[Model] BPHGNN #246

LaooBaoo · 2024-08-06T12:57:24Z

No description provided.

gyzhou2000 · 2024-08-09T09:00:58Z

需要在提交之前，重新拉取主分支上的代码，保证代码是最新的
拉取最新代码之后，解决代码冲突

docs/source/api/model.rst

docs/source/api/trainerflow.rst

openhgnn/trainerflow/MHGCN_trainer.py

gyzhou2000 · 2024-08-09T09:09:33Z

openhgnn/trainerflow/MHGCN_trainer.py

+        train_dataset = TextDataset('OpenHGNN/openhgnn/dataset/data/MHGCN_dataset/train.txt')
+        valid_dataset = TextDataset('OpenHGNN/openhgnn/dataset/data/MHGCN_dataset/valid.txt')
+        test_dataset = TextDataset('OpenHGNN/openhgnn/dataset/data/MHGCN_dataset/test.txt')
+
+        train_loader = DataLoader(train_dataset, batch_size=32, shuffle=True)
+        valid_loader = DataLoader(valid_dataset, batch_size=32, shuffle=False)
+        test_loader = DataLoader(test_dataset, batch_size=32, shuffle=False)


这部分代码考虑一下是否要放在 __init__ 函数中

gyzhou2000 · 2024-08-09T09:11:44Z

openhgnn/trainerflow/MHGCN_trainer.py

+
+                output = self.model(features=features, hg=self.hg, encode=features)
+
+                loss = F.cross_entropy(output, labels)  # 计算损失


不要出现中文注释

gyzhou2000 · 2024-08-19T07:52:19Z

docs/source/api/model.rst

@@ -10,7 +10,7 @@ Models
    {% for cls in openhgnn.models.classes %}
      {{ cls }}
    {% endfor %}
-
+   


不需要添加无意义的空行

gyzhou2000 · 2024-08-19T07:52:28Z

docs/source/api/trainerflow.rst

@@ -10,7 +10,7 @@ Trainerflow
    {% for cls in openhgnn.trainerflow.classes %}
      {{ cls }}
    {% endfor %}
-
+    


gyzhou2000 · 2024-08-19T07:53:17Z

main.py

@@ -7,7 +7,7 @@

 if __name__ == '__main__':
    parser = argparse.ArgumentParser()
-    parser.add_argument('--model', '-m', default='RGCN', type=str, help='name of models')
+    parser.add_argument('--model', '-m', default='BPHGNN', type=str, help='name of models')


不要修改 main.py 中默认的超参数，如果想运行自己的模型的话，可以在命令行中设置，python mian.py --model BPHGNN

gyzhou2000 · 2024-08-19T07:57:34Z

openhgnn/tasks/node_classification.py

@@ -28,7 +28,7 @@ class NodeClassification(BaseTask):
    def __init__(self, args):
        super(NodeClassification, self).__init__()
        self.logger = args.logger
-        self.dataset = build_dataset(args.dataset, 'node_classification', logger=self.logger)
+        self.dataset = build_dataset(args.dataset, 'node_classification', logger=self.logger) # 返回node_classificationDataset对象


删除中文注释

gyzhou2000 · 2024-08-19T07:59:30Z

openhgnn/trainerflow/base_flow.py

@@ -137,11 +137,11 @@ def init_feature(self, act):
        if isinstance(self.hg.ndata['h'], dict):
            # The heterogeneous contains more than one node type.
            input_feature = HeteroFeature(self.hg.ndata['h'], get_nodes_dict(self.hg),
-                                            self.args.hidden_dim, act=act).to(self.device)
+                                            self.args.hidden, act=act).to(self.device)


不要修改 base_flow.py 中的默认参数名称，可将 config.ini 文件中对应的超参数名改成 hidden_dim

gyzhou2000 · 2024-08-19T08:03:48Z

openhgnn/trainerflow/BPHGNN_trainer.py

+            opt = torch.optim.Adam([{'params': self.model.parameters(), 'lr': 0.01}, {'params': log.parameters()}], lr=0.001, weight_decay=0.0005)
+            log.to(device)


现在优化器中的学习率使用的是固定参数，这样做不准确。需要将值设置为 config.ini 中的参数值

gyzhou2000 · 2024-08-19T08:04:40Z

openhgnn/trainerflow/BPHGNN_trainer.py

+        except:
+            pass
+        labels = labels.astype(np.int16)
+        device=torch.device('cuda')


在 main.py 中已经通过 --gpu 这个参数来设置tensor的位置了，不需要重复设置

gyzhou2000 · 2024-08-19T08:06:38Z

openhgnn/trainerflow/BPHGNN_trainer.py

+class TextDataset(Dataset):
+    def __init__(self, filepath):
+        self.data = []
+        with open(filepath, 'r') as f:
+            for line in f:
+
+                split_line = list(map(int, line.strip().split()))
+                self.data.append(split_line)
+
+    def __len__(self):
+        return len(self.data)


