v0.5.4

• Bug fix for SparseArrayRef::iterator
• Fix getting integral vectors with mlstream
• Documentation fixes

v0.5.3

• Bug fix: in some cases, the row index was not correctly computed when using SparseArrayRef::iterator

v0.5.2

• Bug fixes
• Support for ByteArray type specification

v0.5.1

• GenericImageRef split into GenericImageRef / GenericImage3DRef
• Bug fixes
• Expanded documentation and examples

v0.5

Version 0.5

• Expanded SparseArray support, added SparseMatrixRef.
• Added experimental RawArray support.
• Added experimental Image and Image3D support.
• Uniform type specifications through LType, e.g. LType[List, Real] or LType[SparseArray, Integer, 2]. Added NormalizeTemplate which converts to this form. The usual LibraryLink compatible type specifications also work.
• More functions for creating Tensors.
• Class instances can now be retrieved by ID; access to the class-collection.
• Greatly expanded documentation with many examples.
• Several other small improvements.

Incompatible changes:

• A compiler with C++11 support is now required. GCC 4.8 and MSVC 2015 both work.
• CompileTempate automatically selects C++11 using a compiler flag. If you need a newer standard, such as C++14, you must now select this explicitly using a flag similar to -std=c++14, even if C++14 is the default for your compiler.
• CubeRef is now indexed as (slices, rows, columns) for consistency with Image. In practice, this means that operator () has not changed, but slices(), rows() and cols() now return the tensor size along the 1st, 2nd and 3rd index, respectively.