| """QuakeFlow DAS: Read DAS event waveforms from HuggingFace. |
| |
| Files are downloaded on first access and cached locally. |
| """ |
|
|
| |
| import functools |
|
|
| import h5py |
| import numpy as np |
| from huggingface_hub import hf_hub_download, list_repo_files |
|
|
| try: |
| import torch |
|
|
| _TORCH_AVAILABLE = True |
| except ImportError: |
| _TORCH_AVAILABLE = False |
|
|
| REPO_ID = "AI4EPS/quakeflow_das" |
|
|
|
|
| def read_event(filepath): |
| """Read a single DAS event HDF5 file.""" |
| with h5py.File(filepath, "r") as f: |
| result = {"data": f["data"][:].astype(np.float32)} |
| for key, val in f["data"].attrs.items(): |
| result[key] = val.decode("utf-8", errors="replace") if isinstance(val, bytes) else val |
| return result |
|
|
|
|
| @functools.lru_cache(maxsize=None) |
| def list_h5(subset): |
| """List all .h5 files for a subset from the HuggingFace repo (cached).""" |
| prefix = f"{subset}/data/" |
| return sorted(f for f in list_repo_files(REPO_ID, repo_type="dataset") if f.startswith(prefix) and f.endswith(".h5")) |
|
|
|
|
| def download(repo_path): |
| """Download a file from HuggingFace (cached after first download).""" |
| return hf_hub_download(REPO_ID, repo_path, repo_type="dataset", local_dir=".") |
|
|
|
|
| _base_class = torch.utils.data.Dataset if _TORCH_AVAILABLE else object |
|
|
|
|
| class DASDataset(_base_class): |
| """PyTorch Dataset for DAS events. Downloads files on first access.""" |
|
|
| def __init__(self, subset, max_events=None): |
| self.files = list_h5(subset) |
| if max_events is not None: |
| self.files = self.files[:max_events] |
|
|
| def __len__(self): |
| return len(self.files) |
|
|
| def __getitem__(self, idx): |
| filepath = download(self.files[idx]) |
| return read_event(filepath) |
|
|
|
|
| |
| if __name__ == "__main__": |
| for subset in ["ridgecrest_north", "arcata"]: |
| print(f"\n=== {subset} ===") |
| dataset = DASDataset(subset, max_events=3) |
| for i in range(len(dataset)): |
| event = dataset[i] |
| print(f" {event['event_id']}: shape={event['data'].shape}, mag={event.get('magnitude', 'N/A')}") |
|
|
| |
|
|
