Gradients of Rolling View

rv_grad.py

import numpy as np
from skimage.util.shape import view_as_windows, view_as_blocks
from sthor.util import filter_pad2d


def f_g(x):

    xr = view_as_windows(x, (2, 2)).copy()
    #print xr
    loss = 0.5 * (xr**2.).sum()

    #xr[:] = 1
    xrT = xr.transpose((0, 2, 1, 3))
    #print xrT

    xrT2d = xrT.reshape(4, 4)
    #print xrT2d

    xrT2dp = filter_pad2d(np.atleast_3d(xrT2d), (3, 3))[..., 0]
    #print xrT2dp

    g = view_as_blocks(xrT2dp, (2, 2))
    g = g.sum(-1).sum(-1)

    return loss, g

x_shape = (3, 3)
x = np.arange(np.prod(x_shape)).reshape(x_shape).astype('f')

#print g


def v(i):
    o = np.zeros(x.size)
    o[i] = 1
    return o.reshape(x.shape)
#print x

e = 1e-9

gt = np.array([((f_g(x + e*v(i))[0] - f_g(x)[0]) / e) for i in xrange(x.size)])
gt = gt.reshape(x.shape)

gv = f_g(x)[1]
print gt
print gv

currently wrong (only working with 3x3 => 2x2)