Loading failures in social networks and combinatorial optimization datasets

[0xZivMor]

Hey,
Unfortunately, I keep getting different errors (all of the following happened multiple times) when downloading different datasets using graphbench:

loader = graphbench.Loader(GRAPHBENCH_ROOT, "socialnetwork") dataset = loader.load() [INFO] Downloading https://zenodo.org/records/14669616/files/graphs.tar.gz -> datasets/graphbench/bluesky/bluesky_graphs/raw/graphs.tar.gz Traceback (most recent call last): File "<stdin>", line 1, in <module> File "miniforge3/envs/gfm/lib/python3.10/site-packages/graphbench/loader.py", line 111, in load data = self._loader(dataset) File "miniforge3/envs/gfm/lib/python3.10/site-packages/graphbench/loader.py", line 135, in _loader train_dataset = BlueSkyDataset(root=self.root, name=dataset_name, pre_transform=self.pre_transform, transform=self.transform, split="train", follower_subgraph=False, cleanup_raw=True,load_preprocessed=True) File "miniforge3/envs/gfm/lib/python3.10/site-packages/graphbench/datasets/bluesky.py", line 279, in __init__ self._prepare() # (i) downloads, unpacks, load data + (ii) timestep handle + (e) subgraph + collate File "miniforge3/envs/gfm/lib/python3.10/site-packages/graphbench/datasets/bluesky.py", line 286, in _prepare _download_and_unpack(self.source, self._raw_dir, Path(self.processed_dir), logger=logger) File "miniforge3/envs/gfm/lib/python3.10/site-packages/graphbench/helpers/download.py", line 26, in _download_and_unpack _stream_download(url, local_path, logger) File "miniforge3/envs/gfm/lib/python3.10/site-packages/graphbench/helpers/download.py", line 48, in _stream_download r.raise_for_status() File "miniforge3/envs/gfm/lib/python3.10/site-packages/requests/models.py", line 1026, in raise_for_status raise HTTPError(http_error_msg, response=self) requests.exceptions.HTTPError: 429 Client Error: Too Many Requests for url: https://zenodo.org/records/14669616/files/graphs.tar.gz

The combinatorial optimization dataset either crashes Python completely (I tried different kernels) or has some weird error if some of the dataset was already downloaded:

loader = graphbench.Loader(GRAPHBENCH_ROOT, "co") ds = loader.load() [INFO] Downloading https://huggingface.co/datasets/log-rwth-aachen/Graphbench_CO/resolve/main/ba_large_mis_labeled.tar.gz -> datasets/graphbench/co/co_ba_large/raw/ba_large_mis_labeled.tar.gz Downloaded and unpacked data to datasets/graphbench/co/co_ba_large/raw [INFO] Saved processed dataset -> datasets/graphbench/co/co_ba_large/processed/[data.pt](http://data.pt/) Done! [INFO] Cleaning up: datasets/graphbench/co/co_ba_large/raw Processing... [INFO] Downloading https://huggingface.co/datasets/log-rwth-aachen/Graphbench_CO/resolve/main/ba_small_mis_labeled.tar.gz -> datasets/graphbench/co/co_ba_small/raw/ba_small_mis_labeled.tar.gz Downloaded and unpacked data to datasets/graphbench/co/co_ba_small/raw [INFO] Saved processed dataset -> datasets/graphbench/co/co_ba_small/processed/[data.pt](http://data.pt/) Done! [INFO] Cleaning up: datasets/graphbench/co/co_ba_small/raw Processing... [INFO] Downloading https://huggingface.co/datasets/log-rwth-aachen/Graphbench_CO/resolve/main/er_large_mis_labeled.tar.gz -> datasets/graphbench/co/co_er_large/raw/er_large_mis_labeled.tar.gz Downloaded and unpacked data to datasets/graphbench/co/co_er_large/raw Killed

And
loader = graphbench.Loader(GRAPHBENCH_ROOT, "co") ds = loader.load() Processing... Traceback (most recent call last): File "<stdin>", line 1, in <module> File "miniforge3/envs/gfm/lib/python3.10/site-packages/graphbench/loader.py", line 111, in load data = self._loader(dataset) File "miniforge3/envs/gfm/lib/python3.10/site-packages/graphbench/loader.py", line 151, in _loader dataset = CODataset(root=self.root, name=dataset_name, pre_transform=self.pre_transform, transform=self.transform, split="train", generate=self.generate) File "miniforge3/envs/gfm/lib/python3.10/site-packages/graphbench/datasets/combinatorial_optimization.py", line 134, in __init__ super().__init__(str(self.algoreas_dir), transform, pre_transform) File "miniforge3/envs/gfm/lib/python3.10/site-packages/torch_geometric/data/in_memory_dataset.py", line 81, in __init__ super().__init__(root, transform, pre_transform, pre_filter, log, File "miniforge3/envs/gfm/lib/python3.10/site-packages/torch_geometric/data/dataset.py", line 115, in __init__ self._process() File "miniforge3/envs/gfm/lib/python3.10/site-packages/torch_geometric/data/dataset.py", line 265, in _process self.process() File "miniforge3/envs/gfm/lib/python3.10/site-packages/graphbench/datasets/combinatorial_optimization.py", line 257, in process self._prepare() File "miniforge3/envs/gfm/lib/python3.10/site-packages/graphbench/datasets/combinatorial_optimization.py", line 191, in _prepare _download_and_unpack(source=self.source, raw_dir=self._raw_dir, processed_dir=self.processed_path, logger=logger) File "miniforge3/envs/gfm/lib/python3.10/site-packages/graphbench/helpers/download.py", line 24, in _download_and_unpack if not processed_dir.exists() or not any(Path(processed_dir).iterdir()): File "miniforge3/envs/gfm/lib/python3.10/pathlib.py", line 1017, in iterdir for name in self._accessor.listdir(self): NotADirectoryError: [Errno 20] Not a directory: 'datasets/graphbench/co/co_ba_large/processed/[data.pt](http://data.pt/)'

[Timo-SH]

Hi,

I already looked a bit into the issues described.

For the bluesky/social media dataset:
This is sadly a known current issue with data obtained from Zenodo. We are currently looking into hosting the data elsewhere with permission of the authors and/or adding a retry in the download method to allow for these types of issues.
In the meantime I can propose to wait a minute and to try the download again.

For CO:
In case data was partially downloaded or the raw directory exists but the data was not processed correctly it is currently advisable to remove the data from the folder (or remove the folder entirely). We are working on a fix which does not require the user to remove any files.
Further, the download of all CO files requires significant RAM as the downloaded datasets are quite large (the er_large_mis dataset is 65GB for example) and loaded into RAM by the loader method. With insufficient RAM the above error occurs as the python process is killed. To download and/or evaluate the datasets separately, you can also name them directly as shown in the datasets.csv file.
With sufficient RAM the process should work as expected.

It would be great if you could try the proposed intermediate solutions and let me know whether everything works as intended.

Best regards
Timo

[ArmanMielke]

The very last error was a bug in CODataset, should be fixed in version 0.1.2.6.