LoginSignup
0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?

More than 1 year has passed since last update.

@netinetikun( くん ねちねち)

[GNN pyg] を使用してみる

Last updated at Posted at 2024-06-30

自分のメモ程度にまとめたものになります。
参考になれば幸いです。

1. karateデータセットの確認

karate_dataset.py
from torch_geometric.datasets import KarateClub

data = KarateClub()[0]

for key in list(data):
    print(key)
コンソール画面
$ python karate_dataset.py

('x', tensor([[1., 0., 0.,  ..., 0., 0., 0.],
        [0., 1., 0.,  ..., 0., 0., 0.],
        [0., 0., 1.,  ..., 0., 0., 0.],
        ...,
        [0., 0., 0.,  ..., 1., 0., 0.],
        [0., 0., 0.,  ..., 0., 1., 0.],
        [0., 0., 0.,  ..., 0., 0., 1.]]))
('edge_index', tensor([[ 0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  1,  1,
          1,  1,  1,  1,  1,  1,  1,  2,  2,  2,  2,  2,  2,  2,  2,  2,  2,  3,
          3,  3,  3,  3,  3,  4,  4,  4,  5,  5,  5,  5,  6,  6,  6,  6,  7,  7,
          7,  7,  8,  8,  8,  8,  8,  9,  9, 10, 10, 10, 11, 12, 12, 13, 13, 13,
         13, 13, 14, 14, 15, 15, 16, 16, 17, 17, 18, 18, 19, 19, 19, 20, 20, 21,
         21, 22, 22, 23, 23, 23, 23, 23, 24, 24, 24, 25, 25, 25, 26, 26, 27, 27,
         27, 27, 28, 28, 28, 29, 29, 29, 29, 30, 30, 30, 30, 31, 31, 31, 31, 31,
         31, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 32, 33, 33, 33, 33, 33,
         33, 33, 33, 33, 33, 33, 33, 33, 33, 33, 33, 33],
        [ 1,  2,  3,  4,  5,  6,  7,  8, 10, 11, 12, 13, 17, 19, 21, 31,  0,  2,
          3,  7, 13, 17, 19, 21, 30,  0,  1,  3,  7,  8,  9, 13, 27, 28, 32,  0,
          1,  2,  7, 12, 13,  0,  6, 10,  0,  6, 10, 16,  0,  4,  5, 16,  0,  1,
          2,  3,  0,  2, 30, 32, 33,  2, 33,  0,  4,  5,  0,  0,  3,  0,  1,  2,
          3, 33, 32, 33, 32, 33,  5,  6,  0,  1, 32, 33,  0,  1, 33, 32, 33,  0,
          1, 32, 33, 25, 27, 29, 32, 33, 25, 27, 31, 23, 24, 31, 29, 33,  2, 23,
         24, 33,  2, 31, 33, 23, 26, 32, 33,  1,  8, 32, 33,  0, 24, 25, 28, 32,
         33,  2,  8, 14, 15, 18, 20, 22, 23, 29, 30, 31, 33,  8,  9, 13, 14, 15,
         18, 19, 20, 22, 23, 26, 27, 28, 29, 30, 31, 32]]))
('y', tensor([1, 1, 1, 1, 3, 3, 3, 1, 0, 1, 3, 1, 1, 1, 0, 0, 3, 1, 0, 1, 0, 1, 0, 0,
        2, 2, 0, 0, 2, 0, 0, 2, 0, 0]))
('train_mask', tensor([ True, False, False, False,  True, False, False, False,  True, False,
        False, False, False, False, False, False, False, False, False, False,
        False, False, False, False,  True, False, False, False, False, False,
        False, False, False, False]))

