Multidimensional data

The two general purpose multidimensional data classes are data_t and cdata_t. The former is for dynamically allocated data, while the latter is for data that has a known size at compile time. Both classes are implemented in the matpack namespace. They are both template classes that can define their rank and data type.

All operations available for view_t are available for data_t and cdata_t.

data_t

The data_t class is a dynamically allocated multidimensional array. It is a template of the data type and of the rank of the array. The rank of an array is the number of access dimensions it has. For example, a 2D array (e.g., a matrix) has a rank 2, and a 3D array has a rank 3.

The class holds a std::vector of the data type and a view_t describing its shape. If the rank of the data_t is 1, several of the std::vector operations, e.g., emplace_back, pop_back, etc., are available. Please see std::vector documentation for more information - and add missing operations as you need them. For all ranks, the data_t has the following additional member methods beyond those offered by view_t:

  • resize - change the size of the array.

  • flatten - shorthand for reshape({size()}).

  • reshape - change the shape of the array. This is only available for rvalue arrays.

Note

Copying and assigning to a data_t will, unlike for view_t, first call resize and then copy the data.

Some types made available via the python interface are:

Tip

The std::vector guarantees that the held data is contiguous in memory as long as the data type itself is contiguous. This can have large performance benefits for contiguous data, such as floating-point numbers. Beware that this guarantee is forfeit if the data type is not itself contiguous in memory, so a data_t<data_t<Numeric, 1>, 1>, for example, is not contiguous since std::vector hold pointers to its data.

cdata_t

The cdata_t class is a statically allocated multidimensional array. It is a template of the data type and of the shape of the array. For example, a 4x4 matrix would be defined as cdata_t<Numeric, 4, 4>, and a 3x3x3 tensor would be defined as cdata_t<Numeric, 3, 3, 3>.

The class holds a std::array of the data type. It operates similarly to data_t, but with a fixed size. The class supports all the same operations as data_t, but without the ability to resize or reshape the array. In addition, pure + and - operations are available for compatible arrays. To help with use of the array, the standard tuple-interface is also available, making it possible to write, for instance, auto& [a, b, c] = vec3;, for a cdata_t<Numeric, 3> vec3{};.

You will mostly find cdata_t used for fixed-size arrays solving specific problems in ARTS. Often, this means it is inherited from as a base class, so that specialized operations do not contaminate the general-purpose cdata_t.

Some types made available via the python interface are:

  • Stokvec - a 1D array of 4 Numeric, defined as cdata_t<Numeric, 4>. Holds spectral radiance in ARTS.

  • Propmat - a 1D array of 7 Numeric, defined as cdata_t<Numeric, 7>. Holds a sparsified propagation matrix in ARTS.

  • Muelmat - a 2D array of 4x4 Numeric, defined as cdata_t<Numeric, 4, 4>. Holds a Mueller matrix in ARTS.

  • Specmat - a 2D array of 4x4 Numeric, defined as cdata_t<Complex, 4, 4>. Holds a spectral matrix in ARTS.

  • Vector2 - a 1D array of 2 Numeric, defined as cdata_t<Numeric, 2>. Used for 2D vectors in ARTS. Not to be confused with Vector. An example is the line-of-sight vector (zenith, azimuth).

  • Vector3 - a 1D array of 3 Numeric, defined as cdata_t<Numeric, 3>. Used for 3D vectors in ARTS. Not to be confused with Vector. An example is the position vector (altitude, latitude, longitude).

  • …

Tip

The std::array guarantees that the held data is contiguous in memory. This can have large performance benefits. Unlike for data_t, a std::array is always contiguous, so a chain of cdata_t<cdata_t<Numeric, N>, M> is also contiguous and would basically be a MxN matrix in memory. This is a very useful property for fixed-size arrays, since it means that data_t<cdata_t<Numeric, N>, M> is also a contiguous array. This array-type has runtime size M and compile-time size N.

Relevant files

The relevant files for the data holding core matpack types are:

  • matpack/matpack_mdspan_data_t.h - the data_t class.

  • matpack/matpack_mdspan_cdata_t.cpp - the cdata_t class.