File size: 15,932 Bytes
7da2ecb | 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 | """Object-based validation loop."""
from __future__ import annotations
from collections import defaultdict
from dataclasses import dataclass
from datetime import datetime
from typing import Any
import numpy as np
from scipy.ndimage import distance_transform_edt
from tqdm import tqdm
from .clusterers import Clusterer, ModelCluster
from .loaders import CloudLabelLoader, CloudTarget, PredictionProvider
from .metrics import compute_scores
from .utils import dilate, format_dt, km_to_pixels, use_edt_backend
@dataclass
class RawValidationResult:
label_records: list[dict[str, Any]]
model_cluster_records: list[dict[str, Any]]
missing_predictions: list[dict[str, Any]]
class Validator:
"""Validate backtracked cloud labels against clustered model predictions."""
def __init__(
self,
targets: list[CloudTarget],
label_loader: CloudLabelLoader,
prediction_provider: PredictionProvider,
clusterer: Clusterer,
pixel_size_km: float = 2.0,
label_buffer_km: float = 0.0,
model_false_buffer_km: float = 0.0,
leadtime_min: int = 10,
leadtime_max: int = 120,
time_step: int = 10,
buffer_backend: str = "auto",
):
self.targets = targets
self.label_loader = label_loader
self.prediction_provider = prediction_provider
self.clusterer = clusterer
self.pixel_size_km = float(pixel_size_km)
self.label_buffer_pixels = km_to_pixels(label_buffer_km, self.pixel_size_km)
self.model_false_buffer_pixels = km_to_pixels(model_false_buffer_km, self.pixel_size_km)
self.leadtime_min = int(leadtime_min)
self.leadtime_max = int(leadtime_max)
self.time_step = int(time_step)
self.buffer_backend = str(buffer_backend or "auto")
@property
def leadtimes(self) -> list[int]:
return list(range(self.leadtime_max, self.leadtime_min - 1, -self.time_step))
def evaluate_raw(self) -> RawValidationResult:
targets_by_dt: dict[datetime, list[CloudTarget]] = defaultdict(list)
for target in self.targets:
targets_by_dt[target.dt].append(target)
label_records: list[dict[str, Any]] = []
model_cluster_records: list[dict[str, Any]] = []
missing_predictions: list[dict[str, Any]] = []
for dt in tqdm(sorted(targets_by_dt), desc="Validate timesteps", dynamic_ncols=True):
dt_targets = targets_by_dt[dt]
label_arr = self.label_loader.load(dt)
try:
field = self.prediction_provider.load(dt)
except FileNotFoundError as exc:
missing_predictions.append(
{
"time": format_dt(dt),
"reason": "missing_prediction",
"message": str(exc),
"num_labels": len(dt_targets),
"cloud_ids": [target.cloud_id for target in dt_targets],
}
)
for target in dt_targets:
label_records.append(
self._base_label_record(
target,
status="impossible",
matched_cluster_ids=[],
prediction_path=None,
label_pixel_count=None,
reason="missing_prediction",
)
)
continue
if field.data.shape != label_arr.shape:
raise ValueError(
f"Shape mismatch at {format_dt(dt)}: prediction={field.data.shape}, label={label_arr.shape}"
)
clusters = self.clusterer.cluster(field.data, field.valid_mask)
target_masks = self._build_target_masks(label_arr, dt_targets)
target_distance_maps = self._build_target_distance_maps(target_masks)
for target in dt_targets:
label_mask = target_masks[target.cloud_id]
label_distance = target_distance_maps[target.cloud_id]
matched_clusters = self._matched_clusters_for_label(label_mask, label_distance, clusters)
status = "hit" if matched_clusters else "miss"
label_records.append(
self._base_label_record(
target,
status=status,
matched_cluster_ids=[cluster.cluster_id for cluster in matched_clusters],
prediction_path=field.path,
label_pixel_count=int(label_mask.sum()),
reason="matched" if matched_clusters else "no_matching_model_cluster",
)
)
model_cluster_records.extend(
self._count_model_clusters(
dt=dt,
clusters=clusters,
target_masks=target_masks,
target_distance_maps=target_distance_maps,
targets_by_cloud_id={target.cloud_id: target for target in dt_targets},
prediction_path=field.path,
)
)
return RawValidationResult(
label_records=label_records,
model_cluster_records=model_cluster_records,
