I was comparing the time about two ways of an array of zeros using np.zeros_like and np.zeros.
I don't know if I'm missing something, but it seems that np.zeros_like is slower and, IMO, it seems there is no reason to be... Test has been made using an Jupyter Notebook, and is shown below:

import numpy as np; a=np.zeros(shape=(1000,1000))
%timeit -n100 np.zeros(shape=a.shape)

40.7 µs ± 5.72 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

import numpy as np; a=np.zeros(shape=(1000,1000))
%timeit -n100 np.zeros_like(a)

2.18 ms ± 47.1 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

Implementation of zeros_like is here: https://github.com/numpy/numpy/blob/master/numpy/core/numeric.py#L138

This issue has been opened to suggest changing the whole implementation with

def zeros_like(a, dtype=None, order='K', subok=True):
  if not subok:
    return np.zeros(shape=a.shape, dtype=dtype, order=order)
  else:
    res = empty_like(a, dtype=dtype, order=order, subok=subok)
    # needed instead of a 0 to get same result as zeros for for string dtypes
    z = zeros(1, dtype=res.dtype)
    multiarray.copyto(res, z, casting='unsafe')
    return res