[Rcpp-devel] Matrix columns

| I?ve been trying the following, with the main part ripped from rcpp-quickref:
| 
|     SEXP dfr2Mat(SEXP dfr)

A much easier way is to, say, call as.matrix() in R and then assign to an
NumericMatrix object in C++.  

Another way is turn the Data.Frame into a list via as.list() as just pick off
column after column.

|     {
|         DataFrame df = dfr;
|         int* dm = INTEGER( ::Rf_getAttrib( df, R_DimSymbol ) ) ;
|         int rows = dm[0];
|         int cols = dm[1];
|         NumericMatrix retMat (rows, cols);
|         for(int i = 0; i < cols; i++)
|         {
|             NumericVector curcol = df(i);
|             NumericMatrix::Column cl = retMat.( _, i);

Why is retMat on the right-hand side if you try to fill it?  Makes no real sense.

|             cl = curcol;
|         }
|         return retMat;
|     }

I can see no straightforward way to write given how objects are organized
internally.  Once a Rcpp::DataFrame object is instantiated, you can call
size() which gives you the number of columns.  So here is a really
complicated way to do this, but let me reiterate that I think you are
starting from the wrong starting point: if your data.frame really is a
matrix, just use a matrix:

R> require(inline)
Loading required package: inline
R> 
R> df2mat <- cxxfunction(signature(Dsexp="ANY"), plugin="Rcpp", body='
+    // construct the data.frame object
+    Rcpp::DataFrame DF = Rcpp::DataFrame(Dsexp);
+ 
+    // we get ncol() from DF
+    int k = DF.size();
+ 
+    // but for nrow() we need to assign first; one way is:
+    Rcpp::NumericVector V = DF[0];
+    int n = V.size();
+ 
+    Rcpp::NumericMatrix M(n,k);
+ 
+    for (int i=0; i<k; i++) {
+        V = DF[i];
+        M(_,i) = V;  // one way to assign using sugar operator _
+    }
+ 
+    return M;
+ ')
R> M <- df2mat(data.frame(a=1:5, b=seq(1.1,5.5,by=1.1)))
R> M
     [,1] [,2]
[1,]    1  1.1
[2,]    2  2.2
[3,]    3  3.3
[4,]    4  4.4
[5,]    5  5.5
R> class(M)
[1] "matrix"
R> 


Hope this helps,  Dirk

Gauss once played himself in a zero-sum game and won $50.
                      -- #11 at http://www.gaussfacts.com

Hi Dirk. Thanks for your time and trouble.
I agree there are simpler ways to achieve this functionality, but if I don't
get around this, I won't be able to produce the thing I'm really after.

Why is retMat on the right-hand side if you try to fill it?  Makes no real

sense.

Nick,

| Hi Dirk. Thanks for your time and trouble.
| I agree there are simpler ways to achieve this functionality, but if I don't
| get around this, I won't be able to produce the thing I'm really after.

I fear that you think that because data.frame is a preferred structure in R,
it also must be in C++. That is not the case, and the reason is the internal
representation of a data.frame at the C (and hence C++) level.  You may be
better off decomposing your data.frame into columns or different types (in R
or C++) or just use a matrix if there is a common type.
 
| > Why is retMat on the right-hand side if you try to fill it?  Makes no real
| sense.
| 
| >From rcpp-quickref:
| // Reference the second column
| // Changes propagate to xx (same applies for Row)
| NumericMatrix::Column zzcol = xx( _, 1);
| zzcol = zzcol * 2;
| 
| Only difference in my code was that I was assigning from a NumericVector
| instead of from an altered version of the original NumericMatrix::Column. Or
| am I mistaken here? I've put it here again:
| NumericVector curcol = df(i);
| NumericMatrix::Column cl = retMat( _, i);
| cl = curcol;//from rcpp-quickref, I understand this alters the matrix?

I believe you to be mistaken. Twice assigning overwrites. Feel free to debug
and test to convince yourself.
 
| Then you appear to suggest that my approach using the "dim" attribute is no
| good, or at least you get the dimensions in a very different way (by using
| the data.frame as a list). I agree your way works, but why shouldn't mine?
| In any case, this sort of thing would be a valuable addition to Rcpp in my
| opinion.
| 
| Finally, you use:
| >+        V = DF[i];
| >+        M(_,i) = V;  // one way to assign using sugar operator _
| Which makes me wonder:
| * why DF[i] and not DF(i) (what's the difference - I still have not found a
| clear explanation)

It's only stylistic. You can use either one.

| * is M(_,i) = V; any different from NumericVector cl = M(_,i);cl=V; ? If so,
| how so?

