import duckdb
import torch
from torch_geometric.data import Data
from torch_geometric.data.data import DataEdgeAttr, DataTensorAttr
from torch_geometric.data.storage import BaseStorage, EdgeStorage, GlobalStorage
from torch_geometric.utils import is_undirected

from paths import DB_PATH, GRAPH_PATH

PATH = GRAPH_PATH

# torch>=2.6 defaults to weights_only=True; allowlist PyG's classes to keep it on.
torch.serialization.add_safe_globals(
    [Data, DataEdgeAttr, DataTensorAttr, BaseStorage, EdgeStorage, GlobalStorage]
)
data = torch.load(PATH, weights_only=True)

print(data)
print("validate():", data.validate())
# is_undirected with edge_attr also requires the attrs to match on reversed pairs
print("is_undirected (attr-aware):",
      is_undirected(data.edge_index, data.edge_attr, num_nodes=data.num_nodes))

# Round-trip three edges back against the source table.
con = duckdb.connect(DB_PATH, read_only=True)
print("\nround-trip check (saved value vs. table value):")
for i in (0, 1_000_000, 5_418_417):
    s, t = data.edge_index[:, i].tolist()
    a, b = data.node_id[s], data.node_id[t]
    saved = data.edge_attr[i].tolist()
    row = con.execute(
        """SELECT OddsRatio_log2, ChiTestsPValue FROM edge_ML_expected_ge5
           WHERE (start_id = ? AND end_id = ?) OR (start_id = ? AND end_id = ?)""",
        [a, b, b, a],
    ).fetchone()
    print(f"  col {i}: {a} <-> {b}")
    print(f"    saved={saved}")
    print(f"    table={list(row)}  match={saved == list(row)}")