SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2406.10310 · NeurIPS · 2024

TEG-DB: A Comprehensive Dataset and Benchmark of Textual-Edge Graphs

Liang Zhao, Zheng Zhang, Chen Ling, Sirui Li, Zhuofeng Li, Zixing Gou, Xiangnan Zhang, Zhongyuan Liu, Yuntong Hu

arXiv · PDF · Open in the Atlas

Code that ran

8 of 11 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 6 failures of our instrument; 3 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2406.10310.json · badge: /api/ran/2406.10310.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
Zhuofeng-Li/TEG-Benchmark canonical (classifier label)
classifier basis: census
8 ran · 3 no recorded run
FunctionStatusWhere it lives
split_graph Ran
ran; our contract draft was wrong, not the code
Zhuofeng-Li/TEG-Benchmark/GNN/model/Dataloader.py
code served (permissive licence) · get_code("2ead908edee5bf31")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np

ARGS = {
    "nodes_num": 100,
    "train_ratio": 0.7,
    "val_ratio": 0.2
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
test Ran
ran; our contract draft was wrong, not the code
Zhuofeng-Li/TEG-Benchmark/GNN/GNN_Link_HitsK.py
code served (permissive licence) · get_code("f0581e5668ba1873")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.utils.data import DataLoader

torch.manual_seed(0)

# Model: needs to accept (x, adj_t) and return node embeddings h
class SimpleGNNModel(nn.Module):
    def __init__(self, input_dim, hidden_dim):
        super().__init__()
        self.linear = nn.Linear(input_dim, hidden_dim)
    
    def forward(self, x, adj_t):
        # Simple pass-through for testing
        return self.linear(x)

# Predictor: needs to accept two node embeddings and return scores
class SimplePredictor(nn.Module):
    def __init__(self, hidden_dim):
        super().__init__()
        self.linear = nn.Linear(hidden_dim * 2, 1)
    
    def forward(self, h1, h2):
        combined = torch.cat([h1, h2], dim=-1)
        return self.linear(combined)

# Evaluator: needs eval method that accepts y_pred_pos, y_pred_neg and returns dict with hits@K
class SimpleEvaluator:
    def __init__(self):
        self.K = 10
    
    def eval(self, data):
        # Simple mock evaluator that returns random hits
        pos_pred = data['y_pred_pos']
        neg_pred = data['y_pred_neg']
        # Mock hits calculation
        hits = torch.sigmoid(pos_pred.mean() - neg_pred.mean()).item()
        return {f'hits@{self.K}': hits}

# Set up parameters
num_nodes = 6
input_dim = 4
hidden_dim = 3
num_train_edges = 2
num_valid_edges = 2
num_test_edges = 2

# Create model and predictor
model = SimpleGNNModel(input_dim, hidden_dim)
predictor = SimplePredictor(hidden_dim)

# Create node features
x = torch.randn(num_nodes, input_dim)

# Create adjacency (sparse tensor format)
adj_t = torch.sparse_coo_tensor(
    indices=torch.tensor([[0, 1, 2], [1, 2, 3]]),
    values=torch.ones(3),
    size=(num_nodes, num_nodes)
)

# Create edge split dictionary
edge_split = {
    'train': {
        'edge': torch.tensor([[0, 1], [1, 2], [2, 3], [3, 4]], dtype=torch.long)
    },
    'valid': {
        'edge': torch.tensor([[0, 2], [1, 3], [2, 4], [3, 5]], dtype=torch.long),
        'edge_neg': torch.tensor([[0, 3], [1, 4], [2, 5], [3, 0]], dtype=torch.long)
    },
    'test': {
        'edge': torch.tensor([[0, 4], [1, 5], [2, 0], [3, 1]], dtype=torch.long),
        'edge_neg': torch.tensor([[0, 5], [1, 0], [2, 1], [3, 2]], dtype=torch.long)
    }
}

# Create evaluator
evaluator = SimpleEvaluator()

# Batch size
batch_size = 2

ARGS = {
    'model': model,
    'predictor': predictor,
    'x': x,
    'adj_t': adj_t,
    'edge_split': edge_split,
    'evaluator': evaluator,
    'batch_size': batch_size
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
test Ran
ran; our contract draft was wrong, not the code
Zhuofeng-Li/TEG-Benchmark/GNN/GNN_Link_HitsK_loader.py
code served (permissive licence) · get_code("64d97bf20e2f647a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Define a minimal model that takes x and adj_t
class MinimalModel(nn.Module):
    def __init__(self, input_dim, hidden_dim):
        super().__init__()
        self.linear = nn.Linear(input_dim, hidden_dim)
    
    def forward(self, x, adj_t):
        # x: node features, adj_t: adjacency
        return self.linear(x)