As I recall, the former ("one-step") created a compiler error as the compiler
could not disambiguate some intermediate type. The second ("two-step")
worked, so I used that,  So the difference is that one works :)   [ In
theory, both should. Patches to make the former work are welcome. ]

| * thanks for letting me know _ is part of sugar. I may be able to figure out
| its use now, but isn't this worthy of a mention in rcpp-sugar.pdf?

As I said a few days ago, patches which enhance documentation or code are
always welcome and will always be properly credited (that is, if we agree
with them and incorporate them).

Dirk

Gauss once played himself in a zero-sum game and won $50.
                      -- #11 at http://www.gaussfacts.com

Dirk,

Thanks again!
I thought I ought to let you know: the main problem was that my compilation
still pointed to older versions of R and Rcpp that did not contain some of
the additions in case. Goes to show!

But still some comments/questions below...

-----Original Message-----
From: Dirk Eddelbuettel [mailto:edd at debian.org]
Sent: zondag 15 mei 2011 21:22
To: Nick Sabbe
Cc: 'Dirk Eddelbuettel'; rcpp-devel at r-forge.wu-wien.ac.at
Subject: RE: [Rcpp-devel] Matrix columns


Nick,

I fear that you think that because data.frame is a preferred structure
in R,
it also must be in C++. That is not the case, and the reason is the
internal
representation of a data.frame at the C (and hence C++) level.  You may
be
better off decomposing your data.frame into columns or different types
(in R
or C++) or just use a matrix if there is a common type.

What I'm actually hoping to achieve is an improvement to the answer to my
question in
http://stackoverflow.com/questions/5980240/performance-of-rbind-data-frame
(though in doubt that I will improve on it). Apart from that, I'm just
trying to learn Rcpp.

| > Why is retMat on the right-hand side if you try to fill it?  Makes
no real
| sense.
|
| >From rcpp-quickref:
| // Reference the second column
| // Changes propagate to xx (same applies for Row)
| NumericMatrix::Column zzcol = xx( _, 1);
| zzcol = zzcol * 2;
|
| Only difference in my code was that I was assigning from a
NumericVector
| instead of from an altered version of the original
NumericMatrix::Column. Or
| am I mistaken here? I've put it here again:
| NumericVector curcol = df(i);
| NumericMatrix::Column cl = retMat( _, i);
| cl = curcol;//from rcpp-quickref, I understand this alters the
matrix?

I believe you to be mistaken. Twice assigning overwrites. Feel free to
debug
and test to convince yourself.

Then what's with the comment in rcpp-quickref (// Changes propagate to xx
(same applies for Row))?
As far as I recall (that old effective C++ has gathered quite a bit of dust
by now), the first = is not an assignment but initialization. So I only
assign once (in the last lign). But I'll debug and test to convince myself.

| * is M(_,i) = V; any different from NumericVector cl = M(_,i);cl=V; ?
If so,
| how so?

As I recall, the former ("one-step") created a compiler error as the
compiler
could not disambiguate some intermediate type. The second ("two-step")
worked, so I used that,  So the difference is that one works :)   [ In
theory, both should. Patches to make the former work are welcome. ]

Truly lol.

| * thanks for letting me know _ is part of sugar. I may be able to
figure out
| its use now, but isn't this worthy of a mention in rcpp-sugar.pdf?

As I said a few days ago, patches which enhance documentation or code
are
always welcome and will always be properly credited (that is, if we
agree
with them and incorporate them).

Intriguingly, I could not find any information on how to get my suggested
"patches" to you, besides a note on your site to "post them on the mailing
list" (admittedly I could have looked better). Do I create a LaTeX file with
additions to the helpfiles and send it here? Or is there a more formal
(read: less annoying for you guys who would then still have to edit it until
it fits the rest of the docs) way? I _do_ intend to try and improve the
documentation, though I don't feel worthy to improve upon the
implementation.

Thanks again!


Nick Sabbe
--
ping: nick.sabbe at ugent.be
link: http://biomath.ugent.be
wink: A1.056, Coupure Links 653, 9000 Gent
ring: 09/264.59.36

-- Do Not Disapprove