File size: 34,928 Bytes
2a8a916 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 | #!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Evaluate PhaseNet/AI picks against continuous-HDF5 annotation JSON.
Main features
-------------
1. Stream-read huge JSONL auto-pick files, e.g. >40 GB.
2. Build a fast per-station / per-phase / per-second SQLite index for auto picks.
3. Compare human/label picks with auto picks using a configurable TP tolerance.
4. Report P/S recall where labels P/S map to auto Pg/Sg by default.
5. Report travel-time residual distribution within a wider error window.
6. Separate statistics for manual labels only and all labels.
7. Count automatic picks in the SQLite index.
8. Fit Gaussian and Student-t residual models.
9. Plot residual histograms with fitted Gaussian and Student-t PDFs.
Typical usage
-------------
python scripts/evaluate_picks.py \
--auto-jsonl data/picks/phasenet.pick.jsonl \
--label-json data/label/annotations_for_continuous_hdf5.json \
--index-db ~/phasenet.pick.index.sqlite \
--outdir eval_picks/eval_phasenet \
--build-index \
--tp-tol 1.5 \
--err-window 5.0
If the SQLite index already exists, omit --build-index.
"""
from __future__ import annotations
import argparse
import json
import math
import os
import sqlite3
import sys
from collections import Counter, defaultdict
from dataclasses import dataclass
from datetime import datetime, timezone
from pathlib import Path
from typing import Any, Dict, Iterable, List, Optional, Sequence, Tuple
import bisect
import numpy as np
# matplotlib is only needed for --plot. Keep import lazy in plot_results().
try:
from scipy import stats as scipy_stats
except Exception: # pragma: no cover
scipy_stats = None
# -----------------------------
# Time and phase normalization
# -----------------------------
def parse_utc_to_epoch_seconds(value: str) -> float:
"""Parse ISO UTC string to epoch seconds.
Supports strings with or without trailing Z. Naive timestamps are treated as UTC.
"""
if value is None:
raise ValueError("time value is None")
s = str(value).strip()
if s.endswith("Z"):
s = s[:-1] + "+00:00"
dt = datetime.fromisoformat(s)
if dt.tzinfo is None:
dt = dt.replace(tzinfo=timezone.utc)
else:
dt = dt.astimezone(timezone.utc)
return dt.timestamp()
def norm_location(loc: Optional[str]) -> str:
if loc is None or loc == "":
return "--"
return str(loc)
def norm_station_id(station_id: Optional[str], network: Optional[str] = None,
station: Optional[str] = None, location: Optional[str] = None) -> str:
"""Normalize station id to network.station.location, using -- for empty location."""
if station_id:
parts = str(station_id).split(".")
if len(parts) >= 3:
return f"{parts[0]}.{parts[1]}.{norm_location(parts[2])}"
return str(station_id)
return f"{network}.{station}.{norm_location(location)}"
DEFAULT_PHASE_MAP = {
"P": ["Pg"],
"S": ["Sg"],
"Pg": ["Pg"],
"Sg": ["Sg"],
"Pn": ["Pg", "Pn", "P"],
"Sn": ["Sg", "Sn", "S"],
}
def parse_phase_map(text: Optional[str]) -> Dict[str, List[str]]:
"""Parse phase map string like 'P:Pg,Pn;S:Sg,Sn'."""
if not text:
return dict(DEFAULT_PHASE_MAP)
phase_map: Dict[str, List[str]] = {}
for item in text.split(";"):
item = item.strip()
if not item:
continue
left, right = item.split(":", 1)
phase_map[left.strip()] = [x.strip() for x in right.split(",") if x.strip()]
return phase_map
# -----------------------------
# SQLite index for huge JSONL
# -----------------------------
def connect_db(db_path: Path) -> sqlite3.Connection:
"""Open SQLite DB and automatically create its parent directory."""
