问题
I am using griddata to resample a numpy 2 dimensional array on a grid.
z.shape = (1000, 1000)
x, y = np.arange(-5, 5, 0.01), np.arange(-5, 5, 0.01)
newx, newy = np.arange(-2, 2, 0.1), np.arange(-2, 2, 0.1)
griddata((x, y), z, (newx[None, :], newy[:, None]))
The code should:
- resample z (which represents an image) to a new coarser or finer grid
- the new grid does not necessarily cover all of the original one.
However griddata cannot manage a regular input grid. Does anyone know an easy alternative?
回答1:
Use any of the methods suitable for data on a grid listed in the documentation: https://docs.scipy.org/doc/scipy/reference/interpolate.html#multivariate-interpolation
That is:
https://docs.scipy.org/doc/scipy/reference/generated/scipy.interpolate.RegularGridInterpolator.html
or https://docs.scipy.org/doc/scipy/reference/generated/scipy.interpolate.RectBivariateSpline.html
or https://docs.scipy.org/doc/scipy/reference/generated/scipy.ndimage.interpolation.map_coordinates.html
Note also that you are using griddata incorrectly. Your code corresponds to interpolating from a line defined by your 1000 (x, y) coordinates, where each point has 1000 values associated with it. However, interpolation to 2D from a 1D line is badly defined, and the failure results from trying to triangulate a set of points that are along a line.
You should do
import numpy as np
from scipy.interpolate import griddata
z = np.random.rand(100, 100)
z.shape = (100, 100)
x, y = np.arange(-5, 5, 0.1), np.arange(-5, 5, 0.1)
xx, yy = np.meshgrid(x, y, indexing='ij')
newx, newy = np.arange(-2, 2, 0.1), np.arange(-2, 2, 0.1)
griddata((xx.ravel(), yy.ravel()), z.ravel(), (newx[None, :], newy[:, None]))
This will work correctly --- however, 1000x1000 = 1000000 points in 2D is simply way too much data for triangulation-based unstructured interpolation (needs large amounts of memory for the triangulation + it's slow), so you should use the gridded data algorithms.
来源：https://stackoverflow.com/questions/26276550/alternatives-to-python-griddata