isonetpp_loader.py

# isonetpp_loader.py
from __future__ import annotations
import os
from typing import Optional, Dict
from huggingface_hub import hf_hub_download

from subiso_dataset import (
    SubgraphIsomorphismDataset,
    TRAIN_MODE, VAL_MODE, TEST_MODE, BROAD_TEST_MODE,
    GMN_DATA_TYPE, PYG_DATA_TYPE
)

# Normalize names users pass ("aids" or "aids240k" → stored names are aids240k)
def _normalize_name(name: str) -> str:
    if name.endswith("240k") or name.endswith("80k"):
        return name
    # assume large dataset default = 240k
    return name + "240k"

def _folder(dataset_size: str) -> str:
    return "small_dataset" if dataset_size == "small" else "large_dataset"

def _ensure_paths(
    repo_id: str,
    mode: str,
    dataset_name: str,
    dataset_size: str,
    local_root: Optional[str] = None,
) -> Dict[str, str]:

    dataset_name = _normalize_name(dataset_name)
    folder = _folder(dataset_size)              # "large_dataset" or "small_dataset"
    prefix = "test" if "test" in mode.lower() else mode
    pairs = "80k" if dataset_size == "small" else "240k"

    query_fname  = f"{prefix}_{dataset_name}{pairs}_query_subgraphs.pkl"
    rel_fname    = f"{prefix}_{dataset_name}{pairs}_rel_nx_is_subgraph_iso.pkl"
    corpus_fname = f"{dataset_name}{pairs}_corpus_subgraphs.pkl"

    repo_query_path  = f"{folder}/splits/{mode}/{query_fname}"
    repo_rel_path    = f"{folder}/splits/{mode}/{rel_fname}"
    repo_corpus_path = f"{folder}/corpus/{corpus_fname}"

    kwargs = dict(repo_id=repo_id, repo_type="dataset", local_dir=local_root, local_dir_use_symlinks=False)

    query_path  = hf_hub_download(filename=repo_query_path,  **kwargs)
    rel_path    = hf_hub_download(filename=repo_rel_path,    **kwargs)
    corpus_path = hf_hub_download(filename=repo_corpus_path, **kwargs)

    return {"query": query_path, "rel": rel_path, "corpus": corpus_path}

def load_isonetpp_benchmark(
    repo_id: str = "structlearning/isonetpp-benchmark",
    mode: str = "train",
    dataset_name: str = "aids",
    dataset_size: str = "large",
    batch_size: int = 128,
    data_type: str = "pyg",
    device: Optional[str] = None,
    download_root: Optional[str] = None,
):
    mode_map = {
        "train": TRAIN_MODE, "val": VAL_MODE, "test": TEST_MODE,
        "extra_test_300": BROAD_TEST_MODE, "Extra_test_300": BROAD_TEST_MODE
    }
    mode_norm = mode_map.get(mode, mode)

    paths = _ensure_paths(
        repo_id=repo_id,
        mode=mode_norm,
        dataset_name=dataset_name,
        dataset_size=dataset_size,
        local_root=download_root
    )

    # The downloaded structure is:
    # <cache>/.../<folder>/splits/<mode>/<files>
    # <cache>/.../<folder>/corpus/<files>
    #
    # So dataset_base_path = parent of <folder>
    base_path = os.path.dirname(os.path.dirname(paths["query"]))   # .../<folder>/splits
    dataset_base_path = os.path.dirname(base_path)                 # .../<folder>

    dataset_config = dict(
        mode=mode_norm,
        dataset_name=_normalize_name(dataset_name),
        dataset_size=dataset_size,
        batch_size=batch_size,
        data_type=data_type,
        dataset_base_path=dataset_base_path,
        dataset_path_override=_folder(dataset_size),   # 🟢 critical fix
        experiment=None,
        device=device,
    )

    return SubgraphIsomorphismDataset(**dataset_config)