db_path = Path(db_path).expanduser().resolve()
db_path.parent.mkdir(parents=True, exist_ok=True)
conn = sqlite3.connect(str(db_path))
conn.execute("PRAGMA journal_mode=WAL;")
conn.execute("PRAGMA synchronous=NORMAL;")
conn.execute("PRAGMA temp_store=MEMORY;")
conn.execute("PRAGMA cache_size=-200000;") # about 200 MB
return conn
def init_pick_index(conn: sqlite3.Connection, drop_existing: bool = False) -> None:
cur = conn.cursor()
if drop_existing:
cur.execute("DROP TABLE IF EXISTS auto_picks")
cur.execute(
"""
CREATE TABLE IF NOT EXISTS auto_picks (
id INTEGER PRIMARY KEY AUTOINCREMENT,
station_id TEXT NOT NULL,
network TEXT,
station TEXT,
location TEXT,
phase_name TEXT NOT NULL,
time_epoch REAL NOT NULL,
sec_key INTEGER NOT NULL,
phase_prob REAL,
polarity TEXT,
polarity_prob REAL,
snr REAL,
amplitude REAL,
h5_file TEXT,
raw_json TEXT
)
"""
)
cur.execute("CREATE INDEX IF NOT EXISTS idx_station_phase_sec ON auto_picks(station_id, phase_name, sec_key)")
cur.execute("CREATE INDEX IF NOT EXISTS idx_station_sec ON auto_picks(station_id, sec_key)")
cur.execute("CREATE INDEX IF NOT EXISTS idx_time ON auto_picks(time_epoch)")
conn.commit()
def iter_jsonl(path: Path) -> Iterable[Dict[str, Any]]:
with path.open("r", encoding="utf-8", errors="replace") as f:
for line_no, line in enumerate(f, 1):
line = line.strip()
if not line:
continue
try:
yield json.loads(line)
except json.JSONDecodeError as exc:
print(f"[WARN] skip bad JSONL line {line_no}: {exc}", file=sys.stderr)
def build_auto_pick_index(
auto_jsonl: Path,
db_path: Path,
batch_size: int = 50000,
drop_existing: bool = False,
keep_raw_json: bool = False,
progress_every: int = 200000,
) -> None:
auto_jsonl = Path(auto_jsonl).expanduser().resolve()
db_path = Path(db_path).expanduser().resolve()
if not auto_jsonl.exists():
raise FileNotFoundError(f"auto JSONL not found: {auto_jsonl}")
conn = connect_db(db_path)
init_pick_index(conn, drop_existing=drop_existing)
cur = conn.cursor()
insert_sql = (
"INSERT INTO auto_picks "
"(station_id, network, station, location, phase_name, time_epoch, sec_key, "
"phase_prob, polarity, polarity_prob, snr, amplitude, h5_file, raw_json) "
"VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?, ?)"
)
batch = []
n = 0
skipped = 0
for rec in iter_jsonl(auto_jsonl):
if rec.get("record_type") != "phase_pick":
continue
try:
station_info = rec.get("station_info") or {}
station_id = norm_station_id(
station_info.get("station_id"),
station_info.get("network"),
station_info.get("station"),
station_info.get("location"),
)
phase_name = str(rec.get("phase_name"))
t = parse_utc_to_epoch_seconds(rec.get("phase_time"))
sec_key = int(math.floor(t))
raw_json = json.dumps(rec, ensure_ascii=False) if keep_raw_json else None
batch.append((
station_id,
station_info.get("network"),
station_info.get("station"),
norm_location(station_info.get("location")),
phase_name,
t,
sec_key,
rec.get("phase_prob"),
rec.get("polarity"),
rec.get("polarity_prob"),
rec.get("snr"),
rec.get("amplitude"),
rec.get("h5_file"),
raw_json,
))
except Exception as exc:
skipped += 1
if skipped <= 10:
print(f"[WARN] skip record: {exc}", file=sys.stderr)
continue
if len(batch) >= batch_size:
cur.executemany(insert_sql, batch)
conn.commit()
n += len(batch)
batch.clear()
if n % progress_every < batch_size:
print(f"[INDEX] inserted {n:,} picks, skipped {skipped:,}")
if batch:
cur.executemany(insert_sql, batch)
conn.commit()
n += len(batch)
conn.execute("ANALYZE")
conn.commit()
conn.close()
print(f"[INDEX] done. inserted {n:,} picks, skipped {skipped:,}. db={db_path}")
# -----------------------------
# Label loading
# -----------------------------
@dataclass
