feat(csv-parser): Introduce header signature ranking metrics for enhanced CSV analysis
Deploy Development / deploy (push) Successful in 52s
Build Test / lint-backend (push) Successful in 0s
Build Test / build-frontend (push) Successful in 16s

- Added new functions for calculating header signature recall and ranking metrics, improving the analysis of CSV templates.
- Updated existing CSV analysis endpoints to utilize the new ranking metrics, enhancing the accuracy of template matching.
- Refactored related code to replace Jaccard score calculations with the new metrics, providing a more comprehensive evaluation of CSV structure.
- Improved documentation for new functions to clarify their purpose and usage in the context of CSV template analysis.
This commit is contained in:
Lars committed 2026-04-10 07:08:21 +02:00
1 parent c10da55ec6
commit b4cc3cb934
7 files changed
+223 -60

No files matched your search

+42 -2
View File
@@ -7,7 +7,7 @@ from __future__ import annotations
import csv
import io
import re
from typing import Any, Dict, Iterator, List, Tuple
from typing import Any, Dict, Iterator, List, Sequence, Tuple
_DEFAULT_DELIMS = [",", ";", "\t"]
@@ -115,7 +115,7 @@ def column_signature(headers: List[str]) -> List[str]:
def headers_signature_match_score(sig_csv: List[str], sig_template: List[str]) -> float:
"""Jaccard-Überlappung 0..1."""
"""Jaccard-Überlappung 0..1 (|A∩B|/|A∪B|). Fällt stark, wenn die CSV viele Zusatzspalten hat."""
a, b = set(sig_csv), set(sig_template)
if not a and not b:
return 1.0
@@ -126,6 +126,46 @@ def headers_signature_match_score(sig_csv: List[str], sig_template: List[str]) -
return inter / union if union else 0.0
def headers_signature_template_recall(sig_csv: Sequence[str], sig_template: Sequence[str]) -> float:
"""
Anteil der Template-Spalten (Signatur), die in der CSV vorkommen: |A∩B|/|B|.
100 %, sobald alle für die Vorlage relevanten Spalten in der Datei sind — unabhängig von
Zusatzspalten (Gewicht + Ernährung in einer Datei erzeugt keinen „Abzug“ für die jeweilige Vorlage).
"""
a = set(sig_csv)
b = {normalize_header_for_signature(str(x)) for x in sig_template}
b.discard("")
if not b:
return 1.0 if not a else 0.0
inter = len(a & b)
return inter / len(b)
def headers_signature_rank_metrics(sig_csv: List[str], sig_template: List[str]) -> dict[str, Any]:
"""
Einheitliche Kennzahlen für Vorlagen-Ranking und UI.
confidence = template_recall (empfohlen für Anzeige / Sortierung primär).
"""
a = set(sig_csv)
b = {normalize_header_for_signature(str(x)) for x in sig_template}
b.discard("")
inter = a & b
n_inter = len(inter)
n_b = len(b)
n_a = len(a)
union = len(a | b)
template_recall = n_inter / n_b if n_b else (1.0 if not n_a else 0.0)
jaccard = n_inter / union if union else 0.0
return {
"confidence": round(template_recall, 4),
"template_recall": round(template_recall, 4),
"jaccard": round(jaccard, 4),
"columns_matched": n_inter,
"columns_in_template": n_b,
"columns_in_csv": n_a,
}
def get_csv_import_limits(conn_row: dict | None) -> dict[str, int]:
"""Liest Limits aus system_config.csv_import; Fallback bei fehlendem Key."""
defaults = {"max_rows_per_file": 50_000, "max_file_bytes": 52_428_800}
+16 -12
View File
@@ -15,7 +15,7 @@ from csv_parser.core import (
column_signature,
decode_raw_bytes,
get_csv_import_limits,
headers_signature_match_score,
headers_signature_rank_metrics,
normalize_header_for_signature,
parse_csv_sample,
)
@@ -134,7 +134,7 @@ async def admin_analyze_csv_for_template(
):
"""
CSV hochladen wie im Nutzer-Import: Spalten + Vorschau + Vorschläge für field_mappings
und type_conversions. Optional Seed-Vorlage (ID) oder beste Jaccard-Systemvorlage für das Modul.
und type_conversions. Optional Seed-Vorlage (ID) oder beste Systemvorlage (Abdeckung, dann Jaccard).
