====== Feature Storage ======



===== Numpy =====

In case of python-only solutions a clustering by an attribute like the category provides an efficient way to storage and retrieve features. With masking rows can be selected.


===== Memory-Mapped Numpy =====

Memorry mapped numpy arrays are pretty fast if using a fast SSD. Otherwise, it's limited by i/o speed. Multhreading also works but is very limited by i/o speed limits.

keywords: sqlite numpy database postgresql