class LabelPick:
label_phase: str
label_time_epoch: float
station_id: str
status: str
event_id: Optional[str]
distance_km: Optional[float]
raw: Dict[str, Any]
def iter_label_picks(label_json: Path) -> Iterable[LabelPick]:
with label_json.open("r", encoding="utf-8") as f:
data = json.load(f)
years = data.get("years", {})
for _year_id, year_obj in years.items():
for _day_id, day_obj in (year_obj.get("days") or {}).items():
for event_id, event_obj in (day_obj.get("events") or {}).items():
for station_id0, sta_obj in (event_obj.get("stations") or {}).items():
for p in (sta_obj.get("picks") or []):
try:
station_id = norm_station_id(
p.get("station_id") or station_id0,
p.get("network"),
p.get("station"),
p.get("location"),
)
yield LabelPick(
label_phase=str(p.get("phase")),
label_time_epoch=parse_utc_to_epoch_seconds(p.get("time")),
station_id=station_id,
status=str(p.get("status", "unknown")),
event_id=p.get("event_id") or event_id,
distance_km=p.get("distance_km"),
raw=p,
)
except Exception as exc:
print(f"[WARN] skip label pick in event={event_id}, station={station_id0}: {exc}", file=sys.stderr)
# -----------------------------
# Matching and statistics
# -----------------------------
# -----------------------------
# Waveform coverage index
# -----------------------------
class WaveformCoverageIndex:
"""In-memory coverage index built from the waveform SQLite index.
For each station_key (network.station), stores a sorted list of
(start_epoch, end_epoch) tuples. Coverage queries use bisect for
O(log N) lookup, so the 290 k label-pick loop stays fast.
Station matching ignores the location code, because the waveform index
uses 2-part keys (e.g. 'CI.AVM') while label station_ids are 3-part
(e.g. 'CI.AVM.--').
"""
def __init__(self, waveform_db: Path) -> None:
waveform_db = Path(waveform_db).expanduser().resolve()
if not waveform_db.exists():
raise FileNotFoundError(f"Waveform index DB not found: {waveform_db}")
self._index: Dict[str, List[Tuple[float, float]]] = {}
self._load(waveform_db)
def _load(self, db_path: Path) -> None:
"""Load all waveform segments into memory, merging by station_key."""
conn = sqlite3.connect(str(db_path))
# Load one row per (station_key, start_epoch, end_epoch); channel/location
# don't matter for coverage — any channel present means the station was processed.
sql = """
SELECT station_key, MIN(start_epoch) AS t0, MAX(end_epoch) AS t1
FROM waveform_segments
GROUP BY station_key, CAST(start_epoch / 86400 AS INTEGER)
ORDER BY station_key, t0
"""
raw: Dict[str, List[Tuple[float, float]]] = {}
for row in conn.execute(sql):
key, t0, t1 = str(row[0]), float(row[1]), float(row[2])
raw.setdefault(key, []).append((t0, t1))
conn.close()
# Merge overlapping/adjacent intervals per station
for key, segs in raw.items():
segs.sort()
merged: List[Tuple[float, float]] = []
for t0, t1 in segs:
if merged and t0 <= merged[-1][1] + 1.0: # 1 s gap tolerance
merged[-1] = (merged[-1][0], max(merged[-1][1], t1))
else:
merged.append((t0, t1))
self._index[key] = merged
n_sta = len(self._index)
n_seg = sum(len(v) for v in self._index.values())
print(f"[COVERAGE] loaded {n_sta} stations, {n_seg} merged segments from {db_path.name}")
@staticmethod
def _station_key_from_id(station_id: str) -> str:
"""Extract 'network.station' from a 3-part 'network.station.location' id."""
parts = station_id.split(".")
if len(parts) >= 2:
return f"{parts[0]}.{parts[1]}"
return station_id
def has_coverage(self, station_id: str, time_epoch: float) -> bool:
"""Return True if any waveform segment covers *time_epoch* for this station."""
