scikit-learn 的 BallTree 的类型错误

Question

我有这个玩具示例 scipy 的 cKDTree which works very well and I want to do a similar piece of code with scikit-learn'sBallTree

import numpy as np
from scipy import spatial

min_neighbors = 3
x,y = np.mgrid[0:5,0:5]

 grid_x,grid_y = np.mgrid[1:6,1:6]
 points = np.c_[x.ravel(),y.ravel()]

 grid_points = np.c_[grid_x.ravel(),grid_y.ravel()]
 tree = spatial.cKDTree(points)
 indices = tree.query_ball_point(grid_points,r=1)
 for idx,(matches,grid) in enumerate(zip(indices,grid_points)):
     if len(matches) >= min_neighbors:
         x1,y1 = tree.data[matches].T

当我用 BallTree 做类似的玩具示例时，如下所示

import numpy as np

from sklearn.neighbors.ball_tree import BallTree
from sklearn.neighbors import NearestNeighbors

import sys

def main():
    min_neighbors = 3
    x,y = np.mgrid[0:5,0:5]  
    grid_x,grid_y = np.mgrid[1:6,1:6]
    points = np.c_[x.ravel(),y.ravel()]
    grid_points = np.c_[grid_x.ravel(),grid_y.ravel()]
    bt = BallTree(points,leaf_size=1, metric='haversine')
    indices = bt.query_radius(grid_points,1)

    for idx,(matches,grid) in enumerate(zip(indices,grid_points)):
        #print(matches)
        if len(matches) >= min_neighbors:
           x1,y1 = bt.data[matches].T

 main()

我收到以下错误 -

   Traceback (most recent call last):
   File "testballtree.py", line 25, in <module>
   main()
   File "testballtree.py", line 23, in main
    x1,y1 = bt.data[matches].T
   File "stringsource", line 406, in      View.MemoryView.memoryview.__getitem__
   File "stringsource", line 746, in View.MemoryView.memview_slice
   TypeError: only integer scalar arrays can be converted to a scalar index

以与 scipy 相同的方式访问 scikit-learn 的 BallTree 中的数据属性的确切方法是什么？

  scikit-learn version is 0.19.2

Answer 1

看起来文档有误，bt.data 是内存视图而不是 numpy 数组。它可能应该是一个 numpy 数组或者是私有的。不过，您可以改用 points 来修复您的代码段。已打开 https://github.com/scikit-learn/scikit-learn/issues/11728

Answer 2

BallTree.data 是 a view or a copy of the training data，因此在上面的示例中，您可以直接使用 points 数组。

data 属性文档字符串确实不正确：它是内存视图而不是数组。您可以使用 numpy.asarray(bt.data).

将其转换回 numpy 数组

scikit-learn 的 BallTree 的类型错误

TypeError with scikit-learn's BallTree

python

typeerror

nearest-neighbor

scikit-learn

python-3.6