Datasets:
File size: 18,484 Bytes
d922ac9 | 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 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 | """Merge quarter-cell tables into 5 km windows at a 2.5 km stride.
Reads the ``quarter-v0.1`` tables one band group at a time and writes the
``terrain-v0.3`` window table. No raster is touched and nothing is downloaded:
every number here comes out of the quarter tables, which is the point of
storing additive statistics in the first place.
Each 5 km cell anchors four windows:
=========== =======================================================
offset_id what it covers
=========== =======================================================
0 the cell itself, reproducing the published 5 km values
1 its eastern half and the next column's western half
2 its southern half and band - 1's (southern neighbour) northern half
3 both, the diagonal step
=========== =======================================================
Band numbers run SOUTH to NORTH (band 0 touches the south pole), so "the next
band south" is ``band - 1``. Offsets 2 and 3 reach into it, and its columns do
not line up with this one's. They use that band's *anchor* quarters - its
northern half, indexed on this band's column edges when it was written - so
the window stays a rectangle. The first band of a group finds them in the
previous group's file.
Usage:
python compose_windows.py --planet earth --quarters DIR --out DIR
"""
from __future__ import annotations
import argparse
import datetime as dt
import json
import math
from pathlib import Path
import numpy as np
import pyarrow as pa
import pyarrow.compute as pc
import pyarrow.parquet as pq
import quarter_cells as qc
import terrain_features as tf
from planetary_grid import PLANETS, PlanetBandGrid
FLOAT_COLUMNS = [name for name in tf.FLOAT_FEATURES]
EARTH_EXTRA = ('ns_samples_averaged', 'water_mask_frac')
def scales_for(planet: str):
"""``(lon_scale(lat), lat_scale(lat))`` in metres per degree.
Imported from the drivers rather than restated, so the window table cannot
drift away from the geometry the quarters were measured in.
"""
if planet == 'mars':
radius = PLANETS['mars']['a_m']
return (lambda lat: math.radians(1.0) * radius * np.cos(np.deg2rad(lat)),
lambda lat: math.radians(1.0) * radius)
import earth_terrain as et
return (lambda lat: math.radians(1.0) * et.prime_vertical_radius(lat)
* np.cos(np.deg2rad(lat)),
lambda lat: math.radians(1.0) * float(et.meridional_radius(lat)))
def empty_part(n: int, columns, has_earth: bool) -> dict:
"""A quarter that holds nothing, for cells the extractor never wrote."""
part = {name: np.zeros(n, dtype=np.float64) for name in columns}
for name in ('elev_min', 'elev_max', 'slope_min', 'slope_max'):
part[name] = np.full(n, np.nan)
part['elev_hist_level'] = np.full(n, qc.LEVEL_MIN, dtype=np.int64)
part['elev_hist_first'] = np.zeros(n, dtype=np.int64)
part['slope_hist_level'] = np.full(n, qc.LEVEL_MIN, dtype=np.int64)
part['slope_hist_first'] = np.zeros(n, dtype=np.int64)
for name in qc.QUARTER_LIST_COLUMNS:
part[name] = np.zeros((n, qc.N_BINS), dtype=np.int64)
return part
def scatter(table: pa.Table, n_cells: int, has_earth: bool) -> dict:
"""Dense per-column arrays for one (qrow, qcol) slice of one band."""
columns = list(qc.SCALARS) + (list(qc.EARTH_SCALARS) if has_earth else [])
columns += [f'm_{name}' for name in qc.MOMENTS]
columns += ['plane_z0', 'origin_lon', 'origin_lat', 'lat_scale']
part = empty_part(n_cells, columns, has_earth)
if table.num_rows == 0:
return part
index = table['column'].to_numpy()
for name in columns:
if name in table.column_names:
part[name][index] = table[name].to_numpy(zero_copy_only=False)
for name in qc.QUARTER_INT_COLUMNS:
part[name][index] = table[name].to_numpy()
for name in qc.QUARTER_LIST_COLUMNS:
flat = np.asarray(table[name].combine_chunks().flatten().to_numpy(
zero_copy_only=False), dtype=np.int64)
part[name][index] = flat.reshape(table.num_rows, qc.N_BINS)
return part
def split_band(table: pa.Table, band: int, n_cells: int, has_earth: bool):
"""The band's own four quarter slices, keyed by ``(qrow, qcol)``."""
mask = (table['band'].to_numpy() == band) & (table['anchor_band'].to_numpy() == band)
rows = table.filter(pa.array(mask))
qrow, qcol = rows['qrow'].to_numpy(), rows['qcol'].to_numpy()
return {(r, c): scatter(rows.filter(pa.array((qrow == r) & (qcol == c))),
n_cells, has_earth)
for r in (0, 1) for c in (0, 1)}
def anchor_slices(table: pa.Table, band: int, n_cells: int, has_earth: bool):
"""The band to the SOUTH's northern quarters, already on this band's columns.
