Instructions to use msherry/pakflood-lora with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use msherry/pakflood-lora with PEFT:
Task type is invalid.
- Notebooks
- Google Colab
- Kaggle
Download code/population.py from msherry/pakflood-lora: direct link, hf CLI and curl.
- Browser
- Download file 3.64 kB
-
https://huggingface.co/msherry/pakflood-lora/resolve/main/code/population.py
- Command line
-
hf download hf://msherry/pakflood-lora/code/population.py
-
curl -L -o population.py https://huggingface.co/msherry/pakflood-lora/resolve/main/code/population.py
3.64 kB
| """Affected-population estimate: overlay predicted and official (label) flood pixels on WorldPop. | |
| Usage (from ~/pakflood/code, CPU): | |
| .venv/bin/python pakflood/population.py <pred_dir> <worldpop.tif> [--root data/geoid-flood] [--out pop.csv] | |
| For every <label_id>_pred.tif in pred_dir: reproject the binary flood masks (prediction and label, both | |
| restricted to mapped pixels) onto the WorldPop grid with average resampling, so each ~100 m population | |
| pixel gets its flooded fraction; affected people = sum(fraction x population). Prints per-AOI and total | |
| affected population for prediction vs official outline, flooded area in km2, and the percentage difference. | |
| """ | |
| import argparse | |
| from pathlib import Path | |
| import numpy as np | |
| import pandas as pd | |
| import rasterio | |
| from rasterio.enums import Resampling | |
| from rasterio.warp import reproject, transform_bounds | |
| from rasterio.windows import from_bounds | |
| def flooded_people(mask: np.ndarray, src_transform, src_crs, pop_ds) -> float: | |
| """mask: float32 (H,W) in tile CRS, 1 = flooded. Returns people in flooded fraction of WorldPop pixels.""" | |
| b = transform_bounds(src_crs, pop_ds.crs, *rasterio.transform.array_bounds(mask.shape[0], mask.shape[1], src_transform)) | |
| win = from_bounds(*b, transform=pop_ds.transform).round_offsets().round_lengths() | |
| win = win.intersection(rasterio.windows.Window(0, 0, pop_ds.width, pop_ds.height)) | |
| if win.width <= 0 or win.height <= 0: | |
| return 0.0 | |
| pop = pop_ds.read(1, window=win).astype(np.float64) | |
| pop[(pop == pop_ds.nodata) | ~np.isfinite(pop) | (pop < 0)] = 0.0 | |
| frac = np.zeros(pop.shape, dtype=np.float32) | |
| reproject(mask, frac, src_transform=src_transform, src_crs=src_crs, | |
| dst_transform=pop_ds.window_transform(win), dst_crs=pop_ds.crs, resampling=Resampling.average) | |
| return float((frac * pop).sum()) | |
| def main() -> None: | |
| ap = argparse.ArgumentParser() | |
| ap.add_argument("pred_dir") | |
| ap.add_argument("worldpop") | |
| ap.add_argument("--root", default="data/geoid-flood") | |
| ap.add_argument("--out", default=None) | |
| a = ap.parse_args() | |
| pred_dir, root = Path(a.pred_dir), Path(a.root) | |
| rows = [] | |
| with rasterio.open(a.worldpop) as pop: | |
| for p in sorted(pred_dir.glob("*_pred.tif")): | |
| lid = p.name.replace("_pred.tif", "") | |
| ev = lid.replace("_label", "").rsplit("-", 1)[0] | |
| with rasterio.open(p) as r: | |
| pred = r.read(1) | |
| tr, crs, res = r.transform, r.crs, r.res | |
| with rasterio.open(root / ev / "label" / f"{lid}.tif") as r: | |
| label = r.read(1) | |
| mapped = label != 255 | |
| m_pred = ((pred == 2) & mapped).astype(np.float32) | |
| m_lab = ((label == 2) & mapped).astype(np.float32) | |
| px_km2 = abs(res[0] * res[1]) / 1e6 | |
| rows.append({ | |
| "label_id": lid, "aoi": ev, | |
| "area_pred_km2": float(m_pred.sum() * px_km2), "area_label_km2": float(m_lab.sum() * px_km2), | |
| "pop_pred": flooded_people(m_pred, tr, crs, pop), "pop_label": flooded_people(m_lab, tr, crs, pop), | |
| "pop_mapped": flooded_people(mapped.astype(np.float32), tr, crs, pop), | |
| }) | |
| df = pd.DataFrame(rows) | |
| if a.out: | |
| df.to_csv(a.out, index=False) | |
| agg = df.groupby("aoi")[["area_pred_km2", "area_label_km2", "pop_pred", "pop_label", "pop_mapped"]].sum() | |
| agg.loc["TOTAL"] = agg.sum() | |
| agg["pop_diff_pct"] = 100 * (agg.pop_pred - agg.pop_label) / agg.pop_label.replace(0, np.nan) | |
| pd.set_option("display.width", 160) | |
| print(agg.round(1).to_string()) | |
| if __name__ == "__main__": | |
| main() | |