key = self._station_key_from_id(station_id)
segs = self._index.get(key)
if not segs:
return False
# Find the last segment whose start <= time_epoch
starts = [s[0] for s in segs]
idx = bisect.bisect_right(starts, time_epoch) - 1
if idx < 0:
return False
return segs[idx][1] >= time_epoch
def covered_networks(self) -> List[str]:
return sorted({k.split(".")[0] for k in self._index})
@dataclass
class MatchResult:
subset: str
label_phase: str
station_id: str
event_id: Optional[str]
label_time_epoch: float
matched: bool
auto_phase: Optional[str] = None
auto_time_epoch: Optional[float] = None
residual_s: Optional[float] = None # auto - label
phase_prob: Optional[float] = None
snr: Optional[float] = None
distance_km: Optional[float] = None
has_waveform: Optional[bool] = None # True/False when waveform_db used; None otherwise
def query_nearest_auto_pick(
conn: sqlite3.Connection,
station_id: str,
auto_phases: Sequence[str],
label_time_epoch: float,
search_window_s: float,
min_prob: Optional[float] = None,
) -> Optional[Tuple[str, float, float, Optional[float], Optional[float]]]:
"""Return nearest auto pick as (phase, time_epoch, residual, prob, snr)."""
sec0 = int(math.floor(label_time_epoch - search_window_s))
sec1 = int(math.floor(label_time_epoch + search_window_s))
placeholders = ",".join("?" for _ in auto_phases)
params: List[Any] = [station_id, *auto_phases, sec0, sec1]
prob_clause = ""
if min_prob is not None:
prob_clause = " AND phase_prob >= ?"
params.append(float(min_prob))
sql = f"""
SELECT phase_name, time_epoch, phase_prob, snr
FROM auto_picks
WHERE station_id = ?
AND phase_name IN ({placeholders})
AND sec_key BETWEEN ? AND ?
{prob_clause}
ORDER BY ABS(time_epoch - ?) ASC
LIMIT 1
"""
params.append(float(label_time_epoch))
row = conn.execute(sql, params).fetchone()
if row is None:
return None
phase, t, prob, snr = row
residual = float(t) - float(label_time_epoch)
if abs(residual) > search_window_s:
return None
return phase, float(t), residual, prob, snr
def evaluate(
label_json: Path,
db_path: Path,
outdir: Path,
phase_map: Dict[str, List[str]],
tp_tol: float = 1.5,
err_window: float = 5.0,
min_prob: Optional[float] = None,
waveform_db: Optional[Path] = None,
) -> Tuple[List[MatchResult], Dict[str, Any]]:
outdir = Path(outdir).expanduser().resolve()
outdir.mkdir(parents=True, exist_ok=True)
db_path = Path(db_path).expanduser().resolve()
if not db_path.exists():
raise FileNotFoundError(
f"SQLite index DB not found: {db_path}. "
"Run once with --build-index, or check --index-db."
)
conn = connect_db(db_path)
# Optional waveform coverage filter
cov_index: Optional[WaveformCoverageIndex] = None
if waveform_db is not None:
cov_index = WaveformCoverageIndex(Path(waveform_db))
results: List[MatchResult] = []
label_counter = Counter()
label_status_counter = Counter()
for lab in iter_label_picks(label_json):
if lab.label_phase not in phase_map:
continue
label_counter[lab.label_phase] += 1
label_status_counter[(lab.status, lab.label_phase)] += 1
# Waveform coverage check
has_waveform: Optional[bool] = None
if cov_index is not None:
has_waveform = cov_index.has_coverage(lab.station_id, lab.label_time_epoch)
auto_phases = phase_map[lab.label_phase]