# Define a minimal predictor that takes two node embeddings
class MinimalPredictor(nn.Module):
    def __init__(self, hidden_dim):
        super().__init__()
        self.linear = nn.Linear(hidden_dim * 2, 1)
    
    def forward(self, h_src, h_dst):
        # Concatenate source and destination embeddings
        combined = torch.cat([h_src, h_dst], dim=-1)
        return self.linear(combined)

# Define a minimal batch class
class MinimalBatch:
    def __init__(self, num_nodes, num_edges, feature_dim):
        self.x = torch.randn(num_nodes, feature_dim)
        self.adj_t = torch.randn(num_nodes, num_nodes)
        self.edge_label_index = torch.tensor([[0, 1, 2], [1, 2, 0]], dtype=torch.long)
    
    def to(self, device):
        self.x = self.x.to(device)
        self.adj_t = self.adj_t.to(device)
        self.edge_label_index = self.edge_label_index.to(device)
        return self

# Define a minimal evaluator
class MinimalEvaluator:
    def __init__(self):
        self.K = 10
    
    def eval(self, data):
        # data contains 'y_pred_pos' and 'y_pred_neg'
        pos_pred = data['y_pred_pos']
        neg_pred = data['y_pred_neg']
        # Simple mock evaluation: return hits based on K
        hits = min(self.K, len(pos_pred)) / max(1, len(pos_pred))
        return {f'hits@{self.K}': hits}

# Setup components
device = 'cpu'
input_dim = 4
hidden_dim = 3
num_nodes = 5
num_edges = 3
feature_dim = 4

model = MinimalModel(input_dim, hidden_dim)
model.eval()

predictor = MinimalPredictor(hidden_dim)
predictor.eval()

# Create dataloaders (lists of batches)
train_loader = [MinimalBatch(num_nodes, num_edges, feature_dim) for _ in range(2)]
valid_loader = [MinimalBatch(num_nodes, num_edges, feature_dim) for _ in range(2)]
valid_negative_loader = [MinimalBatch(num_nodes, num_edges, feature_dim) for _ in range(2)]
test_loader = [MinimalBatch(num_nodes, num_edges, feature_dim) for _ in range(2)]
test_negative_loader = [MinimalBatch(num_nodes, num_edges, feature_dim) for _ in range(2)]

dataloaders = {
    'train': train_loader,
    'valid': valid_loader,
    'valid_negative': valid_negative_loader,
    'test': test_loader,
    'test_negative': test_negative_loader
}

evaluator = MinimalEvaluator()

ARGS = {
    'model': model,
    'predictor': predictor,
    'dataloaders': dataloaders,
    'evaluator': evaluator,
    'device': device
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
test Ran
ran; our contract draft was wrong, not the code
Zhuofeng-Li/TEG-Benchmark/GNN/GNN_Link_MRR.py
code served (permissive licence) · get_code("8b3cabb047d30449")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.utils.data import DataLoader

torch.manual_seed(0)

# Build a minimal model that takes x and adj_t and returns embeddings h
class SimpleModel(nn.Module):
    def __init__(self, input_dim, output_dim):
        super().__init__()
        self.linear = nn.Linear(input_dim, output_dim)
    
    def forward(self, x, adj_t):
        # Just return a simple embedding based on x
        return self.linear(x)

# Build a minimal predictor that takes two embeddings and returns a score
class SimplePredictor(nn.Module):
    def __init__(self, embedding_dim):
        super().__init__()
        self.linear = nn.Linear(embedding_dim * 2, 1)
    
    def forward(self, h_src, h_dst):
        combined = torch.cat([h_src, h_dst], dim=-1)
        return self.linear(combined)

# Build a minimal evaluator
class SimpleEvaluator:
    def eval(self, input_dict):
        # Returns a dict with 'mrr_list' containing a tensor
        y_pred_pos = input_dict['y_pred_pos']
        y_pred_neg = input_dict['y_pred_neg']
        # Return dummy MRR values
        mrr_list = torch.tensor([0.8])
        return {'mrr_list': mrr_list}

