如何创建具有时间重叠的邻接矩阵?

时间:2018-11-06 17:19:12

标签: python r pandas dplyr networkx

考虑这个简单的例子

#python bros
pd.DataFrame({'id' : [1,1,2,3],
                       'time_in' : [0,30,1,5],
                       'time_out' : [2,35,3,6]})
Out[66]: 
   id  time_in  time_out
0   1        0         2
1   1       30        35
2   2        1         3
3   3        5         6


#R bros
dplyr::data_frame(id = c(1,1,2,3),
                  time_in = c(0,30,1,5),
                  time_out = c(2,35,3,6))

在这里,解释很简单。

个人1在时间0和时间2之间停留在给定的位置。个体2在时间1和时间3之间停留。因此,个人2与个人1相遇,并在我的网络中与其连接。

也就是说,我网络的节点是id,如果两个节点的[time_in, time_out]间隔重叠,则在两个节点之间会有一条边。

是否有一种有效的方法可以从此输入数据中生成adjacency matrixedge list,以便可以在诸如networkx之类的网络程序包中使用它们?我的真实数据集比那还要大。

谢谢!

1 个答案:

答案 0 :(得分:2)

我认为这是制作邻接矩阵的可能解决方案。这样做的目的是将每个时隙相互比较,然后按顶点组减少比较。

import numpy as np
import pandas as pd

df = pd.DataFrame({'id' : [1, 1, 2, 3],
                   'time_in' : [0, 30, 1, 5],
                   'time_out' : [2, 35, 3, 6]})
# Sort so equal ids are together
df.sort_values('id', inplace=True)
# Get data arrays
ids = df.id.values
t_in = df.time_in.values
t_out = df.time_out.values
# Graph vertices
vertices = np.unique(ids)
# Find time slot overlaps
overlaps = (t_in[:, np.newaxis] <= t_out) & (t_out[:, np.newaxis] >= t_in)
# Find vertex group slices
reduce_idx = np.concatenate([[0], np.where(np.diff(ids) != 0)[0] + 1])
# Reduce by vertex groups to make adjacency matrix
connect = np.logical_or.reduceat(overlaps, reduce_idx, axis=1)
connect = np.logical_or.reduceat(connect, reduce_idx, axis=0)
# Clear diagonal if you want to remove self-connection
i = np.arange(len(vertices))
connect[i, i] = False
# Adjacency matrix as data frame
graph_df = pd.DataFrame(connect, index=vertices, columns=vertices)
print(graph_df)

输出:

       1      2      3
1  False   True  False
2   True  False  False
3  False  False  False