# Search by wider window for residual distribution. TP uses tp_tol later.
nearest = query_nearest_auto_pick(
conn, lab.station_id, auto_phases, lab.label_time_epoch,
search_window_s=err_window, min_prob=min_prob,
)
for subset in ("all", "manual", "automatic"):
if subset == "manual" and lab.status != "manual":
continue
if subset == "automatic" and lab.status != "automatic":
continue
if nearest is None:
results.append(MatchResult(
subset=subset, label_phase=lab.label_phase, station_id=lab.station_id,
event_id=lab.event_id, label_time_epoch=lab.label_time_epoch,
matched=False, distance_km=lab.distance_km,
has_waveform=has_waveform,
))
else:
auto_phase, auto_time, residual, prob, snr = nearest
results.append(MatchResult(
subset=subset, label_phase=lab.label_phase, station_id=lab.station_id,
event_id=lab.event_id, label_time_epoch=lab.label_time_epoch,
matched=abs(residual) <= tp_tol,
auto_phase=auto_phase, auto_time_epoch=auto_time,
residual_s=residual, phase_prob=prob, snr=snr,
distance_km=lab.distance_km, has_waveform=has_waveform,
))
conn.close()
summary = summarize_results(results, label_counter, label_status_counter, tp_tol, err_window)
summary["auto_pick_count"] = get_auto_counts(db_path, min_prob=min_prob, phase_map=phase_map)
write_outputs(results, summary, outdir)
return results, summary
def fit_student_t(residuals: np.ndarray) -> Dict[str, Optional[float]]:
if residuals.size < 3 or scipy_stats is None:
return {"df": None, "loc": None, "scale": None}
try:
df, loc, scale = scipy_stats.t.fit(residuals)
return {"df": float(df), "loc": float(loc), "scale": float(scale)}
except Exception:
return {"df": None, "loc": None, "scale": None}
def fit_gaussian(residuals: np.ndarray) -> Dict[str, Optional[float]]:
"""Maximum-likelihood Gaussian fit for residuals."""
if residuals.size < 2:
return {"mean": None, "std_mle": None, "std_unbiased": None}
return {
"mean": float(np.mean(residuals)),
"std_mle": float(np.std(residuals, ddof=0)),
"std_unbiased": float(np.std(residuals, ddof=1)),
}
def get_auto_counts(
db_path: Path,
min_prob: Optional[float] = None,
phase_map: Optional[Dict[str, List[str]]] = None,
) -> Dict[str, Any]:
"""Count automatic picks in the SQLite index.
Returns original automatic phase counts, e.g. Pg/Sg, plus optional counts
mapped to label phases, e.g. P->Pg and S->Sg. If min_prob is set, the
counts are computed after applying phase_prob >= min_prob.
"""
conn = connect_db(db_path)
where = ""
params: List[Any] = []
if min_prob is not None:
where = "WHERE phase_prob >= ?"
params.append(float(min_prob))
rows = conn.execute(
f"SELECT phase_name, COUNT(*) FROM auto_picks {where} GROUP BY phase_name ORDER BY phase_name",
params,
).fetchall()
by_auto_phase = {str(ph): int(c) for ph, c in rows}
total = int(sum(by_auto_phase.values()))
mapped: Dict[str, int] = {}
if phase_map:
for label_phase, auto_phases in phase_map.items():
mapped[label_phase] = int(sum(by_auto_phase.get(ap, 0) for ap in auto_phases))
conn.close()
return {
"filter": {"min_prob": min_prob},
"total": total,
"by_auto_phase": by_auto_phase,
"mapped_to_label_phase": mapped,
}
def summarize_results(
results: List[MatchResult],
label_counter: Counter,
label_status_counter: Counter,
tp_tol: float,
err_window: float,
) -> Dict[str, Any]:
summary: Dict[str, Any] = {
"tp_tolerance_s": tp_tol,
"residual_window_s": err_window,
"label_phase_count_all_status": dict(label_counter),
"label_phase_count_by_status": {f"{k[0]}:{k[1]}": v for k, v in label_status_counter.items()},
"subsets": {},
}
for subset in sorted({r.subset for r in results}):
subset_results = [r for r in results if r.subset == subset]
phases = sorted({r.label_phase for r in subset_results})
subset_summary: Dict[str, Any] = {}
for ph in phases:
ph_results = [r for r in subset_results if r.label_phase == ph]
n_label = len(ph_results)
n_tp = sum(1 for r in ph_results if r.matched)
residuals = np.array([r.residual_s for r in ph_results if r.residual_s is not None and abs(r.residual_s) <= err_window], dtype=float)
tp_residuals = np.array([r.residual_s for r in ph_results if r.matched and r.residual_s is not None], dtype=float)
