如何提高在大型三维网格上存储值的速度

Answer 1

可以删除for循环并使用numpy操作来处理它。与您的代码相同，没有for循环和索引，速度快约60倍：

def ver_2(xCoord, yCoord, zCoord, xSpacing, ySpacing, zSpacing):
    Grid = numpy.empty([len(xCoord), 8, 4])
    Grid[:, 0, 0] = xCoord
    Grid[:, 0, 1] = yCoord
    Grid[:, 0, 2] = zCoord
    Grid[:, 0, 3] = 0
    #
    vX = floorPartial(xCoord, xSpacing)
    vY = floorPartial(yCoord, ySpacing)
    vZ = floorPartial(zCoord, zSpacing)
    #
    d1 = EucDistSq(xCoord, yCoord, zCoord, vX, vY, vZ)
    d2 = EucDistSq(xCoord, yCoord, zCoord, vX+xSpacing, vY, vZ)
    d3 = EucDistSq(xCoord, yCoord, zCoord, vX-xSpacing, vY, vZ)
    d4 = EucDistSq(xCoord, yCoord, zCoord, vX, vY+ySpacing, vZ)
    d5 = EucDistSq(xCoord, yCoord, zCoord, vX, vY-ySpacing, vZ)
    d6 = EucDistSq(xCoord, yCoord, zCoord, vX, vY, vZ+zSpacing)
    d7 = EucDistSq(xCoord, yCoord, zCoord, vX, vY, vZ-zSpacing)
    #
    dt = numpy.sum([d1, d2, d3, d4, d5, d6, d7], axis=0)
    # VoxelX,VoxelY,VoxelZ,intensity
    Grid[:, 1] = numpy.stack((vX, vY, vZ, d1/dt), axis=-1)
    Grid[:, 2] = numpy.stack((vX+xSpacing, vY, vZ, d2/dt), axis=-1)
    Grid[:, 3] = numpy.stack((vX-xSpacing, vY, vZ, d3/dt), axis=-1)
    Grid[:, 4] = numpy.stack((vX, vY+ySpacing, vZ, d4/dt), axis=-1)
    Grid[:, 5] = numpy.stack((vX, vY-ySpacing, vZ, d5/dt), axis=-1)
    Grid[:, 6] = numpy.stack((vX, vY, vZ+zSpacing, d6/dt), axis=-1)
    Grid[:, 7] = numpy.stack((vX, vY, vZ-zSpacing, d7/dt), axis=-1)
    return Grid

我想更多的优化是可能的，有更多的面向矩阵的计算，但我没有准确地掌握代码要做什么：-/可能在计算距离之前设置网格的[：，1-7，0-2]值，而仅仅使用网格值进行距离计算可以减少一些时间，因为省略了一些不需要的分配。你知道吗