2. TransformerConvのアーキテクチャ

TFConv.py
import pickle

import torch
import torch.nn.functional as F

from torch_geometric.nn import GCNConv, TransformerConv, GATConv
from torch_geometric.nn import Linear

from torch_geometric.datasets import KarateClub

torch.manual_seed(1)

num_node_features = KarateClub().num_node_features
num_classes = KarateClub().num_classes

hidden_size = 64

class TFConv(torch.nn.Module):
    def __init__(self):
        super(TFConv, self).__init__()
        
        self.conv1 = TransformerConv(num_node_features, hidden_size)
        self.conv2 = TransformerConv(hidden_size, num_classes)
    
    def forward(self, data):
        x, edge_index = data.x, data.edge_index
        x = self.conv1(x, edge_index)
        x = F.relu(x)
        x = F.dropout(x, training=self.training)
        x = self.conv2(x, edge_index)
        
        return F.log_softmax(x, dim=1)

class GCN(torch.nn.Module):
    def __init__(self):
        super(GCN, self).__init__()
        
        self.conv1 = GCNConv(num_node_features, hidden_size)
        self.conv2 = GCNConv(hidden_size, num_classes)
    
    def forward(self, data):
        x, edge_index = data.x, data.edge_index
        x = self.conv1(x, edge_index)
        x = F.relu(x)
        x = F.dropout(x, training=self.training)
        x = self.conv2(x, edge_index)
        
        return F.log_softmax(x, dim=1)
    
class GAT(torch.nn.Module):
    def __init__(self):
        super(GAT, self).__init__()
        
        self.conv1 = GATConv(num_node_features, hidden_size)
        self.conv2 = GATConv(hidden_size, num_classes)
    
    def forward(self, data):
        x, edge_index = data.x, data.edge_index
        x = self.conv1(x, edge_index)
        x = F.relu(x)
        x = F.dropout(x, training=self.training)
        x = self.conv2(x, edge_index)
        
        return F.log_softmax(x, dim=1)

3. 学習用サンプルソース

learn.py
import time
import torch
import pickle
import torch.nn.functional as F

from sklearn.metrics import accuracy_score
from torch_geometric.datasets import KarateClub

from Net.TFConv import TFConv

PICKLE_PATH = './model/classification_model.pickle'

def main():
    dataset = KarateClub()
    
    data = dataset[0]

    model = TFConv(dataset)
    model.train()

    optimizer = torch.optim.Adam(model.parameters(), lr=0.01)
    #device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

    # learnig loop
    for epoch in range(100):
        optimizer.zero_grad()
        out = model(data)
        loss = F.nll_loss(out, data.y)
        loss.backward()
        optimizer.step()
        if epoch % 9 == 0:
            print('Epoch %d | Loss: %.4f' % (epoch, loss.item()))

    # モデルを評価モードに設定
    model.eval()

    # 推論
    _, pred = model(data).max(dim=1)

    print("結果：", pred)
    print("真値：", data.y)
    
    accuracy = accuracy_score(data.y, pred)
    print(f"正解率: {round(accuracy, 3)}")

    model.save_model(model, PICKLE_PATH)

if __name__ == "__main__":
    start = time.time()
    main()
    end = time.time()
    print(round(end - start, 3), "秒")

3-1. 実行してみる

コンソール結果画面
$ python learn.py

Epoch 0 | Loss: 1.4885
Epoch 9 | Loss: 1.3408
Epoch 18 | Loss: 1.2545
Epoch 27 | Loss: 1.0197
Epoch 36 | Loss: 0.8998
Epoch 45 | Loss: 0.6569
Epoch 54 | Loss: 0.5792
Epoch 63 | Loss: 0.5220
Epoch 72 | Loss: 0.4881
Epoch 81 | Loss: 0.3948
Epoch 90 | Loss: 0.3162
Epoch 99 | Loss: 0.4034
結果： tensor([1, 1, 1, 1, 3, 3, 3, 1, 0, 1, 3, 1, 1, 1, 0, 0, 3, 1, 0, 1, 0, 1, 0, 0,
        2, 2, 0, 0, 2, 0, 0, 2, 0, 0])
真値： tensor([1, 1, 1, 1, 3, 3, 3, 1, 0, 1, 3, 1, 1, 1, 0, 0, 3, 1, 0, 1, 0, 1, 0, 0,
        2, 2, 0, 0, 2, 0, 0, 2, 0, 0])
正解率: 1.0
0.203 秒