"""
_ = session
if not get_module_definition(module):
@@ -191,15 +191,15 @@ async def admin_analyze_csv_for_template(
)
rows = [r2d(r) for r in cur.fetchall()]
best: dict | None = None
best_score = -1.0
best_key: tuple[float, int, float] = (-1.0, -1, -1.0)
for t in rows:
t_sig = list(t.get("column_signature") or [])
t_norm = sorted({normalize_header_for_signature(str(s)) for s in t_sig})
score = headers_signature_match_score(sig, t_norm)
if score > best_score:
best_score = score
m = headers_signature_rank_metrics(sig, t_sig)
key = (m["confidence"], m["columns_matched"], m["jaccard"])
if key > best_key:
best_key = key
best = t
if best and best_score > 0:
if best and best_key[0] > 0:
seed_row = best
seed_fm = (seed_row or {}).get("field_mappings") or {}
@@ -214,13 +214,17 @@ async def admin_analyze_csv_for_template(
seed_meta = None
if seed_row:
t_sig = [normalize_header_for_signature(str(s)) for s in (seed_row.get("column_signature") or [])]
t_sig = list(seed_row.get("column_signature") or [])
sm = headers_signature_rank_metrics(sig, t_sig)
seed_meta = {
"id": seed_row["id"],
"mapping_name": seed_row.get("mapping_name"),
"confidence": round(headers_signature_match_score(sig, sorted(set(t_sig))), 4)
if t_sig
else 0.0,
"confidence": sm["confidence"],
"template_recall": sm["template_recall"],
"jaccard": sm["jaccard"],
"columns_matched": sm["columns_matched"],
"columns_in_template": sm["columns_in_template"],
"columns_in_csv": sm["columns_in_csv"],
}
return {
+16 -6
View File
@@ -20,7 +20,7 @@ from csv_parser.core import (
decode_raw_bytes,
column_signature,
get_csv_import_limits,
headers_signature_match_score,
headers_signature_rank_metrics,
normalize_header_for_signature,
parse_csv_sample,
)
@@ -247,19 +247,29 @@ async def analyze_csv(
ranked = []
for t in templates:
t_sig = list(t["column_signature"]) if t["column_signature"] else []
t_norm = sorted({normalize_header_for_signature(str(s)) for s in t_sig})
score = headers_signature_match_score(sig, t_norm)
metrics = headers_signature_rank_metrics(sig, t_sig)
ranked.append(
{
"mapping_id": t["id"],
"module": t["module"],
"mapping_name": t["mapping_name"],
"is_system": bool(t.get("is_system")),
"confidence": round(score, 4),
"match_type": "signature_jaccard",
"confidence": metrics["confidence"],
"template_recall": metrics["template_recall"],
"jaccard": metrics["jaccard"],
"columns_matched": metrics["columns_matched"],
"columns_in_template": metrics["columns_in_template"],
"columns_in_csv": metrics["columns_in_csv"],
"match_type": "template_recall",
}
)
ranked.sort(key=lambda x: -x["confidence"])
ranked.sort(
key=lambda x: (
-(x.get("confidence") or 0),
-(x.get("columns_matched") or 0),
-(x.get("jaccard") or 0),
),
)
top = ranked[:25]
recommended = top[0] if top and (top[0]["confidence"] or 0) > 0 else None
+15
View File
@@ -8,6 +8,7 @@ from csv_parser.core import (
parse_csv_sample,
column_signature,
headers_signature_match_score,
headers_signature_rank_metrics,
get_csv_import_limits,
iter_csv_dict_rows,
)
@@ -46,6 +47,20 @@ def test_jaccard():
assert headers_signature_match_score(s1, s2) == pytest.approx(2 / 3)
def test_template_recall_full_when_csv_has_extra_columns():
"""Alle Template-Spalten in der CSV → Recall 1.0; Jaccard niedriger bei vielen Zusatzspalten."""
csv_sig = column_signature(
["D", "E", "F", "Extra1", "Extra2", "Extra3", "Extra4", "Extra5"]
)
tmpl_sig = column_signature(["d", "e", "f"])
m = headers_signature_rank_metrics(csv_sig, tmpl_sig)
assert m["confidence"] == 1.0
assert m["template_recall"] == 1.0
assert m["columns_matched"] == 3
assert m["columns_in_template"] == 3
assert m["jaccard"] == pytest.approx(3 / 8)
def test_get_csv_import_limits_default():
assert get_csv_import_limits(None)["max_rows_per_file"] == 50_000