How to serialize armadillo's vector

Question

How can I serialize arma::Col? Below are a MWE and the error output.

MWE:

#include 
#i         


        

Answer 1

The real crux of the issue here is that you want to add a serialize() member function to the various Armadillo objects, but that doesn't appear to be possible... except that thanks to a clever use of the preprocessor in Armadillo, it is!

Take a look at Mat_bones.hpp and Col_bones.hpp... you'll see something like this, inside of the class definitions of Mat and Col:

public:

#ifdef ARMA_EXTRA_COL_PROTO
  #include ARMA_INCFILE_WRAP(ARMA_EXTRA_COL_PROTO)
#endif

It made me very happy when I found this, because now I can do something like define a file called Mat_extra_bones.hpp:

//! Add a serialization operator.
template
void serialize(Archive& ar, const unsigned int version);

and then Mat_extra_meat.hpp:

// Add a serialization operator.
template
template
void Mat::serialize(Archive& ar, const unsigned int /* version */)
{
  using boost::serialization::make_nvp;
  using boost::serialization::make_array;

  const uword old_n_elem = n_elem;

  // This is accurate from Armadillo 3.6.0 onwards.
  // We can't use BOOST_SERIALIZATION_NVP() because of the access::rw() call.
  ar & make_nvp("n_rows", access::rw(n_rows));
  ar & make_nvp("n_cols", access::rw(n_cols));
  ar & make_nvp("n_elem", access::rw(n_elem));
  ar & make_nvp("vec_state", access::rw(vec_state));

  // mem_state will always be 0 on load, so we don't need to save it.
  if (Archive::is_loading::value)
  {
    // Don't free if local memory is being used.
    if (mem_state == 0 && mem != NULL && old_n_elem > arma_config::mat_prealloc)
    {
      memory::release(access::rw(mem));
    }

    access::rw(mem_state) = 0;

    // We also need to allocate the memory we're using.
    init_cold();
  }

  ar & make_array(access::rwp(mem), n_elem);
}

Then, in your program, all you need to do is

#define ARMA_EXTRA_MAT_PROTO mat_extra_bones.hpp
#define ARMA_EXTRA_MAT_MEAT mat_extra_meat.hpp

and the serialize() function will be a member of the Mat class. You can easily adapt this solution for other Armadillo types.

In fact this is exactly what the mlpack library (http://www.mlpack.org/) does, so if you are interested you can take a closer look at the exact solution I implemented there:

https://github.com/mlpack/mlpack/tree/master/src/mlpack/core/arma_extend