Re: Matrix elements returned in wrong order

[Ralph Heinkel]

I'm not completely sure about this, it looks like that R has fortran-style arrays (left-most index changes fastest), while in Python we have c-style arrays ...
A solution would be to reshape the array in 'F'ortran order:

>>> arr = conn.r("matrix(1:12, nrow=4, ncol=3)")
>>> arr.reshape((4, 3), order='F')
array([[ 1, 5, 9],
       [ 2, 6, 10],
       [3, 7, 11],
       [4, 8, 12]])

Not sure yet whether this should be a default transformation in pyRserve ... any comments about this?

Ralph

Am Montag, 18. Juni 2012 22:14:58 UTC+2 schrieb Vebjorn Ljosa:

I think matrix elements are being returned in the wrong order.

In R, I get the following:

> matrix(1:12, nrow=4, ncol=3)
     [,1] [,2] [,3]
[1,] 1 5 9
[2,] 2 6 10
[3,] 3 7 11
[4,] 4 8 12

When transfering the matrix to Python and pyRserve, the values are filled in row-wise instead of column-wise:

$ python

Python 2.7.1 (r271:86832, Jul 31 2011, 19:30:53) 

[GCC 4.2.1 (Based on Apple Inc. build 5658) (LLVM build 2335.15.00)] on darwin

Type "help", "copyright", "credits" or "license" for more information.

>>> import pyRserve

>>> pyRserve.__version__

'0.5.2'

>>> conn = pyRserve.connect()

>>> print conn.r("matrix(1:12, nrow=4, ncol=3)")

[[ 1  2  3]

 [ 4  5  6]

 [ 7  8  9]

 [10 11 12]]

>>> 

Thanks,
Vebjorn

[Ralph Heinkel]

sorry, I just figured out that - for some strange reasons - the array must be 1-d before the reshape() method with the fortran order works. So the correct code is:

>>> arr = conn.r("matrix(1:12, nrow=4, ncol=3)")

>>> shape = arr.shape
>>> arr.shape = arr.size  # this makes it 1-d

>>> arr.reshape((4, 3), order='F')
array([[ 1, 5, 9],
       [ 2, 6, 10],
       [3, 7, 11],
       [4, 8, 12]])

cheers,

Ralph

[Ralph Heinkel]

Hi Kent,

thanks for your explanations of your problem. I'll look into this to see how a real clean solution would look like.

Ralph

Am Freitag, 4. Januar 2013 20:35:35 UTC+1 schrieb Kent Johnson:

A partial fix to the problem in my previous post seems to be to make RSerializer serialize arrays in Fortran order. This makes sense as the Rserve documentation at http://www.rforge.net/Rserve/faq.html#matrix says, "In general all arrays and matrices are one-dimensional vectors in R stored in column-major format with dimensions as attributes."

I modified rserializer.py at lines 193 and 212, adding the parameter "order='F'" to the calls to tostring(). This fixes my particular problem. It is only a partial fix because the array now is garbled when read back into Python.

Kent

[Ralph Heinkel]

Hi Kent,

please check my latest post about pyRserve V 0.7.0 and give it a try - it should fix your problems!

[Kent Johnson]

On Wednesday, January 9, 2013 2:41:10 AM UTC-5, Ralph Heinkel wrote:

Hi Kent,

please check my latest post about pyRserve V 0.7.0 and give it a try - it should fix your problems!

Yes, that is better, thanks!

Kent 

[Ralph Heinkel]

Hi Kent,

is "better" good enough? Or correct enough? For me it would be helpful if you could confirm that the arrays definitely behave as you would expect.

Thanks a lot,
Ralph

[Kent Johnson]

Sorry, I didn't mean to be vague. It now behaves the way I expect both on the case I posted to the list and for my real application.

Kent

[Ralph Heinkel]

Thanks Kent, that's great to hear.
If other users also confirm that their problems have been solved I'll
make an official package ready for download.

ciao ciao

Ralph