# Waveform-coverage-corrected recall: denominator is only labels that
# have waveform data available (has_waveform=True). Labels where
# has_waveform is None (waveform_db not supplied) are excluded from
# the covered subset so the field stays None rather than 0/1.
cov_results = [r for r in ph_results if r.has_waveform is True]
n_label_cov = len(cov_results)
n_tp_cov = sum(1 for r in cov_results if r.matched)
# Distinguish "waveform_db not used" (all None) from "used but 0 covered"
use_cov = any(r.has_waveform is not None for r in ph_results)
phase_summary = {
"n_label": int(n_label),
"n_matched_within_tp_tol": int(n_tp),
"recall": float(n_tp / n_label) if n_label else None,
# Coverage-corrected metrics (only present when --waveform-db is used)
"n_label_with_waveform": int(n_label_cov) if use_cov else None,
"n_tp_with_waveform": int(n_tp_cov) if use_cov else None,
"recall_covered": float(n_tp_cov / n_label_cov) if (use_cov and n_label_cov) else None,
"n_residual_within_err_window": int(residuals.size),
"residual_mean_s": float(np.mean(residuals)) if residuals.size else None,
"residual_std_s": float(np.std(residuals, ddof=1)) if residuals.size > 1 else None,
"residual_median_s": float(np.median(residuals)) if residuals.size else None,
"residual_abs_p90_s": float(np.percentile(np.abs(residuals), 90)) if residuals.size else None,
"residual_abs_p95_s": float(np.percentile(np.abs(residuals), 95)) if residuals.size else None,
"tp_residual_std_s": float(np.std(tp_residuals, ddof=1)) if tp_residuals.size > 1 else None,
"gaussian_fit_all_within_err_window": fit_gaussian(residuals),
"student_t_fit_all_within_err_window": fit_student_t(residuals),
}
subset_summary[ph] = phase_summary
# Combined P+S / Pg+Sg style summary
n_label_all = len(subset_results)
n_tp_all = sum(1 for r in subset_results if r.matched)
cov_all = [r for r in subset_results if r.has_waveform is True]
n_label_cov_all = len(cov_all)
n_tp_cov_all = sum(1 for r in cov_all if r.matched)
use_cov_all = any(r.has_waveform is not None for r in subset_results)
subset_summary["P_S_combined"] = {
"n_label": int(n_label_all),
"n_matched_within_tp_tol": int(n_tp_all),
"recall": float(n_tp_all / n_label_all) if n_label_all else None,
"n_label_with_waveform": int(n_label_cov_all) if use_cov_all else None,
"n_tp_with_waveform": int(n_tp_cov_all) if use_cov_all else None,
"recall_covered": float(n_tp_cov_all / n_label_cov_all) if (use_cov_all and n_label_cov_all) else None,
}
summary["subsets"][subset] = subset_summary
return summary
def write_outputs(results: List[MatchResult], summary: Dict[str, Any], outdir: Path) -> None:
with (outdir / "summary.json").open("w", encoding="utf-8") as f:
json.dump(summary, f, ensure_ascii=False, indent=2)
with (outdir / "matches.jsonl").open("w", encoding="utf-8") as f:
for r in results:
f.write(json.dumps(r.__dict__, ensure_ascii=False) + "\n")
with (outdir / "summary.tsv").open("w", encoding="utf-8") as f:
f.write("subset\tphase\tn_label\tn_tp\trecall\tn_label_with_waveform\tn_tp_with_waveform\trecall_covered\tn_residual\tmean_s\tstd_s\tmedian_s\tabs_p90_s\tabs_p95_s\tgauss_mean\tgauss_std_mle\tgauss_std_unbiased\tt_df\tt_loc\tt_scale\n")
for subset, ss in summary["subsets"].items():
for phase, d in ss.items():
if phase == "P_S_combined":
continue
gfit = d.get("gaussian_fit_all_within_err_window", {}) or {}
tfit = d.get("student_t_fit_all_within_err_window", {}) or {}
f.write("\t".join([
subset, phase,
str(d.get("n_label")), str(d.get("n_matched_within_tp_tol")), str(d.get("recall")),
