how to perform max/mean pooling on a 2d array using numpy

You could use scikit-image block_reduce:

import numpy as np
import skimage.measure

a = np.array([
      [  20,  200,   -5,   23],
      [ -13,  134,  119,  100],
      [ 120,   32,   49,   25],
      [-120,   12,    9,   23]
])
skimage.measure.block_reduce(a, (2,2), np.max)

Gives:

array([[200, 119],
       [120,  49]])