Re: [eigen] Feature suggestion: interior row and column ranges [patch attached]

[Benoit Jacob <jacob.benoit.1@xxxxxxxxx>]

How about middleRows(first, count)?

by analogy with topRows(count) and bottomRows(count).

Benoit

2010/6/26 Gael Guennebaud <gael.guennebaud@xxxxxxxxx>:
> hi,
>
> thanks for the patch.
>
> Let's discuss function names: to be consistent with
> bottomRows/topRows/leftCols/rightCols, they should be: rows(int,int)
> and cols(int,int), but then there is a name clash with rows() and
> cols() returning the matrix sizes.  In the sparse module I used
> subRows, and subCols for the same feature because I wanted names that
> finish by "rows" and "cols", but they are not very good names! So yes
> why not colRange and rowRange. I'm not 100% convinced, but since I
> have nothing better to propose...
>
> gael
>
> On Sat, Jun 26, 2010 at 11:52 PM, Manoj Rajagopalan <rmanoj@xxxxxxxxx> wrote:
>>
>>
>>
>> On Saturday 26 June 2010 05:32:31 pm FMDSPAM wrote:
>>>   Am 26.06.2010 22:58, schrieb Manoj Rajagopalan:
>>> > Hi,
>>> >
>>> >    I've noticed leftCols(), rightCols(), topRows() and bottomRows() in
>>> > the DenseBase API. I have, on occasion felt the need to access a
>>> > contiguous range of rows/columns that start and end in the interior - the
>>> > above functions allow access only to the extremes. Could DenseBase member
>>> > functions, possibly named colRange(Index start, Index size) and
>>> > rowRange(Index start, Index size), be supported sometime in the future?
>>> >
>>> > Thanks,
>>> > Manoj
>>>
>>> It's already in upcomming eigen 3.0:
>>> Use dynamic sized block: mat1.block(i,j,rows,cols)
>>> or fixed sized block: mat1.block<rows,cols>(i,j)
>>> to have Read-write access to sub-matrices.
>>>
>>> see devel-doc:Tutorial Matrix Blocks
>>> <http://eigen.tuxfamily.org/dox-devel/TutorialCore.html#TutorialCoreMatrixB
>>>locks>
>>>
>>> cheers
>>> Frank
>>
>>
>>
>> Attaching a patch implementing my suggestion for the consideration of the
>> developers/maintainers.
>>
>> @Frank: I wished the API would let me choose, say 3 columns starting with the
>> 4th, in a 10x10 matrix. This should be a 10x3 submatrix starting at column #3
>> (0-base-indexing). With leftCols() (or rightCols()), the starting (or ending)
>> column is fixed to an extreme and there is currently no API method to extract
>> a few contiguous columns or rows purely in the interior.
>>
>> DenseBase::block() is the master method that allows the access of any
>> contiguous 2D sub-region and is used to implement leftCol() and rightCol()
>> but the latter are much more readable functions. So I present
>> DenseBase::rowRange() and DenseBase::colRange() which use DenseBase::block()
>> to extract contiguous interior rows and columns respectively.
>>
>> Example of rowRange usage:
>>
>>        int const N = 10;
>>        MatrixXi A(N,N);
>>        A.setRandom();
>>        cout << "A =\n" << A << '\n' << endl;
>>        cout << "A(3..5,:) =\n" << A.rowRange(3,3) << endl;
>>        // alternatively, since #cols is known at compile time
>>        cout << "A(3..5,:) =\n" << A.rowRange<3>(3) << endl;
>>
>> Output:
>>
>> A =
>>  7   0 -10   0   1   1  -5   1   4  10
>>  -2   3  -5   7   6  -9  10 -10  -7   2
>>  6  -3  -8   2  -2  -6   8  -1  -1   3
>>  6   0   6  -4   8   3   7   9   8   8
>>  9   9  -7   3  -5   8  -5   9   7  -1
>>  -6   9  -2   1  -3  -3   1   5  -4   9
>>  -3   3  -8   0   6  -9  -3  -5  -6  -2
>>  6   5  -8  10   9 -10   5   5   8   7
>>  -5  -8  10  -4  -9  -1   0   3  -3   4
>>  1   2  -6   6   9  -9   4  -3   4   9
>>
>> A(3..5,:) =
>>  0   1   1
>>  7   6  -9
>>  2  -2  -6
>>  -4   8   3
>>  3  -5   8
>>  1  -3  -3
>>  0   6  -9
>>  10   9 -10
>>  -4  -9  -1
>>  6   9  -9
>>
>>
>> I have written small test programs like the above and have tested static and
>> dynamic versions of rowRange() and colRange().
>>
>> Thanks,
>> Manoj
>>
>
>
>