一、数据集介绍

football数据集是根据美国大学生足球联赛而创建的一个数据集，包含115支球队（即图中的点）、616条比赛数据（即图中的边）

二、代码分析

1、加载包

import networkx as nx
import random
from tqdm import tqdm
from sklearn.decomposition import PCA
import matplotlib.pyplot as plt

2、加载数据集

# 数据加载，构造图
G = nx.read_gml('football.gml')
print('----------------')
#球队总数
print('len(G):', len(G)) 
#都有哪些球队
print('G.nodes():', G.nodes()) 
print('----------------')
#都有哪些比赛
# print('G.edges():', G.edges()) 

运行结果：

len(G): 115
G.nodes(): ['TexasElPaso', 'ArizonaState', 'MiamiFlorida', 'Army', 'MichiganState', 'SanDiegoState', 'ColoradoState', 'MiddleTennesseeState', 'Michigan', 'OhioState', 'AlabamaBirmingham', 'Vanderbilt', 'Mississippi', 'Clemson', 'California', 'NorthCarolina', 'Temple', 'Auburn', 'Maryland', 'Baylor', 'Wisconsin', 'Florida', 'WashingtonState', 'CentralMichigan', 'SouthernMississippi', 'Illinois', 'IowaState', 'BallState', 'Tennessee', 'UtahState', 'Nebraska', 'ArkansasState', 'NewMexico', 'KansasState', 'Indiana', 'Akron', 'SouthCarolina', 'Pittsburgh', 'Memphis', 'Colorado', 'Hawaii', 'Duke', 'Alabama', 'Navy', 'VirginiaTech', 'Tulane', 'Syracuse', 'BostonCollege', 'Arizona', 'FloridaState', 'SouthernMethodist', 'SanJoseState', 'Arkansas', 'TexasTech', 'Wyoming', 'Stanford', 'Kentucky', 'Utah', 'Missouri', 'NorthTexas', 'MississippiState', 'EastCarolina', 'Nevada', 'Purdue', 'PennState', 'LouisianaState', 'Virginia', 'Iowa', 'Tulsa', 'Rutgers', 'NotreDame', 'WesternMichigan', 'AirForce', 'Connecticut', 'Oregon', 'NewMexicoState', 'NorthernIllinois', 'Kansas', 'CentralFlorida', 'Northwestern', 'TexasA&M', 'OregonState', 'Cincinnati', 'LouisianaMonroe', 'Oklahoma', 'TexasChristian', 'Idaho', 'BrighamYoung', 'SouthernCalifornia', 'Marshall', 'Ohio', 'NorthCarolinaState', 'Houston', 'Kent', 'Toledo', 'BowlingGreenState', 'OklahomaState', 'WestVirginia', 'BoiseState', 'Minnesota', 'Texas', 'LouisianaLafayette', 'MiamiOhio', 'NevadaLasVegas', 'GeorgiaTech', 'Louisville', 'LouisianaTech', 'Rice', 'EasternMichigan', 'WakeForest', 'Washington', 'Georgia', 'FresnoState', 'UCLA', 'Buffalo']
----------------

3、deepwalk

从一个点出发，随机游走10个顶点。

def get_randomwalk(node, path_length):
    random_walk = [node]
    for i in range(path_length-1):
        temp = list(G.neighbors(node))
        temp = list(set(temp) - set(random_walk))    
        if len(temp) == 0:
            break
        random_node = random.choice(temp)
        random_walk.append(random_node)
        node = random_node        
    return random_walk

print('----------------')
print('get_randomwalk:', get_randomwalk('EastCarolina', 10))

运行结果

get_randomwalk: ['EastCarolina', 'AlabamaBirmingham', 'Memphis', 'Houston', 'Louisville', 'Tulane', 'LouisianaLafayette', 'NorthTexas', 'BoiseState', 'NewMexico']
100%|██████████| 115/115 [00:00<00:00, 3286.87it/s]

4、所有点随机游走

# 从图获取所有节点的列表
all_nodes = list(G.nodes())
# 捕获数据集中所有节点的随机游走序列
random_walks = []
for n in tqdm(all_nodes):
	# 每个节点游走5次，每次最长距离为10
    for i in range(5):
        random_walks.append(get_randomwalk(n, 10))
        
# 输出随机游走序列，及序列个数
print('----------------')
print('random_walks:', random_walks)
print('----------------')
print(len(random_walks))

5、使用word2vec训练随机游走的数据

from gensim.models import Word2Vec
import warnings
warnings.filterwarnings('ignore')

