Random sampling over XY coordinates in R (or in Matlab ??)

Question

My data frame has the following four columns: type("A" or "B"), xvar, longitude, and latitude. It looks like:

      type    xvar    longitude    latitude
[1,]   A       20      -87.81        40.11
[2,]   A       12      -87.82        40.12
[3,]   A       50      -87.85        40.22
....
[21,]  B       24      -87.79        40.04
[22,]  B       30      -87.88        40.10
[23,]  B       12      -87.67        40.32
[24,]  B       66      -87.66        40.44
....

I have 20 rows for type="A", and 25,000 rows for type="B". My task is to randomly assign the values of xvar for 20 "A" data points onto the XY space of type "B" without replacement. For example, the xvar=20 as in the first observation of type="A" can be randomly located in [22,] that is (-87.88,40.10) . Because I am doing that without replacement, in theory, I can do this replication 25,000/20 = 1,250 times. I want a 1,000 replication.

And I have a function (say, myfunc(xvar,longitude,latitude)) that returns one statistical value from one randome sample. I first create an empty matrix (say, myresult) of 1,000x1.

myresult <- array(0,dim=c(1000,1))

Then, for each random sample, I apply my function (myfunc) to calculate the statistic.

for (i in seq(1:1000)) {
  draw one sample, that has three variables: xvar, longitude, latitude.
  apply my function to this selected sample.
  store the calculated statistic in the myresult[i,]
}

I wonder how to do this in R. (And may be in Matlab??) Thanks!

=============================================================

Update: @user. Borrowing your idea, the following is what I want:

dd1 <- df[df$type == "B" ,] 
dd2 <- df[df$type == "A" ,]
v   <- dd2[sample(nrow(dd2), nrow(dd2)), ]
randomXvarOfA <- as.matrix(v[,c("xvar")])  
cols <- c("longitude","latitude")
B_shuffled_XY <- dd1[,cols][sample(nrow(dd1), nrow(dd2)), ]
dimnames(randomXvarOfA)=list(NULL,c("xvar"))
sampledData <- cbind(randomXvarOfA,B_shuffled_XY)
sampledData

   xvar longitude latitude
4   20    -87.79    40.04
7   12    -87.66    40.44
5   50    -87.88    40.10

Answer 1

I think the function you're looking for is the 'sample' function. It would work something like this (using your looping approach):

drawn_Sample <- sample(21:25000, 20000, rep=FALSE)
myresult <- integer(1000)    

for (i in seq(1:1000){
index_Values <- (1 + (i-1)*20):(20 + (i-1)*20))
myresult[i] <- myfun(my_Data$xvar[1:20], my_Data$longitude[drawn_Sample[index_Values]], my_Data$latitude[drawn_Sample[index_Values]])
}

In this case, I am randomly assigning rows 1:20 (the one's with value "A") to groups of twenty randomly chosen rows 21:25000 and then applying the function across the groupings.

This feels a bit needlessly complicated, and I think we could condense it all down if we knew a little more about your function ('myfun'). I'm assuming it's vectorized.

Update : At the OP's request, I am adding how to modify this answer to suit data frames that are not so easily sorted.

repetitions <- 1000 # Change this as necessary

A_data <- my_Data[my_Data$type=="A",]
B_data <- my_Data[my_Data$type=="B",]

A_rows <- nrow(A_data)
B_rows <- nrow(B_data)

drawn_Sample <- sample(1:B_rows, repetitions * A_rows, rep=FALSE)
myresult <- integer(repetitions)    

for (i in seq(1:repetitions){
index_Values <- (1 + (i-1)*A_rows):(A_rows + (i-1)*A_rows))
myresult[i] <- myfun(A_data$xvar, B_data$longitude[drawn_Sample[index_Values]], B_data$latitude[drawn_Sample[index_Values]])
}

Answer 2

Read in your data:

  df<- read.table( text="
      type    xvar    longitude    latitude
      A       20      -87.81        40.11
      A       12      -87.82        40.12
      A       50      -87.85        40.22
      B       24      -87.79        40.04
      B       30      -87.88        40.10
       B       12      -87.67        40.32
      B       66      -87.66        40.44", header = TRUE)

I was writing this without splitting and it looked so messy. So I decided just to split your data.frame .

    dd1 <- df[df$type == "B" ,]  # get all rows of just type A
    dd2 <- df[df$type == "A" ,]  # get all rows of just type B

    v   <- dd2[sample(nrow(dd2), 2), ] #sample two rows at random that are type A
    # if you want to sample 20 rows change the 2 to a 20

    cols <- c("longitude", "latitude")
    dd1[,cols][sample(nrow(dd1), 2), ] <- v[,cols] 
    #Add the random long/lat selected from type As into 2 random long/lat of B


# put the As and Bs back together
rbind(dd2,dd1)
#  type xvar longitude latitude
# 1    A   20    -87.81    40.11
# 2    A   12    -87.82    40.12
# 3    A   50    -87.85    40.22
# 4    B   24    -87.79    40.04
# 5    B   30    -87.85    40.22
# 6    B   12    -87.81    40.11
# 7    B   66    -87.66    40.44

As you can see rows 5 and 6 of B have new randomly selected lat and long values from A types. I did not change the xvar values though. I don't know if you want this. If you did want to change the xvars too then you would change cols to cols <- c("xvar","longitude", "latitude") .

Inside a function it would look like:

changestuff <-  function(x){

        dd1 <- x[x$type == "B" ,]  # get just A
        dd2 <- x[x$type == "A" ,]  # get just B
        v   <- dd2[sample(nrow(dd2), 2), ]
        cols <- c("longitude", "latitude")
        dd1[,cols][sample(nrow(dd1), 2), ] <- v[,cols] 
        rbind(dd2,dd1)
                            }

changestuff(df)