missing_predictions=missing_predictions,
)
def _build_target_masks(self, label_arr: np.ndarray, targets: list[CloudTarget]) -> dict[str, np.ndarray]:
masks: dict[str, np.ndarray] = {}
for target in targets:
mask = label_arr == target.number
if not np.any(mask):
raise ValueError(f"Backtracked cloud {target.cloud_id} not found in label")
masks[target.cloud_id] = mask
return masks
def _build_target_distance_maps(self, target_masks: dict[str, np.ndarray]) -> dict[str, np.ndarray]:
return {
cloud_id: distance_transform_edt(~mask.astype(bool, copy=False))
for cloud_id, mask in target_masks.items()
if np.any(mask)
}
def _matches_distance(self, cluster_mask: np.ndarray, target_distance: np.ndarray, radius_pixels: int) -> bool:
if not np.any(cluster_mask):
return False
return bool(float(np.min(target_distance[cluster_mask.astype(bool, copy=False)])) <= int(radius_pixels))
def _matched_clusters_for_label(
self,
label_mask: np.ndarray,
label_distance: np.ndarray,
clusters: list[ModelCluster],
) -> list[ModelCluster]:
if use_edt_backend(self.label_buffer_pixels, self.buffer_backend):
return [
cluster
for cluster in clusters
if self._matches_distance(cluster.mask, label_distance, self.label_buffer_pixels)
]
label_match_mask = dilate(label_mask, self.label_buffer_pixels)
return [cluster for cluster in clusters if self._touches(label_match_mask, cluster.mask)]
def _matched_cloud_ids_for_cluster(
self,
cluster_mask: np.ndarray,
target_masks: dict[str, np.ndarray],
target_distance_maps: dict[str, np.ndarray],
) -> list[str]:
if use_edt_backend(self.model_false_buffer_pixels, self.buffer_backend):
return sorted(
cloud_id
for cloud_id, distances in target_distance_maps.items()
if self._matches_distance(cluster_mask, distances, self.model_false_buffer_pixels)
)
cluster_match_mask = dilate(cluster_mask, self.model_false_buffer_pixels)
return sorted(
cloud_id for cloud_id, mask in target_masks.items() if self._touches(cluster_match_mask, mask)
)
def _base_label_record(
self,
target: CloudTarget,
status: str,
matched_cluster_ids: list[int],
prediction_path: str | None,
label_pixel_count: int | None,
reason: str,
) -> dict[str, Any]:
record = target.to_dict()
record.update(
{
"raw_status": status,
"status": status,
"reason": reason,
"matched_cluster_ids": matched_cluster_ids,
"matched_cluster_count": len(matched_cluster_ids),
"prediction_path": prediction_path,
"label_pixel_count": label_pixel_count,
"label_buffer_pixels": self.label_buffer_pixels,
"label_buffer_km": self.label_buffer_pixels * self.pixel_size_km,
}
)
return record
def _count_model_clusters(
self,
dt: datetime,
clusters: list[ModelCluster],
target_masks: dict[str, np.ndarray],
target_distance_maps: dict[str, np.ndarray] | None,
targets_by_cloud_id: dict[str, CloudTarget],
prediction_path: str,
) -> list[dict[str, Any]]:
records: list[dict[str, Any]] = []
if target_distance_maps is None:
target_distance_maps = self._build_target_distance_maps(target_masks)
for cluster in clusters:
matched_cloud_ids = self._matched_cloud_ids_for_cluster(
cluster.mask,
target_masks,
target_distance_maps,
)
is_false = len(matched_cloud_ids) == 0
nearest_label = (
self._nearest_target_label(cluster.mask, target_distance_maps, targets_by_cloud_id)
if is_false
else None
)
records.append(
{
"time": format_dt(dt),
"cluster_id": int(cluster.cluster_id),
"cluster_key": f"{format_dt(dt)}_{cluster.cluster_id}",
"status": "false" if is_false else "model_hit",
"is_false": is_false,
"matched_cloud_ids": matched_cloud_ids,
"matched_cloud_count": len(matched_cloud_ids),
"pixel_count": int(cluster.pixel_count),
"raw_pixel_count": int(cluster.raw_pixel_count),
"model_false_buffer_pixels": self.model_false_buffer_pixels,
"model_false_buffer_km": self.model_false_buffer_pixels * self.pixel_size_km,
"assigned_false_cloud_id": nearest_label["cloud_id"] if nearest_label else None,
"assigned_false_leadtime": nearest_label["leadtime"] if nearest_label else None,
"assigned_false_distance_pixels": nearest_label["distance_pixels"] if nearest_label else None,
