Feature requests

[Ray Kim]

1. Add matrix reduce (sum) operation: As far as I know, there is no easy way to get a total sum of a matrix. I'm using std::accumulate on a submatrix right now. Hope there is a standard and optimized way of summing a matrix

2. const matrix views: Pulling out a vector from a row or column from a matrix with const is hard. blaze::column and blaze::row are not allowed on a const pretext. Hope I could directly pull out a vector from a matrix, or introduce const column/row views so I can assign them to a vector

3. matrix equality compare: Armadillo have a really great matrix equality compare operator with a tolerance argument. If the difference of elements in each operands is within provided tolerance, the operator returns a boolean true. Hope we could have something like that!

Thanks for this great library!

[Klaus Iglberger]

Hi Ray!

Welcome to Blaze and thanks for the proposals. Feedback about the library in general and missing features is highly appreciated.

Allow me to respond to the feature requests individually.

---

Add matrix (sum) operation

This feature is indeed missing and we will tackle it in one of the next releases (see issue #4). However, unfortunately this is a major task and therefore will take some time to implement. But you can rest assured that it is very high on our priority list.

---

const matrix views

The following code snippet shows how to extract a row from a constant matrix:

// Creating a 'const'-qualified row-major dense 3x3 matrix
const blaze::DynamicMatrix<int> A{ { 1, 2, 3 }, { 4, 5, 6 }, { 7, 8, 9 } };

// Extracting the 1st row into a row vector
blaze::DynamicVector<int,rowVector> x;
x = row( A, 1UL );
std::cerr << " x = " << x << "\n";

// Extracting the 2nd row into a 'const'-qualified column vector
const blaze::DynamicVector<int,columnVector> y( trans( row( A, 2UL ) );
std::cerr << " y = " << y << "\n";

This example will print

x = { 4, 5, 6 }
y = { 7, 8, 9 }

Please note that x is a row vector. Blaze distinguishes between row and column vectors. The row of a matrix is considered a row vector, the column of a matrix is considered a column vector.

The type returned from the row() function in this case is Row< const DynamicMatrix<int> > (note the const on the template argument). However, even if you have a non-const-qualified matrix you can at any point decide that a row should be immutable by using Row< const DynamicMatrix<int> > or const Row< DynamicMatrix<int> > explicitly.

If you want to assign a row to a column vector (such as y), you can do this via the trans() function.

We hope this helps. If you still encounter any problem, please post an according minimum example.

---

matrix equality compare

Blaze currently doesn't provide a matrix comparison with an additional tolerance value. However, this sounds like a reasonable feature. We will consider adding this in one of the next releases. Thanks for the proposal!

---

Best regards,

Klaus!

[Ray Kim]

Seems like I mixed up types on extracting the vector part. Thanks for your reply!

Great regards from Korea

[Marcin Copik]

An addition of reduction operation would allow for a quick implementation of many matrix norms. This feature is very useful for correctness verification of results and Blaze is missing both vector and matrix norms.

[pkerichang]

Sorry to bump old issue, but is the matrix/vector equality comparison with custom tolerance implemented yet? If not I’m happen to submit a PR.

[Klaus Iglberger]

Hi @pkerichang !

No, the feature hasn’t been implemented yet. We of course don’t mind if you provide a pull request. We would expect further overloads of the equal() functions in the following files:

• <blaze/math/expressions/DVecDVecEqualExpr.h>
• <blaze/math/expressions/DVecSVecEqualExpr.h>
• <blaze/math/expressions/SVecSVecEqualExpr.h>
• <blaze/math/expressions/DMatDMatEqualExpr.h>
• <blaze/math/expressions/DMatSMatEqualExpr.h>
• <blaze/math/expressions/SMatSMatEqualExpr.h>

Based on these new functions a lot of the comparison functionality used within Blaze could be refactored. However, the pull request should be limited to the new functions. Thanks,

Best regards,

Klaus!