File size: 3,423 Bytes
74bf532 | 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 | """
Backfill framework alignment scores for existing items and rebuild the
AlignmentAggregate table (per institution + global).
Usage:
python scripts/backfill_alignment.py # DRY RUN — counts only
python scripts/backfill_alignment.py --apply
python scripts/backfill_alignment.py --apply --force # re-score current-version items
Safe to re-run: items already at ALIGNMENT_VERSION are skipped unless --force.
No network needed beyond the one-time embedding-model download.
"""
import argparse
import json
import os
import sys
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from uraas.config.alignment_frameworks import ALIGNMENT_VERSION
from uraas.database import Item, SessionLocal
from uraas.services.alignment_engine import (
recompute_aggregates,
score_item_alignment,
scoring_mode,
)
from uraas.utils.analytics_cache import analytics_cache
def main():
parser = argparse.ArgumentParser(description=__doc__)
parser.add_argument("--apply", action="store_true", help="Write changes (default: dry run)")
parser.add_argument("--force", action="store_true", help="Re-score items already at current version")
parser.add_argument("--batch", type=int, default=500)
args = parser.parse_args()
session = SessionLocal()
try:
q = session.query(Item)
if not args.force:
q = q.filter(
(Item.alignment_version.is_(None))
| (Item.alignment_version < ALIGNMENT_VERSION)
)
todo = q.count()
total = session.query(Item).count()
print("=" * 64)
print(f"Scoring mode: {scoring_mode()} | version: {ALIGNMENT_VERSION}")
print(f"Items to score: {todo} / {total}")
print("=" * 64)
if not args.apply:
print("[DRY RUN] No writes. Re-run with --apply.")
return 0
scored = aligned = 0
framework_hits = {}
while True:
batch = q.limit(args.batch).all()
if not batch:
break
for it in batch:
j, v = score_item_alignment(
it.title or "", it.abstract or "", it.dc_subject or ""
)
it.alignment_scores = j
it.alignment_version = v
scored += 1
if j:
aligned += 1
for fk in json.loads(j):
framework_hits[fk] = framework_hits.get(fk, 0) + 1
session.commit()
print(f" scored {scored}/{todo}")
print("\nPer-framework items with alignment:")
for fk, n in sorted(framework_hits.items(), key=lambda kv: -kv[1]):
print(f" {fk:24s} {n}")
# Aggregates: global + each distinct institution
rows = recompute_aggregates(session, None)
institutions = [
i for (i,) in session.query(Item.institution).distinct() if i
]
for inst in institutions:
rows += recompute_aggregates(session, inst)
print(f"\nAggregate rows written: {rows} ({1 + len(institutions)} scopes)")
analytics_cache.invalidate_all()
print("\n" + "=" * 64)
print(f"DONE. scored={scored} with_alignment={aligned}")
print("=" * 64)
return 0
finally:
session.close()
if __name__ == "__main__":
sys.exit(main())
|