Assigning an multi-dimensional array from Java to R through Rserve

Question

I am using Java/R/Rserve for a project. I am facing the problem of transferring a multi-dimensional array from Java into R workspace for calculation. So far, the assign method of the RConnection object only allows the following to be passed: byte[], double[], int[], REXP, String,and String[].

I sidestepped this by creating a loop in Java, and passed the variables individually. Although this works, it looks ugly and inefficient.

RConnection c = new RConnection();
c.eval("x <- matrix(0,nrow=dimX[1],ncol=dimX[2])");
for (int i = 0; i < dimX[0]; i++){
  c.assign("i",Integer.toString(i+1));
  c.eval("i <- as.numeric(i)");
  for (int j = 0; j < dimX[1]; j++){
    c.assign("j",Integer.toString(j+1));
c.eval("j <- as.numeric(j)");
c.assign("tmp", Double.toString(XOBS[i][j]));
c.eval("x[i,j] <- as.numeric(tmp)");
  }             
}

The document for Rserve on http://www.rforge.net/Rserve/dist/JRclient/JavaDoc/org/rosuda/JRclient/REXP.html seems to be outdated, and the examples for Rserve are rather limited. Could anyone give me a suggestion on how to improve on this code?

Thank you

how big is your data? although, I am not sure about the possible precision loss, one way would be to dump your data into strings row by row and call eval with rbind in R, another way - is to dump all your Java data into the file and do read.table in R. — seninp
– seninp, Commented Oct 27, 2012 at 6:26

Gadget · Accepted Answer · 2014-04-13 16:46:13Z

I found one solution and just made it a little bit more friendly, link on source also attached.

Comments: it's ready-to-use utility method. It based on JRI, which now is a part of rJava.

Source:
http://www.lbgi.fr/wikili/index.php/JRI

    /**
     * Creates and assigns a matrix object in R from 2D table of double
     *
     * @param rEngine        the  R instance used
     * @param sourceArray    the 2D table of double
     *                       the matrix must have always the same column number on every row
     * @param nameToAssignOn the R object name
     * @return R matrix instance or null if R return an error
     */
    public static REXP assignAsRMatrix(Rengine rEngine, double[][] sourceArray, String nameToAssignOn) {
        if (sourceArray.length == 0) {
            return null;
        }

        rEngine.assign(nameToAssignOn, sourceArray[0]);
        REXP resultMatrix = rEngine.eval(nameToAssignOn + " <- matrix( " + nameToAssignOn + " ,nr=1)");
        for (int i = 1; i < sourceArray.length; i++) {
            rEngine.assign("temp", sourceArray[i]);
            resultMatrix = rEngine.eval(nameToAssignOn + " <- rbind(" + nameToAssignOn + ",matrix(temp,nr=1))");
        }

        return resultMatrix;
    }

veryltdbeard · Accepted Answer · 2014-10-23 10:20:57Z

2

For reference (method might not have been available yet at the time question was asked):

REXP REXP.createDoubleMatrix(double[][] arg);

answered Oct 23, 2014 at 10:20

veryltdbeard

4305 silver badges14 bronze badges

Comments

seninp · Accepted Answer · 2012-10-27 07:12:32Z

1

what if you do something like this (altering row and line numbers for your needs)?

RConnection c = new RConnection();

double[][] test = { { 1.0D, 2.0D }, { 3.0D, 4.0D } };

c.assign("res", test[0]);
for (int i = 1; i < 2; i++) {
  c.assign("tmp", test[i]);
  c.eval("res<-rbind(res,tmp)");
}

REXP x = c.eval("sum(res)");
System.out.println(x.asString());

this returns 10, as expected, but, however, this

String s = c.eval("rowSums(res)").asString();
System.out.println(s);

doesnt printout what it suppose, it just returns 3, maybe my Ubuntu-installed RServe is broken and can't print whatever is after space in result string 3 7:

> rowSums(d)
c1 c2 
3  7

and I cant find good examples too :(

edited Oct 27, 2012 at 7:12

answered Oct 27, 2012 at 7:04

seninp

7521 gold badge6 silver badges23 bronze badges

2 Comments

Linh B Ngo Over a year ago

This is a correct approach, but I did not want to follow it because I am wary of the potential run time/memory cost when using rbind. That's why I allocated the matrix in R at the beginning.

seninp Over a year ago

soory, i've posted a wrong timing in the comment, so cleaned that up. My simple wall-time benchmark code is here, seems like rbind is way faster than matrix. i don't know why.

Rich · Accepted Answer · 2014-05-15 00:49:15Z

0

You could:

flatten the array into a vector of integer rows, such that

a11 a12

a21 a22

=>

flat_array = new int[] {a11, a12, a21, a22}

Assign that to a local variable e.g:

rEngine.assign(".values", flat_array);
Call an R function that makes a matrix (or dataframe) in a global, like:

In R:

 make.matrix <- function(nrows, ncols, values) {

        value_mat <- matrix(values, nrow=nrows, ncol=ncols, byrow=TRUE)
        temp.res <<- res
        res
    }

In Java:

rEngine.eval("make.matrix(2,2,.values)");

Now you have the matrix in temp.res

answered May 15, 2014 at 0:49

Rich

98512 silver badges16 bronze badges

Collectives™ on Stack Overflow

Assigning an multi-dimensional array from Java to R through Rserve

4 Answers 4

Comments

Comments

2 Comments

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

4 Answers 4

Comments

Comments

2 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Related