xnd-project / libndtypes

Subsumed into xnd
https://xnd.io/
BSD 3-Clause "New" or "Revised" License
25 stars 17 forks source link

Non-homogeneous/hidden dimensions? #46

Open anton-malakhov opened 6 years ago

anton-malakhov commented 6 years ago

For the sake of hardware efficiency, we might want to store arrays in blocking formats, e.g. (i, j) --> (i, j, j4, i4). But from the point of view of user-experience, it's still good to provide look'n'feel of (i, j). Numpy does not allow creating a low-dimensional view into higher-dimensional array but it could be possible if it'd store internal dimensions for sake of all the C extensions, which would know the real strides while presenting 'interface' dimensions to user. Would this kind of trick possible with this libndtypes? Thanks!