[Python] np.concatenate extremely slow for slightly larger 2D ndarray while small slices run...

I encountered extremely slow performance when calling np.concatenate. The minimal reproducible code is as follows:

a = np.zeros((1200, 500), dtype=np.float64)

# Runs instantly, under 1 second
b = np.concatenate(a[:1000])

# Extremely slow, takes ~10 minutes
c = np.concatenate(a)


np.concatenate(a) (just slightly larger) is extremely slow, even though memory usage is small.

I know there are some faster alternatives, but I'm curious why this behavior occurs.

Is this extremely slow behavior expected for np.concatenate? Why?

Continue reading...