What is this?

This is a single templated class for an N-dimensional array or "tensor". I have been using it in a probabilistic context (as a data structure for discrete joint distributions and discrete conditional distributions).

The implementation is quite straightforward and uses an internal vector as internal data structure. A "good implementation" would define probably some good external iterators over such data structure instead. To go from a "tabular" index to a "linear index" there are two functions defined:

• get_linear_index
• get_tabular_index

There are a bunch of convenience functions for 2D and 3D arrays.

One specific function is really useful in a probabilistic setting, namely:

value_container * get(const tabular_index & index, dimension_type summarize)

It "summarizes" the values of an nd-array in a particular dimension. For example, for a 2D array (a matrix) it returns values column-wise or row-wise. Note that it doesn't actually "sum" the values, it only returns a column or row.

Note that an nd-array can become really large. See the function size(). It is a multiplication of all the cardinalities of all involved dimensions.

Copyrights

• License: LGPLv3, MIT and/or Apache.