterion = 0.998, statistic = 9.347
      13)*  weights = 0 
    12) nativeSpeaker == {}
      14)*  weights = 0 
  11) age > 9
    15) nativeSpeaker == {}; criterion = 1, statistic = 19.124
      16) age <= 10; criterion = 1, statistic = 18.441
        17)*  weights = 0 
      16) age > 10
        18)*  weights = 0 
    15) nativeSpeaker == {}
      19)*  weights = 0 

Why is it empty(weights in each node is equal to zero)?

Skydive

Answer 1

Short answer: the case weights weights in each node are NULL, i.e. not stored. The prettytree function outputs weights = 0, since sum(NULL) equals 0 in R.

---

Consider the following ctree example:

library("party")
x <- ctree(Species ~ ., data=iris)
plot(x, type="simple")

For the resulting object x (class BinaryTree) the case weights are stored in each node:

R> sum(x@tree$left$weights)
[1] 50
R> sum(x@tree$right$weights)
[1] 100
R> sum(x@tree$right$left$weights)
[1] 54
R> sum(x@tree$right$right$weights)
[1] 46

Now lets take a closer look at cforest:

y <- cforest(Species ~ ., data=iris, control=cforest_control(mtry=2))
tr <- party:::prettytree(y@ensemble[[1]], names(y@data@get("input")))
plot(new("BinaryTree", tree=tr, data=y@data, responses=y@responses))

The case weights are not stored in the tree ensemble, which can be seen by the following:

fixInNamespace("print.TerminalNode", "party")

change the print method to

function (x, n = 1, ...)·                                                     
{                                                                             
    print(names(x))                                                           
    print(x$weights)                                                          
    cat(paste(paste(rep(" ", n - 1), collapse = ""), x$nodeID,·               
        ")* ", sep = "", collapse = ""), "weights =", sum(x$weights),·        
        "\n")                                                                 
} 

Now we can observe that weights is NULL in every node:

R> tr
1) Petal.Width <= 0.4; criterion = 10.641, statistic = 10.641
 [1] "nodeID"     "weights"    "criterion"  "terminal"   "psplit"    
 [6] "ssplits"    "prediction" "left"       "right"      NA          
NULL
  2)*  weights = 0 
1) Petal.Width > 0.4
  3) Petal.Width <= 1.6; criterion = 8.629, statistic = 8.629
 [1] "nodeID"     "weights"    "criterion"  "terminal"   "psplit"    
 [6] "ssplits"    "prediction" "left"       "right"      NA          
NULL
    4)*  weights = 0 
  3) Petal.Width > 1.6
 [1] "nodeID"     "weights"    "criterion"  "terminal"   "psplit"    
 [6] "ssplits"    "prediction" "left"       "right"      NA          
NULL
    5)*  weights = 0 

Update this is a hack to display the sums of the case weights:

update_tree <- function(x) {
  if(!x$terminal) {
    x$left <- update_tree(x$left)
    x$right <- update_tree(x$right)
  } else {
    x$weights <- x[[9]]
    x$weights_ <- x[[9]]
  }
  x
}
tr_weights <- update_tree(tr)
plot(new("BinaryTree", tree=tr_weights, data=y@data, responses=y@responses))

Comnenus

Thank you for your answer. But, does it mean that cforest constructs empty forest? And what do you think about the reason of such result? – Skydive

The trees are not "empty". The case weights are not available in the tree ensemble, hence the output is misleading. I don't know exactly why, maybe for efficient memory usage. – Comnenus

Ok. But, how can I get weights? – Skydive

Answer 2

The solution proposed by @rcs in the Update is interesting but does not work with cforest when the dependent variable is numerical. The code:

set.seed(12345)
y <- cforest(score ~ ., data = readingSkills,
       control = cforest_unbiased(mtry = 2, ntree = 50))
tr <- party:::prettytree(y@ensemble[[1]], names(y@data@get("input")))
tr_weights <- update_tree(tr)
plot(new("BinaryTree", tree=tr_weights, data=y@data, responses=y@responses))

generates the following error message

R> Error in valid.data(rep(units, length.out = length(x)), data) :
   no string supplied for 'strwidth/height' unit

and the following plot:

Below I suggest an improved version of the hack proposed by @rcs:

get_cTree <- function(cf, k=1) {
  dt <- cf@data@get("input")
  tr <- party:::prettytree(cf@ensemble[[k]], names(dt))
  tr_updated <- update_tree(tr, dt)
  new("BinaryTree", tree=tr_updated, data=cf@data, responses=cf@responses, 
      cond_distr_response=cf@cond_distr_response, predict_response=cf@predict_response)
}

update_tree <- function(x, dt) {
  x <- update_weights(x, dt)
  if(!x$terminal) {
    x$left <- update_tree(x$left, dt)
    x$right <- update_tree(x$right, dt)   
  } 
  x
}

update_weights <- function(x, dt) {
  splt <- x$psplit
  spltClass <- attr(splt,"class")
  spltVarName <- splt$variableName
  spltVar <- dt[,spltVarName]
  spltVarLev <- levels(spltVar)
  if (!is.null(spltClass)) {
    if (spltClass=="nominalSplit") {
     attr(x$psplit$splitpoint,"levels") <- spltVarLev   
     filt <- spltVar %in% spltVarLev[as.logical(x$psplit$splitpoint)] 
    } else {
     filt <- (spltVar <= splt$splitpoint)
    }
  x$left$weights <- as.numeric(filt)
  x$right$weights <- as.numeric(!filt)
  }
  x
}

plot(get_cTree(y, 1))

Franz