这个类在后续代码中并没有用到，可以删除

gyzhou2000 · 2024-08-19T08:18:53Z

openhgnn/dataset/BPHGNN_dataset.py

+class TextDataset(Dataset):
+    def __init__(self, filepath):
+        self.data = []
+        with open(filepath, 'r') as f:
+            for line in f:
+
+                split_line = list(map(int, line.strip().split()))
+                self.data.append(split_line)
+
+    def __len__(self):
+        return len(self.data)
+
+    def __getitem__(self, idx):
+
+        data = self.data[idx]
+        features = torch.tensor(data[:-1], dtype=torch.float32)
+        labels = torch.tensor(data[-1], dtype=torch.long)
+        return features, labels
+
+def load_edges(file_path):
+    edge_data = {
+        1: [],  
+        2: [],   
+        3: []   
+    }
+
+    with open(file_path, 'r') as f:
+        for line in f:
+
+            split_line = list(map(int, line.strip().split()))
+            edge_type, src_id, dst_id, _ = split_line
+            edge_data[edge_type].append((src_id, dst_id))
+
+    return edge_data
+
+def build_hetero_graph():
+
+    edge_data = load_edges('OpenHGNN/openhgnn/dataset/data/test/BPHGNN_dataset/alibaba_small/train.txt')
+
+    data_dict = {
+        ('node', 'relation1', 'node'): edge_data[1],
+        ('node', 'relation2', 'node'): edge_data[2],
+        ('node', 'relation3', 'node'): edge_data[3],
+    }
+
+    hg = dgl.heterograph(data_dict)
+    return hg
+
+
+train_dataset = TextDataset('OpenHGNN/openhgnn/dataset/data/BPHGNN_dataset/train.txt')
+valid_dataset = TextDataset('OpenHGNN/openhgnn/dataset/data/BPHGNN_dataset/valid.txt')
+test_dataset = TextDataset('OpenHGNN/openhgnn/dataset/data/BPHGNN_dataset/test.txt')
+
+train_loader = DataLoader(train_dataset, batch_size=32, shuffle=True)
+valid_loader = DataLoader(valid_dataset, batch_size=32, shuffle=False)
+test_loader = DataLoader(test_dataset, batch_size=32, shuffle=False)


dataset 文件的内容可以参考其他dataset文件的编写内容，这样写不对

gyzhou2000 · 2024-08-19T08:21:22Z

openhgnn/dataset/NodeClassificationDataset.py

@@ -500,6 +502,7 @@ def __init__(self, dataset_name, *args, **kwargs):
        self.train_idx, self.valid_idx, self.test_idx = split_idx["train"][self.category], split_idx["valid"][
            self.category], split_idx["test"][self.category]
        self.g, self.label_dict = dataset[0]
+        self.BPHGNN_g = self.mag4mhgcn(dataset)


bphgnn模型为什么要调用self.mag4mhgcn(dataset)

第一次提交

efc67b1

gyzhou2000 changed the title ~~MHGCN-check-for 周光煜~~ [Model] BPHGNN Aug 9, 2024

gyzhou2000 reviewed Aug 9, 2024

View reviewed changes

LaooBaoo added 2 commits August 19, 2024 11:36

包骏飞——BPHGNN模型复现

175c06e

取消了中文注释

9759302

gyzhou2000 reviewed Aug 19, 2024

View reviewed changes

LaooBaoo and others added 3 commits August 19, 2024 21:39

第三次提交

475f5db

Merge remote-tracking branch 'upstream/main' into feature-branch

688c691

resolve conflict

4223002

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[Model] BPHGNN #246

[Model] BPHGNN #246

LaooBaoo commented Aug 6, 2024

gyzhou2000 commented Aug 9, 2024 •

edited

Loading

gyzhou2000 Aug 9, 2024

gyzhou2000 Aug 9, 2024

gyzhou2000 Aug 19, 2024

gyzhou2000 Aug 19, 2024

gyzhou2000 Aug 19, 2024

gyzhou2000 Aug 19, 2024

gyzhou2000 Aug 19, 2024

gyzhou2000 Aug 19, 2024

gyzhou2000 Aug 19, 2024

gyzhou2000 Aug 19, 2024

gyzhou2000 Aug 19, 2024

gyzhou2000 Aug 19, 2024


		output = self.model(features=features, hg=self.hg, encode=features)

		loss = F.cross_entropy(output, labels) # 计算损失

		opt = torch.optim.Adam([{'params': self.model.parameters(), 'lr': 0.01}, {'params': log.parameters()}], lr=0.001, weight_decay=0.0005)
		log.to(device)

[Model] BPHGNN #246

Are you sure you want to change the base?

[Model] BPHGNN #246

Conversation

LaooBaoo commented Aug 6, 2024

gyzhou2000 commented Aug 9, 2024 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

gyzhou2000 commented Aug 9, 2024 •

edited

Loading