felixcheung commented on a change in pull request #23746: [SPARK-26761][SQL][R]
Vectorized R gapply() implementation
URL: https://github.com/apache/spark/pull/23746#discussion_r255841911
##########
File path: R/pkg/inst/worker/worker.R
##########
@@ -181,17 +185,35 @@ if (isEmpty != 0) {
colNames, computeFunc, data)
} else {
# gapply mode
+ outputs <- list()
for (i in 1:length(data)) {
# Timing reading input data for execution
inputElap <- elapsedSecs()
output <- compute(mode, partition, serializer, deserializer,
keys[[i]],
colNames, computeFunc, data[[i]])
computeElap <- elapsedSecs()
- outputResult(serializer, output, outputCon)
+ if (deserializer == "arrow") {
+ requireNamespace("arrow", quietly = TRUE)
+ outputs[[length(outputs) + 1L]] <- output
+ } else {
+ outputResult(serializer, output, outputCon)
+ }
outputElap <- elapsedSecs()
computeInputElapsDiff <- computeInputElapsDiff + (computeElap -
inputElap)
outputComputeElapsDiff <- outputComputeElapsDiff + (outputElap -
computeElap)
}
+
+ if (deserializer == "arrow") {
+ requireNamespace("arrow", quietly = TRUE)
+ # See
https://stat.ethz.ch/pipermail/r-help/2010-September/252046.html
+ # rbind.fill might be an anternative to make it faster if plyr is
installed.
+ combined <- do.call("rbind", outputs)
Review comment:
shouldn't we turn it off (as `stringsAsFactors = FALSE`)?
If I recall, we don't handle factor to JVM correctly, and compute() from the
user function doesn't **really** have to return a data.frame per se.
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
With regards,
Apache Git Services
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
