Fast shifting copy using NUMBA

NUMBA_shifting_fill.py

import numba


@numba.njit
def shifting_fill(src, dst):
    """Fill a 2D (or more multi-demensional) array-like object by shifting a 1D array, using NUMBA.
       This is useful when making an input to a RNN from time series data, that is, where
           src = [0, 2, 1, 3, 5]
       and the number of time steps fed in the RNN is M=3,
           dst = [
               [0, 2, 1],
               [2, 1, 3],
               [1, 3, 5]
           ]
       Note that dst's shape must be (L, M, ...), where L >= len(src) - M + 1

    Args:
        src (array): 1D ndarray
        dst (array): 2D ndarray
    """
    assert dst.ndim >= 2

    L, M = dst.shape
    assert L+M-1 <= len(src)

    for i in range(L):
        dst[i] = src[i:i+M]