# Set up dimensions
num_nodes = 8
input_dim = 4
embedding_dim = 3
batch_size = 2
neg_len = 2

# Create model and predictor
model = SimpleModel(input_dim, embedding_dim)
predictor = SimplePredictor(embedding_dim)

# Create node features
x = torch.randn(num_nodes, input_dim)

# Create adjacency (sparse tensor)
adj_t = torch.sparse_coo_tensor(
    indices=torch.tensor([[0, 1, 2], [1, 2, 3]]),
    values=torch.ones(3),
    size=(num_nodes, num_nodes)
)

# Create edge_split with the required structure
edge_split = {
    'eval_train': {
        'source_node': torch.tensor([0, 1, 2, 3]),
        'target_node': torch.tensor([1, 2, 3, 4]),
        'target_node_neg': torch.tensor([5, 6, 7, 5, 6, 7, 5, 6]),
    },
    'valid': {
        'source_node': torch.tensor([2, 3]),
        'target_node': torch.tensor([4, 5]),
        'target_node_neg': torch.tensor([6, 7, 6, 7]),
    },
    'test': {
        'source_node': torch.tensor([4, 5]),
        'target_node': torch.tensor([6, 7]),
        'target_node_neg': torch.tensor([0, 1, 0, 1]),
    }
}

# Create evaluator
evaluator = SimpleEvaluator()

ARGS = {
    'model': model,
    'predictor': predictor,
    'x': x,
    'adj_t': adj_t,
    'edge_split': edge_split,
    'evaluator': evaluator,
    'batch_size': batch_size,
    'neg_len': neg_len,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
train Ran
ran; our contract draft was wrong, not the code
Zhuofeng-Li/TEG-Benchmark/GNN/GNN_Link_HitsK.py
code served (permissive licence) · get_code("b9125d103f13768d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.optim import Adam

torch.manual_seed(0)

# Define a minimal GNN model that takes x and adj_t
class SimpleGNN(nn.Module):
    def __init__(self, in_features, hidden_features):
        super().__init__()
        self.linear = nn.Linear(in_features, hidden_features)
    
    def forward(self, x, adj_t):
        # Just return a transformation of x for this test
        return self.linear(x)

# Define a minimal predictor that takes two node embeddings
class SimplePredictor(nn.Module):
    def __init__(self, hidden_features):
        super().__init__()
        self.linear = nn.Linear(hidden_features * 2, 1)
    
    def forward(self, h_i, h_j):
        # Concatenate and predict link probability
        combined = torch.cat([h_i, h_j], dim=-1)
        return torch.sigmoid(self.linear(combined))

# Create small tensors
num_nodes = 5
input_dim = 4
hidden_dim = 3

x = torch.randn(num_nodes, input_dim)
adj_t = torch.eye(num_nodes)  # Simple identity adjacency for testing

# Create model and predictor
model = SimpleGNN(input_dim, hidden_dim)
predictor = SimplePredictor(hidden_dim)

# Create optimizer
optimizer = Adam(list(model.parameters()) + list(predictor.parameters()), lr=0.01)

# Create edge_split with train edges
# Need at least a few edges for the batch to work
edge_split = {
    'train': {
        'edge': torch.tensor([
            [0, 1],
            [1, 2],
            [2, 3],
            [3, 4],
            [0, 2]
        ], dtype=torch.long)
    }
}

batch_size = 2

ARGS = {
    'model': model,
    'predictor': predictor,
    'x': x,
    'adj_t': adj_t,
    'edge_split': edge_split,
    'optimizer': optimizer,
    'batch_size': batch_size
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
train Ran
ran, honoured the contract we drafted
Zhuofeng-Li/TEG-Benchmark/GNN/GNN_Link_HitsK_loader.py
code served (permissive licence) · get_code("93ed7f8d21768471")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.optim import Adam

torch.manual_seed(0)

# Define a simple model that takes x and adj_t and returns node embeddings
class SimpleGNNModel(nn.Module):
    def __init__(self, input_dim, hidden_dim):
        super().__init__()
        self.linear = nn.Linear(input_dim, hidden_dim)
    
    def forward(self, x, adj_t):
        # x: [num_nodes, input_dim]
        # adj_t: sparse adjacency matrix
        # Return: [num_nodes, hidden_dim]
        return self.linear(x)

