R XML count sub-nodes for each node x

Question

I have large XML with PropertyItemData nodes that can contain ObjectList node. I want to get list/vector of ObjectList subnode count for each PropertyItemData.

Generation of example data:

doc <-
'
<a>
<PropertyItemData>
  <ObjectList>
    <ObjectData><z>1</z></ObjectData>
  </ObjectList>
</PropertyItemData>
<PropertyItemData>
  <ObjectList>
    <ObjectData><z>1</z></ObjectData>
    <ObjectData><z>1</z></ObjectData>
  </ObjectList>
</PropertyItemData>
<PropertyItemData>
</PropertyItemData>
</a>
'

n <- 300 * 1000
doc2 <- paste(lapply(1:n, function(x) doc), collapse = '')
doc2 <- sprintf('<b>%s</b>', doc2)

Current approach:

library(XML)
xx <- xmlParse(doc2)
b <-  getNodeSet(xx, "//PropertyItemData") # get all PropertyItemData
s2 <- sapply(b, xpathSApply, ".//ObjectList", xmlSize) # for each count ObjectList sub-nodes
s2[lengths(s2) == 0L] <- 0L # if no subnodes = 0
s2 <- unlist(s2)
head(s2)
# [1] 1 2 0 1 2 0

Is there faster way to obtain desired results? Can this be done with XPath?

Answer 1

This alternative delivers a time improvement of over 80% with your data.

library(xml2)
n <- 300 * 1000 
doc2 <- paste(lapply(1:n, function(x) doc), collapse = '')
doc2 <- sprintf('<b>%s</b>', doc2)

xx2 <- read_xml(doc2)
b2 <- xml_find_all(xx2, "//PropertyItemData")

result <- sapply(b2, function(y) xml_length(xml_children(y)))
identical(result, s2)
# [1] TRUE

---

Benchmarking

original_sol <- function(){
  xx <- xmlParse(doc2)
 b <-  getNodeSet(xx, "//PropertyItemData")
s2 <- sapply(b, xpathSApply, ".//ObjectList", xmlSize)
s2[lengths(s2) == 0L] <- 0L
s2 <- unlist(s2)
}
new_sol <- function(){
  xx2 <- read_xml(doc2)
  b2 <- xml_find_all(xx2, "//PropertyItemData")
  result <- sapply(b2, function(y) xml_length(xml_children(y)))
}

library(microbenchmark)
microbenchmark(
  original_solution = original_sol(),
  new_solution = new_sol(),
  times=1
)
Unit: seconds
              expr       min        lq      mean    median        uq       max neval
 original_solution 120.47773 120.47773 120.47773 120.47773 120.47773 120.47773     1
      new_solution  25.35973  25.35973  25.35973  25.35973  25.35973  25.35973     1

---

EDIT if you need more specific methods than counting all children of PropertyItemData nodes, here are 2 other methods:

• same algorithm as you used (listing the children of ObjectList nodes),
• Counting the children ObjectData nodes.

# same method as OP's: length of children of ObjectList nodes -------------
objectlist <- lapply(b2, function(x) xml_find_all(x, "ObjectList"))
result_same_algorithm <- sapply(objectlist, function(x) sum(xml_length(xml_children(x))))
identical(result_same_algorithm, s2)
# third possibility: length of Object data children nodes --------------------------
res_objectdata_children <- sapply(b2, function(x) sum(xml_length(xml_find_all(x, ".//ObjectData"))))
identical(res_objectdata_children, s2)