Bands are numbered south to north, so that is ``band - 1``: its northern
half is the strip directly below this band's southern half.
"""
mask = ((table['band'].to_numpy() == band - 1)
& (table['anchor_band'].to_numpy() == band))
rows = table.filter(pa.array(mask))
qcol = rows['qcol'].to_numpy()
return {c: scatter(rows.filter(pa.array(qcol == c)), n_cells, has_earth)
for c in (0, 1)}
def roll(part: dict) -> dict:
"""Shift a per-column array one column west, wrapping at +-180 degrees."""
return {k: np.roll(v, -1, axis=0) for k, v in part.items()}
def compose_band(own, anchor, *, band: int, n_cells: int, lon_step: float,
center_lat: float, south_lat: float, lat_scale, has_partner: bool):
"""Every window anchored on this band, as one record per offset.
``anchor`` holds band - 1's northern quarters on this band's columns, and
``has_partner`` says whether that band was measured at all; offsets 2 and
3 are only composed when it was.
"""
cells = np.arange(n_cells)
west = -180.0 + lon_step * (cells + 0.5)
east = -180.0 + lon_step * (cells + 1.0)
here = np.full(n_cells, float(lat_scale(center_lat)))
boundary = np.full(n_cells, float(lat_scale(south_lat)))
plans = {
0: ([own[(0, 0)], own[(0, 1)], own[(1, 0)], own[(1, 1)]],
(west, np.full(n_cells, center_lat), here)),
1: ([own[(0, 1)], own[(1, 1)], roll(own[(0, 0)]), roll(own[(1, 0)])],
(east, np.full(n_cells, center_lat), here)),
}
if has_partner:
# this band's southern half over band - 1's northern half, both on
# this band's column edges, referenced at the boundary between them
plans[2] = ([own[(1, 0)], own[(1, 1)], anchor[0], anchor[1]],
(west, np.full(n_cells, south_lat), boundary))
plans[3] = ([own[(1, 1)], roll(own[(1, 0)]), anchor[1], roll(anchor[0])],
(east, np.full(n_cells, south_lat), boundary))
out = {}
for offset_id, (parts, reference) in plans.items():
out[offset_id] = qc.compose_windows_array(parts, reference)
return out
def build_table(records, planet: str, provenance: dict) -> pa.Table:
"""One Parquet table for a band group, empty windows dropped."""
prefix = PLANETS[planet]['prefix']
bands, columns, offsets, rows = [], [], [], []
for band, offset_id, record in records:
# A window with no valid sample measures nothing. The grid and terrain
# tables already record that a cell is open ocean, so writing four
# all-NaN rows per such cell would roughly treble the Earth table to
# say nothing new.
keep = np.asarray(record['n_valid_samples_200m']) > 0
if not keep.any():
continue
index = np.flatnonzero(keep)
bands.append(np.full(index.size, band, dtype=np.int32))
columns.append(index.astype(np.int32))
offsets.append(np.full(index.size, offset_id, dtype=np.int8))
rows.append({k: np.asarray(v)[index] for k, v in record.items()
if np.ndim(v) == 1})
if not rows:
bands = [np.zeros(0, dtype=np.int32)]
columns = [np.zeros(0, dtype=np.int32)]
offsets = [np.zeros(0, dtype=np.int8)]
rows = [{}]
band_array = np.concatenate(bands)
column_array = np.concatenate(columns)
offset_array = np.concatenate(offsets)
def gather(name, dtype):
parts = [r[name] for r in rows if name in r]
return (np.concatenate(parts).astype(dtype) if parts
else np.zeros(0, dtype=dtype))
data = {
'grid_version': tf._constant(tf.GRID_VERSION, band_array.size),
'window_id': pa.array([qc.window_id(prefix, int(b), int(c), int(o))
for b, c, o in zip(band_array, column_array, offset_array)]),
'region_id': pa.array([f'{prefix}_{int(b):04d}_{int(c):05d}'
for b, c in zip(band_array, column_array)]),
'planet': tf._constant(planet, band_array.size),
'band': pa.array(band_array),
'column': pa.array(column_array),
'offset_id': pa.array(offset_array),
'offset_row': pa.array(np.array([qc.OFFSETS[int(o)][0] for o in offset_array],
dtype=np.int8)),
'offset_col': pa.array(np.array([qc.OFFSETS[int(o)][1] for o in offset_array],
dtype=np.int8)),
}
names = list(FLOAT_COLUMNS)
if planet == 'earth':
names += list(EARTH_EXTRA)
for name in names:
data[name] = pa.array(gather(name, np.float64).astype(np.float32))
for name in ('n_native_samples', 'n_valid_native_samples'):
data[name] = pa.array(gather(name, np.int64))
data['feature_version'] = tf._constant(qc.WINDOW_VERSION, band_array.size)
data['quarter_version'] = tf._constant(qc.QUARTER_VERSION, band_array.size)
data['code_version'] = tf._constant(provenance['code_version'], band_array.size)
# Two hashes, because two programs made these numbers: the extractor that
# measured the quarters and the composer that merged them.
data['quarter_code_version'] = tf._constant(
provenance.get('quarter_code_version', ''), band_array.size)
data['computed_at'] = tf._constant(provenance['computed_at'], band_array.size)
table = pa.table(data)
return table.replace_schema_metadata({
'feature_version': qc.WINDOW_VERSION,
'quarter_version': qc.QUARTER_VERSION,
'grid_version': tf.GRID_VERSION,
'exact_columns': json.dumps(list(qc.EXACT_COLUMNS)),
'approximate_columns': json.dumps(list(qc.APPROXIMATE_COLUMNS)),
'provenance': json.dumps(provenance, sort_keys=True),
})
def group_path(out_dir: Path, group: int) -> Path:
return Path(out_dir) / qc.WINDOW_VERSION / f'band_group={group:03d}.parquet'
def quarter_group_path(quarter_dir: Path, group: int) -> Path:
return Path(quarter_dir) / f'band_group={group:03d}.parquet'
def covered_bands(path: Path) -> set[int]:
"""Every band a quarter file's run covered, open-ocean bands included.
Earth stores that list in the file metadata, where ``--resume`` reads it.
A file without it says nothing, and callers fall back to the bands that
actually wrote rows.
"""
try:
metadata = pq.read_schema(path).metadata or {}
return {int(b) for b in json.loads(metadata[b'bands'].decode())}
except (KeyError, ValueError, TypeError):
return set()
def quarter_versions(table: pa.Table) -> list[str]:
"""The distinct, non-empty ``quarter_code_version`` values of a table."""
if 'quarter_code_version' not in table.column_names or table.num_rows == 0:
return []
return sorted(v for v in pc.unique(table['quarter_code_version']).to_pylist() if v)
def compose_group(planet: str, grid: PlanetBandGrid, quarter_dir: Path,
group: int, provenance: dict) -> tuple[pa.Table, dict]:
path = quarter_group_path(quarter_dir, group)
table = pq.read_table(path)
problems = qc.audit_quarter_table(table, str(path))
if problems:
raise SystemExit('refusing to compose:\n ' + '\n '.join(problems))
n_bands = len(grid.column_counts)
present = set(table['band'].to_numpy().tolist())
# Every band the run covered, not just those that wrote a row: a band of
# open ocean still bounds the southward windows of the band above it.
covered = covered_bands(path) | present
own_bands = sorted(b for b in covered
if b // tf.BANDS_PER_GROUP == group and 0 <= b < n_bands)