# Define a simple predictor that takes two node embeddings and predicts edge probability
class SimplePredictor(nn.Module):
    def __init__(self, hidden_dim):
        super().__init__()
        self.linear = nn.Linear(2 * hidden_dim, 1)
    
    def forward(self, h_src, h_dst):
        # h_src: [num_edges, hidden_dim]
        # h_dst: [num_edges, hidden_dim]
        # Return: [num_edges, 1]
        combined = torch.cat([h_src, h_dst], dim=1)
        return torch.sigmoid(self.linear(combined)).squeeze(-1)

# Define a simple batch class
class Batch:
    def __init__(self, x, adj_t, edge_label_index):
        self.x = x
        self.adj_t = adj_t
        self.edge_label_index = edge_label_index
    
    def to(self, device):
        self.x = self.x.to(device)
        self.adj_t = self.adj_t.to(device)
        self.edge_label_index = self.edge_label_index.to(device)
        return self

# Create toy data
num_nodes = 5
input_dim = 4
hidden_dim = 3
num_edges = 2

# Node features: [num_nodes, input_dim]
x = torch.randn(num_nodes, input_dim)

# Adjacency matrix (sparse tensor)
indices = torch.tensor([[0, 1], [1, 2]]).t().contiguous()
values = torch.ones(2)
adj_t = torch.sparse_coo_tensor(indices, values, (num_nodes, num_nodes))

# Edge label index: [2, num_edges] where each column is (src, dst)
edge_label_index = torch.tensor([[0, 1], [2, 3]])

# Create batch
batch = Batch(x, adj_t, edge_label_index)

# Create train loader
train_loader = [batch]

# Create model and predictor
model = SimpleGNNModel(input_dim, hidden_dim)
predictor = SimplePredictor(hidden_dim)

# Create optimizer
optimizer = Adam(list(model.parameters()) + list(predictor.parameters()), lr=0.01)

# Device
device = torch.device('cpu')

ARGS = {
    'model': model,
    'predictor': predictor,
    'train_loader': train_loader,
    'optimizer': optimizer,
    'device': device
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
train Ran
ran; our contract draft was wrong, not the code
Zhuofeng-Li/TEG-Benchmark/GNN/GNN_Link_MRR.py
code served (permissive licence) · get_code("ad21068f2893ab5c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from torch.optim import Adam

torch.manual_seed(0)

# Define a minimal GNN model that takes x and adj_t and returns embeddings
class SimpleGNN(nn.Module):
    def __init__(self, input_dim, output_dim):
        super().__init__()
        self.linear = nn.Linear(input_dim, output_dim)
    
    def forward(self, x, adj_t):
        # Simple linear transformation to simulate GNN
        return self.linear(x)

# Define a minimal predictor that takes two embeddings and returns scores
class SimplePredictor(nn.Module):
    def __init__(self, embedding_dim):
        super().__init__()
        self.linear = nn.Linear(embedding_dim * 2, 1)
    
    def forward(self, src_emb, dst_emb):
        # Concatenate embeddings and pass through linear layer, then sigmoid
        combined = torch.cat([src_emb, dst_emb], dim=1)
        return torch.sigmoid(self.linear(combined))

# Create small tensors
num_nodes = 8
feature_dim = 4
embedding_dim = 3

x = torch.randn(num_nodes, feature_dim)
adj_t = torch.eye(num_nodes)  # Simple adjacency matrix (identity for simplicity)

# Create model and predictor
model = SimpleGNN(feature_dim, embedding_dim)
predictor = SimplePredictor(embedding_dim)

# Create edge split with train edges
num_train_edges = 6
edge_split = {
    'train': {
        'source_node': torch.randint(0, num_nodes, (num_train_edges,)),
        'target_node': torch.randint(0, num_nodes, (num_train_edges,)),
    }
}

