Row selection by function

[Daniel Baker]

Hi,

‌

I’ve found the rows/column selection utility quite helpful, though I thought it would be easier if we could return rows satisfying a function rather than by indexes only. (This could make the selection method a little more ergonomic.)

‌

I wrote the following as adapters which produce the desired results, but I thought I’d share the idea/an example implementation in case you wanted to incorporate them into a later blaze release.

‌

template<typename F, typename IT=std::uint32_t, size_t N=16>
auto indices_if(const F &func, size_t n) {
    blaze::SmallArray<IT, N> ret;
    for(IT i = 0; i < n; ++i) if(func(i)) ret.pushBack(i);
    return ret;
}
template<typename M, typename F, typename IT=std::uint32_t>
auto rows_if(const M &mat, const F &func) {
    auto wrapfunc = [&](auto x) -> bool {return func(row(mat, x, blaze::unchecked));};
      return rows(mat,
                 indices_if<decltype(wrapfunc),IT>(wrapfunc, // predicate
                                                   mat.rows())); // nrow
}
template<typename M, typename F, typename IT=std::uint32_t>
auto columns_if(const M &mat, const F &func) {
auto wrapfunc = [&](auto x) -> bool {return func(column(mat, x, blaze::unchecked));};
    return columns(mat, // matrix
                   indices_if<decltype(wrapfunc), IT>(wrapfunc, // predicate
                                                      mat.columns())); // ncol
}

‌

Thanks for your time!

‌

[Klaus Iglberger]

Hi Daniel!

Thanks a lot for your proposal! I admit that I feel a little proud that Blaze already enables you to implement your idea of a row selection in just a few lines of code. I like the idea and will definitely consider this as a nice extension for one of the next releases of Blaze.

Have you considered that issue #285 might provide a different point of view on your problem? The given predicate is nothing but a reduction operation that reduces a row or column to a bool. The resulting vector of bool could be used to produce the necessary indices.

Thanks again,

Best regards,

Klaus!