File size: 4,865 Bytes
ccd1a6d 9f3cbd9 ccd1a6d | 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 | """Score HyperAPI's Extract endpoint on hyperbots-inc/synthetic-invoice-extraction.
Setup:
# install the HyperAPI SDK: https://apis.hyperbots.com/docs/quickstart
export HYPERAPI_KEY=hk_test_... # free key: https://apis.hyperbots.com/dashboard/api-keys
Run:
python evaluate_hyperapi.py --limit 25 # score 25 invoices (uses the OCR text as a .txt upload)
python evaluate_hyperapi.py --limit 25 --pdf # render the HTML to PDF first (needs: pip install weasyprint)
python evaluate_hyperapi.py --dry-run # check the scoring pipeline with gold labels, no API calls
The free tier covers 30 pages per month, so keep --limit at or below 30 on a free key.
"""
import argparse
import json
import os
import re
import tempfile
from datetime import datetime
CORE_FIELDS = ["invoice_number", "date", "due_date", "vendor_name", "total_amount", "currency"]
DATE_FORMATS = ["%Y-%m-%d", "%m/%d/%Y", "%d/%m/%Y", "%b %d, %Y", "%d %B %Y", "%B %d, %Y", "%d-%m-%Y", "%Y/%m/%d"]
def load_rows(path=None):
if path is None:
from huggingface_hub import hf_hub_download
path = hf_hub_download("hyperbots-inc/synthetic-invoice-extraction", "data/test.jsonl", repo_type="dataset")
with open(path, encoding="utf-8") as f:
return [json.loads(line) for line in f if line.strip()]
def norm_text(x):
return re.sub(r"\s+", " ", str(x or "")).strip().lower().rstrip(".")
def norm_amount(x):
if x is None:
return None
s = re.sub(r"[^0-9.\-]", "", str(x).replace(",", ""))
try:
return round(float(s), 2)
except ValueError:
return None
def norm_date(x, hint=None):
s = str(x or "").strip()
fmts = DATE_FORMATS
if hint == "eu": # dd/mm/yyyy is ambiguous with mm/dd/yyyy; prefer the document's convention
fmts = ["%d/%m/%Y"] + [f for f in DATE_FORMATS if f != "%d/%m/%Y"]
for f in fmts:
try:
return datetime.strptime(s, f).strftime("%Y-%m-%d")
except ValueError:
continue
return s
def field_match(name, pred, gold, row):
if name in ("total_amount",):
return norm_amount(pred) is not None and norm_amount(pred) == norm_amount(gold)
if name in ("date", "due_date"):
return norm_date(pred, row.get("date_format")) == gold
return norm_text(pred) == norm_text(gold)
def line_items_match(pred_items, gold_items):
"""Fraction of gold line items whose total appears among predicted line totals."""
pred_totals = [norm_amount(i.get("total") or i.get("amount")) for i in (pred_items or [])]
hits = 0
for g in gold_items:
t = norm_amount(g["total"])
if t in pred_totals:
pred_totals.remove(t)
hits += 1
return hits / max(len(gold_items), 1)
def to_file(row, tmpdir, as_pdf):
if as_pdf:
from weasyprint import HTML
path = os.path.join(tmpdir, row["id"] + ".pdf")
HTML(string=row["html"]).write_pdf(path)
else:
path = os.path.join(tmpdir, row["id"] + ".txt")
with open(path, "w", encoding="utf-8") as f:
f.write(row["ocr_text"])
return path
def predict(client, path):
res = client.extract(path)
result = res.get("result", res) if isinstance(res, dict) else {}
return result.get("entities", {}), result.get("line_items", [])
def main():
ap = argparse.ArgumentParser()
ap.add_argument("--limit", type=int, default=25)
ap.add_argument("--pdf", action="store_true", help="render HTML to PDF before upload (needs weasyprint)")
ap.add_argument("--data", default=None, help="local path to test.jsonl (default: download from the Hub)")
ap.add_argument("--dry-run", action="store_true", help="score gold labels against themselves; no API calls")
args = ap.parse_args()
rows = load_rows(args.data)[: args.limit]
client = None
if not args.dry_run:
from hyperapi import HyperAPIClient
client = HyperAPIClient() # reads HYPERAPI_KEY
hits = {f: 0 for f in CORE_FIELDS}
li_scores = []
with tempfile.TemporaryDirectory() as tmp:
for row in rows:
gold = row["labels"]
if args.dry_run:
ents, items = gold["entities"], gold["line_items"]
else:
ents, items = predict(client, to_file(row, tmp, args.pdf))
for f in CORE_FIELDS:
hits[f] += field_match(f, ents.get(f), gold["entities"][f], row)
li_scores.append(line_items_match(items, gold["line_items"]))
if client is not None:
client.close()
n = len(rows)
print(f"Scored {n} invoices")
for f in CORE_FIELDS:
print(f" {f:<16} {hits[f] / n:6.1%}")
print(f" {'line_items':<16} {sum(li_scores) / n:6.1%} (share of gold line totals recovered)")
if __name__ == "__main__":
main()
|