str(d.get("n_label_with_waveform")), str(d.get("n_tp_with_waveform")), str(d.get("recall_covered")),
str(d.get("n_residual_within_err_window")), str(d.get("residual_mean_s")),
str(d.get("residual_std_s")), str(d.get("residual_median_s")),
str(d.get("residual_abs_p90_s")), str(d.get("residual_abs_p95_s")),
str(gfit.get("mean")), str(gfit.get("std_mle")), str(gfit.get("std_unbiased")),
str(tfit.get("df")), str(tfit.get("loc")), str(tfit.get("scale")),
]) + "\n")
def load_matches(path: Path) -> List[Dict[str, Any]]:
rows = []
with path.open("r", encoding="utf-8") as f:
for line in f:
if line.strip():
rows.append(json.loads(line))
return rows
def plot_results(outdir: Path) -> None:
import matplotlib.pyplot as plt
matches_path = outdir / "matches.jsonl"
if not matches_path.exists():
print(f"[PLOT] missing {matches_path}", file=sys.stderr)
return
rows = load_matches(matches_path)
plot_dir = outdir / "figures"
plot_dir.mkdir(parents=True, exist_ok=True)
# Residual distribution by subset and phase
for subset in sorted({r["subset"] for r in rows}):
for phase in sorted({r["label_phase"] for r in rows if r["subset"] == subset}):
vals = np.array([
r["residual_s"] for r in rows
if r["subset"] == subset and r["label_phase"] == phase and r.get("residual_s") is not None
], dtype=float)
if vals.size == 0:
continue
vals = vals[np.isfinite(vals)]
if vals.size == 0:
continue
fig = plt.figure(figsize=(7.5, 4.8))
plt.hist(vals, bins=100, density=True, alpha=0.55, label=f"Residuals (n={vals.size:,})")
x_min, x_max = float(np.min(vals)), float(np.max(vals))
if x_min == x_max:
x_min -= 1.0
x_max += 1.0
x = np.linspace(x_min, x_max, 800)
mu = float(np.mean(vals))
sigma = float(np.std(vals, ddof=0)) if vals.size > 1 else 0.0
if scipy_stats is not None and sigma > 0:
gauss_pdf = scipy_stats.norm.pdf(x, loc=mu, scale=sigma)
plt.plot(x, gauss_pdf, linewidth=2, label=f"Gaussian μ={mu:.3f}, σ={sigma:.3f}")
if vals.size >= 3:
try:
df, loc, scale = scipy_stats.t.fit(vals)
if scale > 0:
t_pdf = scipy_stats.t.pdf(x, df, loc=loc, scale=scale)
plt.plot(x, t_pdf, linewidth=2, label=f"Student-t df={df:.2f}, loc={loc:.3f}, scale={scale:.3f}")
except Exception as exc:
print(f"[PLOT] Student-t fit failed for {subset}/{phase}: {exc}", file=sys.stderr)
else:
plt.text(
0.02, 0.95,
"Install scipy to overlay Gaussian/Student-t PDFs",
transform=plt.gca().transAxes,
va="top",
)
plt.axvline(0.0, linestyle="--", linewidth=1)
plt.xlabel("Residual time: automatic - label (s)")
plt.ylabel("Probability density")
plt.title(f"Residual distribution with fits: {subset}, {phase}")
plt.legend(fontsize=8)
plt.tight_layout()
fig.savefig(plot_dir / f"residual_fit_{subset}_{phase}.png", dpi=220)
plt.close(fig)
# Recall bar chart
summary = json.loads((outdir / "summary.json").read_text(encoding="utf-8"))
labels, vals = [], []
for subset, ss in summary["subsets"].items():
for phase, d in ss.items():
if phase == "P_S_combined":
continue
if d.get("recall") is not None:
labels.append(f"{subset}-{phase}")
vals.append(d["recall"])
if labels:
fig = plt.figure(figsize=(max(7, 0.8 * len(labels)), 4.5))
plt.bar(labels, vals)
plt.ylim(0, 1)
plt.ylabel("Recall")
plt.title("Phase-pick recall")
plt.xticks(rotation=45, ha="right")
plt.tight_layout()
fig.savefig(plot_dir / "recall_bar.png", dpi=200)
plt.close(fig)
# Automatic pick counts by original automatic phase
auto_counts = summary.get("auto_pick_count", {}).get("by_auto_phase", {})
if auto_counts:
phases = list(auto_counts.keys())
counts = [auto_counts[p] for p in phases]
