1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
|
"""Binary format for the Rust -> Python feature handoff (replaces JSON).
One row per 100 m grid cell, stored as a NumPy structured array in a .npy file:
x, y (EPSG:3035 cell centre, float64) followed by risk_model.FEATURES (float32), little-endian.
No parsing: Python memory-maps the file, Rust writes a fixed header plus raw rows.
The header descr Rust must write (keep in sync with FEATURES):
[('x', '<f8'), ('y', '<f8'), ('elev', '<f4'), ('slope', '<f4'), ('relief', '<f4'), ('hand', '<f4'),
('dist_river', '<f4'), ('rp10', '<f4'), ('rp100', '<f4'), ('rp500', '<f4')]
"""
import io
import numpy as np, pandas as pd
from risk_model import FEATURES
GRID_DTYPE = np.dtype([("x", "<f8"), ("y", "<f8")] + [(f, "<f4") for f in FEATURES])
def to_frame(arr):
"""Structured array -> DataFrame with the columns the model expects."""
if arr.dtype != GRID_DTYPE:
raise ValueError(f"unexpected layout {arr.dtype}, expected {GRID_DTYPE}")
return pd.DataFrame({name: arr[name] for name in GRID_DTYPE.names})
def from_frame(df):
arr = np.empty(len(df), GRID_DTYPE)
for name in GRID_DTYPE.names:
arr[name] = df[name].to_numpy()
return arr
def load(path):
return to_frame(np.load(path, mmap_mode="r"))
def save(df, path):
np.save(path, from_frame(df))
def from_bytes(buf):
"""For an HTTP body (Content-Type: application/octet-stream) instead of a file."""
return to_frame(np.load(io.BytesIO(buf)))
def to_bytes(df):
out = io.BytesIO()
np.save(out, from_frame(df))
return out.getvalue()
|
