learn_graph.py

import json
import torch
from torch_geometric.data import HeteroData

# # Load the JSON data
with open("papers.json", "r") as f:
    data = json.load(f)

node_features = []
edges = []

for i, node in enumerate(data):
    # if i > 1000: break
    node_features.append((node['index'], node["embedding"]))
    for reference in node['references']:
        edges.append((node['index'], reference))

# Create tensors for node features and edges
node_features = sorted(node_features, key=lambda x: x[0])
node_features_tensor = torch.tensor([x[1] for x in node_features], dtype=torch.float)
edge_index_tensor = torch.tensor(edges, dtype=torch.long).t().contiguous()

# Create the HeteroData object
graph = HeteroData()
graph['paper'].x = node_features_tensor
graph['paper', 'cites', 'paper'].edge_index = edge_index_tensor


torch.save(graph, "graph_data.pt")

# cited = {}
# for edge in edges:
#     if edge[1] not in cited:
#         cited[edge[1]] = 0
#     cited[edge[1]] += 1
    
# print(cited)

# print(graph)

# # Load the HeteroData object from a file
# loaded_graph = torch.load("graph_data.pt")
# print(loaded_graph)