# Create optimizer
optimizer = Adam(list(model.parameters()) + list(predictor.parameters()), lr=0.01)

# Batch size for training
batch_size = 2

ARGS = {
    'model': model,
    'predictor': predictor,
    'x': x,
    'adj_t': adj_t,
    'edge_split': edge_split,
    'optimizer': optimizer,
    'batch_size': batch_size,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
train Ran
ran, honoured the contract we drafted
Zhuofeng-Li/TEG-Benchmark/GNN/GNN_Link_MRR_loader.py
code served (permissive licence) · get_code("70a4a68841982fea")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned float, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Small stand-in model that takes x and adj_t and returns node embeddings
class SimpleGNN(nn.Module):
    def __init__(self, input_dim, output_dim):
        super().__init__()
        self.lin = nn.Linear(input_dim, output_dim)
    
    def forward(self, x, adj_t):
        return self.lin(x)

# Small predictor that takes two node embeddings and returns link probability
class SimplePredictor(nn.Module):
    def __init__(self, hidden_dim):
        super().__init__()
        self.lin = nn.Linear(hidden_dim * 2, 1)
    
    def forward(self, src_h, dst_h):
        combined = torch.cat([src_h, dst_h], dim=-1)
        return torch.sigmoid(self.lin(combined)).squeeze(-1)

# Create model and predictor
model = SimpleGNN(input_dim=4, output_dim=8)
predictor = SimplePredictor(hidden_dim=8)

# Create a simple batch class with the attributes the function expects
class SimpleBatch:
    def __init__(self, x, adj_t, edge_label_index):
        self.x = x
        self.adj_t = adj_t
        self.edge_label_index = edge_label_index
    
    def to(self, device):
        self.x = self.x.to(device)
        self.adj_t = self.adj_t.to(device)
        self.edge_label_index = self.edge_label_index.to(device)
        return self

# Create small train_loader with one batch
# 5 nodes with dim 4, 3 edges
x = torch.randn(5, 4)
adj_t = torch.randn(5, 5)
edge_label_index = torch.tensor([[0, 1, 2], [1, 3, 4]], dtype=torch.long)

batch = SimpleBatch(x=x, adj_t=adj_t, edge_label_index=edge_label_index)
train_loader = [batch]

# Create optimizer
optimizer = torch.optim.Adam(
    list(model.parameters()) + list(predictor.parameters()),
    lr=0.01
)

device = 'cpu'

ARGS = {
    'model': model,
    'predictor': predictor,
    'train_loader': train_loader,
    'optimizer': optimizer,
    'device': device,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
gen_loader Not yet run
unverified (harvested, no recorded run)
Zhuofeng-Li/TEG-Benchmark/GNN/GNN_Link_MRR_loader.py
code served (permissive licence) · get_code("4f6f2bae5c0ee660")
How it ran

no invocation recorded.

split_time Not yet run
unverified (harvested, no recorded run)
Zhuofeng-Li/TEG-Benchmark/GNN/model/Dataloader.py
code served (permissive licence) · get_code("d9a1942021124664")
How it ran

no invocation recorded.

train Not yet run
unverified (harvested, no recorded run)
Zhuofeng-Li/TEG-Benchmark/GNN/GNN_Node.py
code served (permissive licence) · get_code("4f3238930893cc90")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 11 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Text-Attributed Graphs (TAGs) augment graph structures with natural language descriptions, facilitating detailed depictions of data and their interconnections across various real-world settings. However, existing TAG datasets predominantly feature textual information only at the nodes, with edges typically represented by mere binary or categorical attributes. This lack of rich textual edge annotations significantly limits the exploration of contextual relationships between entities, hindering deeper insights into graph-structured data. To address this gap, we introduce Textual-Edge Graphs Datasets and Benchmark (TEG-DB), a comprehensive and diverse collection of benchmark textual-edge datasets featuring rich textual descriptions on nodes and edges. The TEG-DB datasets are large-scale and encompass a wide range of domains, from citation networks to social networks. In addition, we conduct extensive benchmark experiments on TEG-DB to assess the extent to which current techniques, including pre-trained language models (PLMs), graph neural networks (GNNs), proposed novel entangled GNNs and their combinations, can utilize textual node and edge information. Our goal is to elicit advancements in textual-edge graph research, specifically in developing methodologies that exploit rich textual node and edge descriptions to enhance graph analysis and provide deeper insights into complex real-world networks. The entire TEG-DB project is publicly accessible as an open-source repository on Github, accessible at https://github.com/Zhuofeng-Li/TEG-Benchmark.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2406.10310")
get_code_for_paper("2406.10310")
have("2406.10310")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2406.10310.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2406.10310.svg)](https://syntology.ai/paper/2406.10310)

Connect an agent — have() is free.