Pauli Virtanen wrote: > Thu, 22 Jul 2010 00:47:20 -0400, Robin Kraft wrote: > [clip] > >> Let's say the image looks like this: np.random.randint(0,2, >> 16).reshape(4,4) >> >> array([[0, 0, 0, 1], >> [0, 0, 1, 1], >> [1, 1, 0, 0], >> [0, 0, 0, 0]]) >> >> I want to use a square, non-overlapping moving window for resampling, so >> that I get a count of all the 1's in each 2x2 window. >> >> 0, 0, 0, 1 >> 0, 0, 1, 1 0 3 >> => 2 0 >> 1, 1, 0, 0 >> 0, 0, 0, 0 >> >> In another situation with similar data I'll need the average, or the >> maximum value, etc.. >> > > Non-overlapping windows can be done by reshaping: > > x = np.array([[0, 0, 0, 1, 1, 1], > [0, 0, 1, 1, 0, 0], > [1, 1, 0, 0, 1, 1], > [0, 0, 0, 0, 1, 1], > [1, 0, 1, 0, 1, 1], > [0, 0, 1, 0, 0, 0]]) > > y = x.reshape(3,2,3,2) > y2 = y.sum(axis=3).sum(axis=1) > > # -> array([[0, 3, 2], > # [2, 0, 4], > # [1, 2, 2]]) > > y2 = x.reshape(3,2,3,2).transpose(0,2,1,3).reshape(3,3,4).sum(axis=-1) > > # -> array([[0, 3, 2], > # [2, 0, 4], > # [1, 2, 2]]) > > > The above requires no copying of data, and should be relatively fast.
Actually, because of the use of reshape(3,3,4), your second example does make a copy. Warren > If > you need overlapping windows, those can be emulated with strides: > > http://mentat.za.net/numpy/scipy2009/stefanv_numpy_advanced.pdf > http://conference.scipy.org/scipy2010/slides/tutorials > /stefan_vd_walt_numpy_advanced.pdf > > _______________________________________________ NumPy-Discussion mailing list NumPy-Discussion@scipy.org http://mail.scipy.org/mailman/listinfo/numpy-discussion
