[Python Graph2Vec]グラフ情報を埋め込んでグラフ予測をしてみる

1. Graph2Vecを試してみる

pythonのGraph2Vecを試してみたので、よければ参考にしてください。

1-1. ライブラリのインストール

$ python -V

Python 3.11.1

$ pip install karateclub

graph2vec.py

import numpy as np
import networkx as nx
import matplotlib.pyplot as plt

from sklearn.decomposition import PCA
from torch_geometric.datasets import TUDataset
from karateclub.graph_embedding import Graph2Vec

# MUTAGデータセットを読み込み
dataset = TUDataset(root="./", name="MUTAG")

# グラフ情報の前処理
graphs = []
for data in dataset:
    e_list = []
    edge_index = data.edge_index
    for i in range(len(edge_index[0])):
        e_list.append((int(edge_index[0][i]), int(edge_index[1][i])))
    g = nx.from_edgelist(e_list)

    x = data.x
    nx.set_node_attributes(g, {j: str(j) for j in range(g.number_of_nodes())}, "feature")

    node_indices = sorted([node for node in g.nodes()])
    numeric_indices = [index for index in range(g.number_of_nodes())]

    if numeric_indices == node_indices:
        graphs.append(g)

# モデルの呼び出し
model = Graph2Vec(wl_iterations=4, epochs=150)

# グラフ情報の予測
model.fit(graphs)

emb128d = model.get_embedding()

print("128d:", emb128d.shape)

# 主成分分析
pca = PCA(n_components=2)
emb2d = pca.fit_transform(emb128d)
print("2d:", emb2d.shape)

np.save('./embedded_data/graph2vec_MUTAG', emb2d)

# 埋め込みデータをグラフで表示
plt.title("node embedding in 2D")
plt.scatter(emb2d[:,0],emb2d[:,1])
plt.show()

1-2. 実行してみる

$ python graph2vec.py

# ("128d:", 188, 128)
# ("2d:", 188, 2)

1-3. 埋め込みグラフを表示する

2. SVMを用いてグラフ予測をしてみる

svm.py

import numpy as np

import torchvision.transforms as transforms

from torch_geometric.datasets import TUDataset
from sklearn.model_selection import LeaveOneOut

from sklearn.svm import SVC
from sklearn.metrics import accuracy_score


DATA_PATH = f'./embedded_data/graph2vec_MUTAG.npy'

# 保存した埋め込みデータを読み込む
x = np.load(DATA_PATH)

# MUTAGデータセットを読み込み
dataset = TUDataset(root="./", name="MUTAG")

y = []

for data in dataset:

    y.append(data.y.item())

# 正解ラベル
y = np.array(y)

# 予測ラベル保存用
pred_list = np.array([])

# SVMモデル呼び出し
model = SVC(kernel='linear')

# Leave One Outを用いる
loo = LeaveOneOut()

i = 1

# Leave One Outを用いる
for train_index, test_index in loo.split(x):
    x_train, x_test = x[train_index], x[test_index]
    y_train, y_test = y[train_index], y[test_index]

    # 埋め込みデータを学習
    model.fit(x_train, y_train)
    
    # 予測
    pred = model.predict(x_test)
    
    print(i, "番目")
    print("予測：", pred)
    print("正解：", y_test)
    print()
    
    pred_list = np.append(pred_list, pred.item())
    i = i + 1

accuracy = accuracy_score(y, pred_list)
print(f"正解率: {round(accuracy, 3)}")

2-1. 実行してみる

$ python svm.py

.
.
.

正解率: 0.665

3. ニューラルネットワークでグラフ分類をしてみる

nn.py

import torch
import numpy as np
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F

from sklearn.model_selection import LeaveOneOut
from torch_geometric.datasets import TUDataset
from sklearn.metrics import accuracy_score

class Net(nn.Module):    
    def __init__(self):
        super(Net, self).__init__()
        self.fc1 = nn.Linear(2, 32)
        self.fc2 = nn.Linear(32, 4)
    
    def forward(self, x):
        x = self.fc1(x)
        x = F.relu(x)
        x = self.fc2(x)
        x = F.softmax(x, dim=1)
        return x

DATA_PATH = f'./embedded_data/graph2vec_MUTAG.npy'

# 保存した埋め込みデータを読み込む
x = torch.from_numpy(np.load(DATA_PATH)).clone()

# MUTAGデータセットを読み込み
dataset = TUDataset(root="./", name="MUTAG")

y = []

for data in dataset:

    y.append(data.y.item())

# 正解ラベル
y = torch.tensor(y)

# 予測ラベル保存用
pred_list = np.array([])

# Leave One Outを用いる
loo = LeaveOneOut()

# モデル呼び出し
model = Net()
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=0.01)

i = 1

# Leave One Outを用いる
for train_index, test_index in loo.split(x):
    
    x_train, x_test = x[train_index], x[test_index]
    y_train, y_test = y[train_index], y[test_index]
    
    # 学習
    model.train()
    for epoch in range(100):
        optimizer.zero_grad()
        out = model(x_train)
        loss = criterion(out, y_train)
        loss.backward(retain_graph=True)
        optimizer.step()
    
    # 予測
    model.eval()
    _, pred = model(x_test).max(dim=1)

    print(i, "番目")
    print("予測：", pred.item())
    print("正解：", y_test.item())
    print("loss:", round(loss.item(), 3))
    print()
    
    pred_list = np.append(pred_list, pred.item())
    i = i + 1

accuracy = accuracy_score(y, pred_list)
print(f"正解率: {round(accuracy, 3)}")

3-1. 実行してみる

$ python nn.py

.
.
.

正解率: 0.846