mgkuhn / npy.jl

Memory-mapped access to Python NYP format files

Read Python/NumPy NPY files via fast memory mapping

This package reads the header of a file in NumPy NPY format, and then uses memory mapping to return the actual data as an Array.

The advantage of memory mapping is that if the data already sits in the blockbuffer cache of the operating system, no data has to be copied in memory to load it. This can significantly improve repeat load times on large datasets.

Example usage:

using Npy
a = loadnpy("data.npy")

Limitations

This package only supports arrays consisting of basic number types (signed/unsigned 8/16/32/64-bit integers and 32/64-bit IEEE floating-point numbers), but not any “pickeled” Python objects.