We have 3 scalar type representations: tensorflow, ctypes, and string names. Not unlikely we may get a fourth, numpy. This patch factors out all the associations into a central place. Modules can build their own maps as needed from it (i.e. name->ctype or dtype->name, etc)
We have 3 scalar type representations: tensorflow, ctypes, and string names. Not unlikely we may get a fourth, numpy. This patch factors out all the associations into a central place. Modules can build their own maps as needed from it (i.e. name->ctype or dtype->name, etc)