如何删除与Scipy稀疏矩阵的数字不同的元素？

Question

I have a COO sparse matrix in which every element is a dictionary. I want to filter that matrix by some conditions, nevertheless when I try to multiply the matrix by the filter I get an exception TypeError: no supported conversion for types: (dtype('O'),) . Is it possible to avoid this issue?

from scipy.sparse import coo_matrix
import numpy as np

row = np.array([0, 0, 1, 2, 2, 2])
col = np.array([0, 2, 2, 0, 1, 2])
data = [{"x": 1}, {"y": -1}, {"x": -1}, {"x": 2}, {"t": -2}, {"z": 2}]
matrix = coo_matrix((data, (row, col)), shape=(3, 3))

matrix.multiply(np.array([0, 1, 0])) # Raises exception

Answer 1

It is possible to filter a coo_matrix , nevertheless it isn't straightforward. First you must create your filter mask with `True/False' values and then use it to index the matrix's column, rows and data vectors.

In [22]: mask = [True, False, False, True, False, False]

In [23]: matrix.data[mask]
Out[23]: array([{'x': 1}, {'x': 2}], dtype=object)

In [24]: matrix.col[mask]
Out[24]: array([0, 0], dtype=int32)

In [25]: matrix.col[row]
Out[25]: array([0, 0, 2, 2, 2, 2], dtype=int32)

In [26]: matrix.col[mask]
Out[26]: array([0, 0], dtype=int32)