"assigned_false_distance_km": nearest_label["distance_km"] if nearest_label else None,
"prediction_path": prediction_path,
}
)
return records
def _nearest_target_label(
self,
cluster_mask: np.ndarray,
target_distance_maps: dict[str, np.ndarray],
targets_by_cloud_id: dict[str, CloudTarget],
) -> dict[str, Any] | None:
if not np.any(cluster_mask) or not target_distance_maps:
return None
best: dict[str, Any] | None = None
cluster_mask = cluster_mask.astype(bool, copy=False)
for cloud_id in sorted(target_distance_maps):
distances = target_distance_maps[cloud_id]
distance_pixels = float(np.min(distances[cluster_mask]))
target = targets_by_cloud_id[cloud_id]
candidate = {
"cloud_id": cloud_id,
"leadtime": int(target.leadtime),
"distance_pixels": distance_pixels,
"distance_km": distance_pixels * self.pixel_size_km,
}
if best is None or distance_pixels < best["distance_pixels"]:
best = candidate
return best
@staticmethod
def _touches(mask_a: np.ndarray, mask_b: np.ndarray) -> bool:
return bool(np.any(mask_a & mask_b))
def apply_leadtime_mode(self, raw_label_records: list[dict[str, Any]], leadtime_mode: str) -> list[dict[str, Any]]:
if leadtime_mode == "exact":
return [dict(record, status=record["raw_status"], leadtime_mode="exact") for record in raw_label_records]
if leadtime_mode != "accumulate":
raise ValueError(f"Unsupported leadtime_mode: {leadtime_mode}")
records = [dict(record, leadtime_mode="accumulate") for record in raw_label_records]
by_mature: dict[str, list[dict[str, Any]]] = defaultdict(list)
for record in records:
by_mature[record["mature_id"]].append(record)
for mature_records in by_mature.values():
possible = [record for record in mature_records if record["raw_status"] != "impossible"]
hit_leadtimes = [int(record["leadtime"]) for record in possible if record["raw_status"] == "hit"]
first_hit_leadtime = max(hit_leadtimes) if hit_leadtimes else None
for record in mature_records:
if record["raw_status"] == "impossible":
record["status"] = "impossible"
record["accumulate_first_hit_leadtime"] = first_hit_leadtime
elif first_hit_leadtime is not None and int(record["leadtime"]) <= first_hit_leadtime:
record["status"] = "hit"
record["reason"] = "accumulated_from_first_hit"
record["accumulate_first_hit_leadtime"] = first_hit_leadtime
else:
record["status"] = "miss"
record["accumulate_first_hit_leadtime"] = first_hit_leadtime
return records
def summarize(
self,
label_records: list[dict[str, Any]],
model_cluster_records: list[dict[str, Any]],
) -> dict[str, Any]:
hits = sum(1 for record in label_records if record["status"] == "hit")
misses = sum(1 for record in label_records if record["status"] == "miss")
impossible = sum(1 for record in label_records if record["status"] == "impossible")
falses = sum(1 for record in model_cluster_records if record["is_false"])
model_hits = sum(1 for record in model_cluster_records if not record["is_false"])
leadtime_metrics: dict[int, dict[str, Any]] = {}
for leadtime in self.leadtimes:
lt_records = [record for record in label_records if int(record["leadtime"]) == leadtime]
lt_hits = sum(1 for record in lt_records if record["status"] == "hit")
lt_misses = sum(1 for record in lt_records if record["status"] == "miss")
lt_impossible = sum(1 for record in lt_records if record["status"] == "impossible")
lt_falses = sum(
1
for record in model_cluster_records
if record["is_false"] and record.get("assigned_false_leadtime") == leadtime
)
lt_scores = compute_scores(lt_hits, lt_misses, lt_falses)
lt_scores.update(
{
"impossible": int(lt_impossible),
"valid_labels": int(lt_hits + lt_misses),
}
)
leadtime_metrics[int(leadtime)] = lt_scores
scores = compute_scores(hits, misses, falses)
scores.update(
{
"valid_labels": int(hits + misses),
"impossible": int(impossible),
"total_input_labels": int(len(label_records)),
"model_hit_clusters": int(model_hits),
"false_model_clusters": int(falses),
"total_model_clusters": int(len(model_cluster_records)),
}
)
return {
"total": scores,
"leadtime": leadtime_metrics,
}
|