fig = plt.figure(figsize=(max(7, 0.8 * len(phases)), 4.5))
plt.bar(phases, counts)
plt.ylabel("Number of automatic picks")
plt.title("Automatic pick counts by phase")
plt.xticks(rotation=45, ha="right")
plt.tight_layout()
fig.savefig(plot_dir / "auto_pick_count_bar.png", dpi=220)
plt.close(fig)
print(f"[PLOT] saved figures to {plot_dir}")
# -----------------------------
# Optional utilities
# -----------------------------
def print_db_info(db_path: Path) -> None:
db_path = Path(db_path).expanduser().resolve()
if not db_path.exists():
raise FileNotFoundError(f"SQLite index DB not found: {db_path}")
conn = connect_db(db_path)
n = conn.execute("SELECT COUNT(*) FROM auto_picks").fetchone()[0]
print(f"auto_picks: {n:,}")
print("phase counts:")
for ph, c in conn.execute("SELECT phase_name, COUNT(*) FROM auto_picks GROUP BY phase_name ORDER BY COUNT(*) DESC"):
print(f" {ph}: {c:,}")
conn.close()
# -----------------------------
# CLI
# -----------------------------
def main() -> None:
parser = argparse.ArgumentParser(description="Evaluate automatic phase picks against continuous-HDF5 annotation JSON.")
parser.add_argument("--auto-jsonl", type=Path, default=Path("data/picks/skynet.phase.jsonl"))
parser.add_argument("--label-json", type=Path, default=Path("data/label/annotations_for_continuous_hdf5.json"))
parser.add_argument("--index-db", type=Path, default=Path("~/skynet.pick.index.sqlite"))
parser.add_argument("--outdir", type=Path, default=Path("eval_picks/eval_skynet"))
parser.add_argument("--build-index", action="store_true", help="Build or update SQLite index from auto JSONL.")
parser.add_argument("--drop-existing", action="store_true", help="Drop existing index table before rebuilding.")
parser.add_argument("--keep-raw-json", action="store_true", help="Store raw JSON in SQLite. Not recommended for 40GB files.")
parser.add_argument("--batch-size", type=int, default=50000)
parser.add_argument("--tp-tol", type=float, default=1.5, help="TP tolerance in seconds.")
parser.add_argument("--err-window", type=float, default=5.0, help="Window for residual distribution in seconds.")
parser.add_argument("--min-prob", type=float, default=None, help="Optional minimum automatic pick probability.")
parser.add_argument("--phase-map", type=str, default=None, help="Example: 'P:Pg;S:Sg' or 'P:Pg,Pn;S:Sg,Sn'.")
parser.add_argument("--waveform-db", type=Path, default=None,
help="Waveform coverage SQLite index (built by hdf5_waveform_index.py). "
"When supplied, each label pick is checked for waveform availability. "
"recall_covered is computed over the subset that has waveform data, "
"so labels from un-processed stations/days are excluded from the denominator.")
parser.add_argument("--plot", action="store_true", help="Generate figures after evaluation.")
parser.add_argument("--db-info", action="store_true", help="Only print index database info.")
args = parser.parse_args()
if args.build_index:
build_auto_pick_index(
args.auto_jsonl, args.index_db,
batch_size=args.batch_size,
drop_existing=args.drop_existing,
keep_raw_json=args.keep_raw_json,
)
if args.db_info:
print_db_info(args.index_db)
return
phase_map = parse_phase_map(args.phase_map)
results, summary = evaluate(
label_json=args.label_json,
db_path=args.index_db,
outdir=args.outdir,
phase_map=phase_map,
tp_tol=args.tp_tol,
err_window=args.err_window,
min_prob=args.min_prob,
waveform_db=args.waveform_db,
)
print(json.dumps(summary, ensure_ascii=False, indent=2))
if args.plot:
plot_results(args.outdir)
if __name__ == "__main__":
main()
|