scikit-learn的BallTree发生TypeError

时间:2018-08-01 07:32:45

标签: python scikit-learn python-3.6 typeerror nearest-neighbor

我有一个带有scipy的cKDTree的toyexample,效果很好,我想用scikit-learn的BallTree做类似的代码

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data属性的确切方法是什么?

  scikit-learn version is 0.19.2

2 个答案:

答案 0 :(得分:2)

看起来文档是错误的,bt.data是内存视图,而不是numpy数组。它可能应该是一个numpy数组或私有的。不过,您可以改为使用points来修复代码段。打开https://github.com/scikit-learn/scikit-learn/issues/11728

答案 1 :(得分:1)

BallTree.dataa 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:它是一个memoryview而不是一个数组。您可以使用numpy.asarray(bt.data)将其转换回numpy数组。