scipy griddata 插值返回一个用 nan 填充的向量

问题描述

我有一个 3d 点列表，我想在 3d 网格上插入这些点.

I have a list of points on in 3d that I would like to interpolate on a 3d grid.

coords = array([[  8.33399963,  12.94800186,  15.22500038],
       [  8.57299995,  13.90000153,  14.14700031],...)

我有网格 x,y,z 坐标，它们与 numpy.meshgrid 一起用于创建网格:

I have the grid x,y,z coordinates, which together with numpy.meshgrid are used to create the grid:

xi,yi,zi = np.meshgrid(bbox[:,0],bbox[:,1],bbox[:,2])

然后当我尝试执行插值时:

and then when I try to perform the interpolation:

griddata(coords,np.random.choice([.1,1,2],size=len(coords)),(xi,yi,zi),method='linear')

我得到了一个 nans 向量:

I get a vector of nans:

array([[[ nan,  nan,  nan,  nan,  nan,  nan,  nan,  nan],
        [ nan,  nan,  nan,  nan,  nan,  nan,  nan,  nan],
        [ nan,  nan,  nan,  nan,  nan,  nan,  nan,  nan],
        [ nan,  nan,  nan,  nan,  nan,  nan,  nan,  nan],
        [ nan,  nan,  nan,  nan,  nan,  nan,  nan,  nan],
        [ nan,  nan,  nan,  nan,  nan,  nan,  nan,  nan],
        [ nan,  nan,  nan,  nan,  nan,  nan,  nan,  nan],
        [ nan,  nan,  nan,  nan,  nan,  nan,  nan,  nan]],....

我在这里做错了什么?

推荐答案

我不知道 xi,yi,zi 值是什么，但很可能它们在 coords 定义的域之外.如果使用meshgrid生成网格，那么注意数组的顺序:

I do not know what xi,yi,zi values are but most likely they are outside the domain defined by coords. If you use meshgrid to generate the grid, then pay attention to the order of the arrays:

在输入长度为 M、N 和 P 的 3-D 情况下，输出的形状为 (N,M, P) 用于 'xy' 索引和 (M, N, P) 用于 'ij' 索引.>

试试这个:

In [61]: coords = 20 * np.random.random((200, 3)) - 1

In [62]: xi, yi, zi = np.meshgrid(np.arange(coords[:, 0].min()+2, coords[:,0].max()-2), np.arange(coords[:, 1].min()+2, coords[:,1
    ...: ].max()-2), np.arange(coords[:, 2].min()+2, coords[:,2].max()-2), indexing='ij')

In [63]: griddata(coords,np.random.choice([.1,1,2],size=len(coords)),(xi.astype(np.float), yi.astype(np.float), zi.astype(np.float
    ...: )),method='linear')

您仍然会得到一些 nan 值，其中点对函数的采样很差，但大多数值都已定义.

You will still get some nan values where points poorly sample the function but most values are defined.

另一种可能性是您只看到第一个平面"，其中可能主要包含 nan.尝试 np.sum(np.isfinite(g)) 以查看所有点中的点如何有效 np.prod(g.shape) where g 是 griddata() 的输出.

Another possibility is that you are just seeing the first "plane" which may contain mostly nan. Try np.sum(np.isfinite(g)) to see how may points are valid out of all points np.prod(g.shape) where g is thr output from griddata().

这篇关于scipy griddata 插值返回一个用 nan 填充的向量的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持！