# Bands run south to north, so the first band of a group pairs with the
# LAST band of the previous group for its southward windows. Only that
# band's anchor rows are wanted, and a quarter file runs to hundreds of
# megabytes, so the filter goes into the read. Without that file the
# first band's offsets 2 and 3 are skipped, and counted.
previous = quarter_group_path(quarter_dir, group - 1)
if own_bands and own_bands[0] > 0 and group > 0 and previous.exists():
first = own_bands[0]
spill = pq.read_table(previous, filters=[('band', '=', first - 1),
('anchor_band', '=', first)])
covered |= covered_bands(previous) | set(spill['band'].to_numpy().tolist())
if spill.num_rows:
table = pa.concat_tables([table, spill], promote_options='default')
versions = quarter_versions(table)
problems = qc.audit_quarter_versions(versions)
if problems:
raise SystemExit(f'refusing to compose group {group:03d}:\n '
+ '\n '.join(problems))
stamped = dict(provenance, quarter_code_version=versions[0] if versions else '')
stats = {'group': group, 'bands': len(own_bands),
'quarter_code_version': stamped['quarter_code_version'],
'bands_without_a_southern_partner': 0}
if table.num_rows == 0:
# a header-only file - every band above 84 N on Earth - holds nothing
return build_table([], planet, stamped), stats
_, lat_scale = scales_for(planet)
has_earth = 'n_native_water' in table.column_names
records, skipped = [], 0
for band in own_bands:
n_cells = int(grid.column_counts[band])
cell = grid.cell(band, 0)
has_partner = band > 0 and (band - 1) in covered
if band > 0 and not has_partner:
skipped += 1
own = split_band(table, band, n_cells, has_earth)
anchor = anchor_slices(table, band, n_cells, has_earth)
composed = compose_band(
own, anchor, band=band, n_cells=n_cells, lon_step=360.0 / n_cells,
center_lat=cell['center_lat'], south_lat=cell['lat_min'],
lat_scale=lat_scale, has_partner=has_partner)
for offset_id, record in composed.items():
records.append((band, offset_id, record))
stats['bands_without_a_southern_partner'] = skipped
return build_table(records, planet, stamped), stats
def main(argv=None) -> int:
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument('--planet', choices=('mars', 'earth'), required=True)
parser.add_argument('--quarters', type=Path, required=True,
help='directory holding band_group=NNN.parquet')
parser.add_argument('--out', type=Path, required=True)
parser.add_argument('--groups', type=str, default=None,
help='inclusive START-END over band groups')
parser.add_argument('--resume', action='store_true')
parser.add_argument('--require-all-groups', action='store_true',
help='refuse unless every band group of the planet has a '
'quarter file (a planet-scale compose must not '
'quietly become a partial one)')
args = parser.parse_args(argv)
grid = PlanetBandGrid.build(args.planet)
available = sorted(int(p.stem.split('=')[1])
for p in Path(args.quarters).glob('band_group=*.parquet'))
if args.require_all_groups:
n_groups = -(-len(grid.column_counts) // tf.BANDS_PER_GROUP)
missing = sorted(set(range(n_groups)) - set(available))
if missing:
print(f'refusing to compose: {len(missing)} of {n_groups} band groups '
'have no quarter file: ' + ', '.join(f'{g:03d}' for g in missing),
flush=True)
return 2
if args.groups:
low, high = (int(v) for v in args.groups.split('-'))
available = [g for g in available if low <= g <= high]
if not available:
print('no quarter band groups found', flush=True)
return 2
# One release from one extractor. Checked across every group before any
# window is written: per group, a mix would only surface halfway through.
versions = set()
for group in available:
path = quarter_group_path(args.quarters, group)
if 'quarter_code_version' in pq.read_schema(path).names:
versions.update(quarter_versions(
pq.read_table(path, columns=['quarter_code_version'])))
problems = qc.audit_quarter_versions(versions)
if problems:
print('refusing to compose:\n ' + '\n '.join(problems), flush=True)
return 2
# The window table names the code that composed it, not the code that
# read the DEM: the quarters already carry that, and these two files are
# the only ones whose arithmetic can change a window's numbers.
provenance = {
'code_version': tf.code_version(Path(__file__),
Path(__file__).with_name('quarter_cells.py')),
'computed_at': dt.datetime.now(dt.timezone.utc).replace(
microsecond=0).isoformat().replace('+00:00', 'Z'),
}
total = 0
for group in available:
destination = group_path(args.out, group)
if args.resume and destination.exists():
print(f'group {group:03d}: already done', flush=True)
continue
table, stats = compose_group(args.planet, grid, args.quarters, group,
provenance)
tf.write_atomic(table, destination)
total += table.num_rows
print(f"group {group:03d}: {table.num_rows:,} windows from "
f"{stats['bands']} bands", flush=True)
print(f'done: {total:,} windows in {len(available)} group(s)', flush=True)
return 0
if __name__ == '__main__':
raise SystemExit(main())
|