apply UDF with more than one argument in spark_apply() function sparklyr

Question

I have R function with more than one argument and uses dplyr functions in it.

Now, I want to apply this UDF to spark data frame.

The sample code

myfun=function(objdf,x,y,k){

  f <- function(x1,y1,x2,y2) {
    d=(x2-x1) + (y2-y1)
  }
  search=function(df,x,y,k){
    df1=data.frame(cbind(df,f(x,y,df$xx,df$yy)))
    colnames(df1)=c(colnames(df),"val")
    colnames(df1)
    new_df=df1 %>% arrange(val) %>% head(k)
    return(new_df)
  }

  searchwithk <- function(x,y,k) {
    force(x,y,k);
    function(df) search(df,x,y,k)
  }

  res <- spark_apply(objdf, function(df) {
    searchwithk(df,x,y,k)
  })
  return(res)
}

#df= spark_dataframe
x=12.12
y=-74.5
k=5
result=myfun(df,x,y,k)
result

it gives me long error / unused parameter in force statement

How to resolve this?

Answer 1

it gives me long error / unused parameter in force statement

force is an unary function. You cannot pass multiple arguments at once:

searchwithk <- function(x,y,k) {
  force(x)
  force(y)
  force(k)
  function(df) search(df,x,y,k)
}

Also:

• f function doesn't return anything. Should be

  f <- function(x1,y1,x2,y2) { (x2-x1) + (y2-y1) }

• dplyr methods will be out of scope in the closure. You'll probably need

  search=function(df,x,y,k){ library(dplyr) ... }

• You incorrectly call the searchwithk and use incorrect object. Should be

  searchwithk(x,y,k)(df)

• Possibly some other issues.

Answer 2

To add onto user9908499's answer, you can effectively pass as many arguments as you want to a two-parameter function by passing in a list of values using the context parameter.

For example,

searchwithk <- function(df, context) # these two parameters are the only two you should need
{
    library(dplyr) # put any other libraries you need here
    x <- context$x; y <- context$y; k <- context$k
    function(df) search(df,x,y,k) # or whatever you want your code to be
}

res <- df %>% spark_apply(searchwithk,
  context = list(x = x, y = y, k = k)) # put as much as you want in this context