Suppose I have the following loop:
struct Obj {
RowVectorXf payload;
Obj(RowvectorXf payload) : payload(std::move(payload)) {};
}
MatrixXd data = ...
std::vector<Obj> payloads;
for (int i = 0; i < n; i++) {
payloads.emplace_back(data.row(i)(seq(1,10,2));
}
This is a simplification of my use case, which absolutely does need to step out into the std:: space with a struct containing a slice for sanity and ease of coding. Anyhow, I have started using the unaryExpr's with abandon, and I am curious if there is something like a rowExpr that could perform this function without the for loop (and thus eliminating the visual noise).
// Goal, in pseudo code:
data.rowExpr([](RowVectorXd row) payloads.emplace_back(row(seq(1,10,2));
With the development branch of Eigen (and the upcoming 3.4 version) you can access the elements of Eigen objects using std-compatible iterators. To iterate row-wise through an Eigen expression, you need to write data.rowwise().begin() and data.rowwise().end() -- (to iterate column-wise, you need to write .colwise().begin() of course). And you can directly pass these to the constructor of std::vector (if the types are compatible).
Since in your example you also need to convert between double and float you can write something like this:
auto const& data_ = data.cast<float>().rowwise();
std::vector<Obj> payloads(data_.begin(), data_.end());
Full working example: https://godbolt.org/z/2x2jFJ