3-2. 保存したモデルを使用してみる

sample.py
import time
import pickle

from sklearn.metrics import accuracy_score

from torch_geometric.datasets import KarateClub

PICKLE_PATH = './model/classification_model.pickle'
        
def load_model(path = PICKLE_PATH):
    return pickle.load(open(path, 'rb'))

def main():
    dataset = KarateClub()
    
    data = dataset[0]
    model = load_model(PICKLE_PATH)

    model.eval()

    # 推論
    _, pred = model(data).max(dim=1)

    print("結果：", pred)
    print("真値：", data.y)
    
    accuracy = accuracy_score(data.y, pred)
    print(f"正解率: {round(accuracy, 3)}")

if __name__ == "__main__":
    start = time.time()
    main()
    end = time.time()
    print(round(end - start, 3), "秒")

3-3. 実行してみる

コンソール結果画面
$ python sample.py

結果： tensor([1, 1, 1, 1, 3, 3, 3, 1, 0, 1, 3, 1, 1, 1, 0, 0, 3, 1, 0, 1, 0, 1, 0, 0,
        2, 2, 0, 0, 2, 0, 0, 2, 0, 0])
真値： tensor([1, 1, 1, 1, 3, 3, 3, 1, 0, 1, 3, 1, 1, 1, 0, 0, 3, 1, 0, 1, 0, 1, 0, 0,
        2, 2, 0, 0, 2, 0, 0, 2, 0, 0])
正解率: 1.0
0.009 秒

3-4. Leave One Outを用いた検証

val.py
import time
import torch
import numpy as np
import pickle

import torch.nn.functional as F

from sklearn.metrics import accuracy_score
from torch_geometric.datasets import KarateClub

from Net.TFConv import TFConv

PICKLE_PATH = './model/classification_model.pickle'

# ノードの個数分の訓練maskを作成
def get_train_mask(dataset, index):
    train_mask = torch.tensor([True for i in range(dataset.data.y.shape[0])])
    train_mask[index] = False
    return train_mask

# ノードの個数分のテストmaskを作成
def get_test_mask(dataset, index):
    test_mask = torch.tensor([False for i in range(dataset.data.y.shape[0])])
    test_mask[index] = True
    return test_mask

def main():
    dataset = KarateClub()
    
    data = dataset[0]

    # 正解率保存用リスト
    accuracy_rate = np.array([])

    # ノードの個数分テストを行う
    for node in range(data.y.shape[0]):
        model = TFConv(dataset)
        model.train()
        
        # 訓練用のmask取得
        train_mask = get_train_mask(dataset, node)
        
        # テスト用のmask取得
        test_mask = get_test_mask(dataset, node)

        optimizer = torch.optim.Adam(model.parameters(), lr=0.01)
        #device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

        # 学習
        for _ in range(100):
            optimizer.zero_grad()
            out = model(data)
            loss = F.nll_loss(out[train_mask], data.y[train_mask])
            loss.backward()
            optimizer.step()
        
        print('Node %d | Loss: %.4f' % (node, loss.item()))

        # モデルを評価モードに設定
        model.eval()

        # 推論
        _, pred = model(data).max(dim=1)
        
        print("ノード：", node, "番目")
        print("結果：", pred[test_mask].item())
        print("真値：", data.y[test_mask].item())
        print()
        
        accuracy = accuracy_score(data.y, pred)
        accuracy_rate = np.append(accuracy, accuracy_rate)
    
    print("Leave One Out:[", round(np.mean(accuracy_rate), 3), "]")

if __name__ == "__main__":
    start = time.time()
    main()
    end = time.time()
    print(round(end - start, 3), "秒")

3-5. 実行してみる

コンソール結果画面
$ python val.py

Node 0 | Loss: 0.3723
ノード： 0 番目
結果： 1
真値： 1

.
.
.

Node 33 | Loss: 0.1673
ノード： 33 番目
結果： 0
真値： 0

Leave One Out:[ 0.984 ]
5.005 秒
0
0
0

Register as a new user and use Qiita more conveniently

  1. You get articles that match your needs
  2. You can efficiently read back useful information
  3. You can use dark theme
What you can do with signing up
Sign upLogin
0
0

Delete article

Deleted articles cannot be recovered.

Draft of this article would be also deleted.

Are you sure you want to delete this article?