如何可视化神经网络的神经元节点之间的连接？附有Python预处理代码

如何可视化神经网络的神经元节点之间的连接？

1 导出表格

将网络的权重矩阵导出为三个表格

（1）nodes.csv

一列表示，为每个节点的编码

（2）links.csv

这个表格表示节点的连接，第一列为顺序编码，唯一表示，第二列是节点的起点，第二个为节点的终点

（3）node_class.csv

这个表格表示每个节点的类别，方便后续对不同的节点绘制不同的颜色

import pandas as pd

# 将小数值的权重矩阵转为0-1表示的矩阵
def convert_to_binary(matrix):
    binary_matrix = []
    for row in matrix:
        binary_row = [1 if value != 0 else 0 for value in row]
        binary_matrix.append(binary_row)
    return np.array(binary_matrix)

def generate_data(W):

    adj_binary_matrix = convert_to_binary(W)
    # 初始化邻接矩阵
    adj_matrix = adj_binary_matrix

    # 获取节点数量
    num_nodes = adj_matrix.shape[0]

    # 生成节点编码
    node_codes = ['node' + str(i+1) for i in range(num_nodes)]

    # 创建 nodes.csv 文件
    nodes_df = pd.DataFrame(node_codes, columns=['node'])


    # 创建 links.csv 文件
    links = []
    id = 1
    for i in range(num_nodes):
        for j in range(num_nodes):
            if adj_matrix[i][j] == 1:
                links.append((id, node_codes[i], node_codes[j]))
                id+=1

    links_df = pd.DataFrame(links, columns=['id', 'source', 'target'])
    links_df.to_csv(f'links.csv', index=False)
    nodes_df.to_csv(f'nodes.csv', index=False)
    # 对网络不同度的节点分类，分为高度节点（（hub））和非高度节点（non-hub）

    # 计算每个节点的度
    degrees = np.sum(adj_matrix != 0, axis=1)

    # 定义度的连接比例为threshold的为hub节点
    threshold  = 0.7
    threshold_degree = np.quantile(degrees,threshold)
    # 根据节点的度将它们分为两类
    node_classes = ['hub node' if degree >= threshold_degree else 'non-hub node' for degree in degrees]

    # 将结果保存到名为node_class.csv的表格中
    result = pd.DataFrame({'id': node_codes, 'class': node_classes})
    result.to_csv('node_class.csv', index=False)
# 假设神经网络每个神经元之间的邻接矩阵
# 初始化一个10*10的邻接矩阵，初始值全为0
adj_matrix = [
    [-0.000000,0.000000,0.000000,0.000000,0.000000,-0.000000,0.122116,-0.029914,0.364874,0.000000],
[0.000000,-0.000000,-0.003972,-0.035493,-0.412647,-0.000000,-0.000000,-0.000000,-0.000000,-0.000000],
[-0.000000,-0.086977,-0.000000,0.190105,0.000000,-0.153798,-0.000000,0.189070,0.000000,-0.000000],
[-0.000000,0.368288,-0.000000,-0.000000,-0.000000,0.057519,0.000000,-0.000000,0.000000,-0.000000],
[0.000000,-0.000000,-0.214501,0.000000,-0.443936,0.000000,0.000000,0.000000,-0.000000,-0.000000],
[-0.000000,-0.172708,0.052662,-0.184544,0.000000,0.024721,0.000000,-0.300713,-0.414200,0.000000],
[-0.137718,0.000000,0.000000,-0.000000,-0.000000,0.258740,0.099708,0.000000,0.032948,-0.190810],
[-0.000000,-0.106000,0.195423,0.000000,-0.000000,0.029322,-0.000000,0.077991,0.000000,-0.000000],
[0.252584,-0.077399,-0.000000,-0.000000,0.000000,-0.338475,0.498515,0.000000,-0.238516,0.597768],
[-0.000000,0.000000,0.048298,-0.000000,-0.000000,0.000000,0.016180,-0.000000,0.176588,0.000000]]
generate_data(adj_matrix)

2 绘图

免费从Chiplot网络绘制，导入以上生成的三个表格，网站有具体的教程
ChiPlot网站