# 训练skip-gram (word2vec)模型
model = Word2Vec(window = 4, sg = 1, hs = 0,
                 negative = 10, # 负采样
                 alpha=0.03, min_alpha=0.0007,
                 seed = 14)
# 从random_walks中创建词汇表
model.build_vocab(random_walks, progress_per=2)
model.train(random_walks, total_examples = model.corpus_count, epochs=20, report_delay=1)
print('model:', model)
# 输出和EastCarolina相似的球队
print('model.similar_by_word:', model.similar_by_word('EastCarolina'))

运行结果

575
model: Word2Vec(vocab=115, size=100, alpha=0.03)
model.similar_by_word: [('Army', 0.9480490684509277), ('FloridaState', 0.9365841746330261), ('Louisville', 0.9252589344978333), ('Tulane', 0.9110944867134094), ('AlabamaBirmingham', 0.9061299562454224), ('GeorgiaTech', 0.9037945866584778), ('WakeForest', 0.9031023979187012), ('Virginia', 0.8917067646980286), ('Duke', 0.8815759420394897), ('Maryland', 0.8811277151107788)]

6、绘图

def plot_nodes(word_list):
	# 每个节点的embedding为100维
    X = model[word_list]
    #print(type(X))
    # 将100维向量减少到2维
    pca = PCA(n_components=2)
    result = pca.fit_transform(X) 
    #print(result)
    # 绘制节点向量
    plt.figure(figsize=(12,9))
    # 创建一个散点图的投影
    plt.scatter(result[:, 0], result[:, 1])
    for i, word in enumerate(word_list):
        plt.annotate(word, xy=(result[i, 0], result[i, 1]))
    plt.show()
# 将所有的球队embedding进行绘制
plot_nodes(model.wv.vocab)

运行结果

三、完整代码

# 使用DeepWalk
import networkx as nx
import random
from tqdm import tqdm
from sklearn.decomposition import PCA
import matplotlib.pyplot as plt


# 数据加载，构造图
G = nx.read_gml('football.gml')
print('----------------')
#球队总数
print('len(G):', len(G)) 
#都有哪些球队
print('G.nodes():', G.nodes()) 
print('----------------')
#都有哪些比赛
print('G.edges():', G.edges()) 


""" 
随机游走
input: 将节点和被遍历的路径的长度作为输入
output: 返回遍历节点的顺序:
"""
def get_randomwalk(node, path_length):
    random_walk = [node]
    for i in range(path_length-1):
        temp = list(G.neighbors(node))
        temp = list(set(temp) - set(random_walk))    
        if len(temp) == 0:
            break
        random_node = random.choice(temp)
        random_walk.append(random_node)
        node = random_node        
    return random_walk

print('----------------')
print('get_randomwalk:', get_randomwalk('EastCarolina', 10))


# 从图获取所有节点的列表
all_nodes = list(G.nodes())
# 捕获数据集中所有节点的随机游走序列
random_walks = []
for n in tqdm(all_nodes):
	# 每个节点游走5次，每次最长距离为10
    for i in range(5):
        random_walks.append(get_randomwalk(n, 10))
        
# 输出随机游走序列，及序列个数
print('----------------')
print('random_walks:', random_walks)
print('----------------')
print(len(random_walks))


# 使用skip-gram，提取模型学习到的权重
from gensim.models import Word2Vec
import warnings
warnings.filterwarnings('ignore')

# 训练skip-gram (word2vec)模型
model = Word2Vec(window = 4, sg = 1, hs = 0,
                 negative = 10, # 负采样
                 alpha=0.03, min_alpha=0.0007,
                 seed = 14)
# 从random_walks中创建词汇表
model.build_vocab(random_walks, progress_per=2)
model.train(random_walks, total_examples = model.corpus_count, epochs=20, report_delay=1)
print('model:', model)
# 输出和EastCarolina相似的球队
print('model.similar_by_word:', model.similar_by_word('EastCarolina'))


# 在二维空间中绘制所选节点的向量
def plot_nodes(word_list):
	# 每个节点的embedding为100维
    X = model[word_list]
    #print(type(X))
    # 将100维向量减少到2维
    pca = PCA(n_components=2)
    result = pca.fit_transform(X) 
    #print(result)
    # 绘制节点向量
    plt.figure(figsize=(12,9))
    # 创建一个散点图的投影
    plt.scatter(result[:, 0], result[:, 1])
    for i, word in enumerate(word_list):
        plt.annotate(word, xy=(result[i, 0], result[i, 1]))
    plt.show()
# 将所有的球队embedding进行绘制
plot_nodes(model.wv.vocab)