dataframe.py

"""Line items → pandas DataFrame → CSV.

The one shape that needs a word of explanation: a line-item array is a *bare* list, and
each cell inside each row carries its own confidence. `unwrap()` flattens a row into the
plain dict pandas wants; keeping the confidences means you can filter before you total.

    pip install pandas
    python examples/dataframe.py invoice.pdf out.csv
"""

import sys

import pandas as pd

from visionapi import VisionAPI, at_least, rows, unwrap, value

path = sys.argv[1] if len(sys.argv) > 1 else "invoice.pdf"
out = sys.argv[2] if len(sys.argv) > 2 else "line-items.csv"

vision = VisionAPI()
res = vision.analyze(file=path, preset="invoice")
result = res["result"]

records = []
for row in rows(result, "line_item"):
    flat = unwrap(row)
    # Carry the header down onto every line so the CSV stands on its own.
    flat["invoice_id"] = value(result, "invoice_id")
    flat["invoice_date"] = value(result, "invoice_date")
    # A per-cell confidence check: anything but "high" on the money column is worth a look.
    flat["amount_is_confident"] = at_least(row.get("amount"), "high")
    records.append(flat)

frame = pd.DataFrame.from_records(records)
frame.to_csv(out, index=False)

print(frame.head().to_string())
print(f"\n{len(frame)} line(s) → {out}")

total = value(result, "total_amount") or value(result, "total")
if total is not None and "amount" in frame:
    summed = frame["amount"].fillna(0).sum()
    print(f"stated total {total} vs summed lines {summed:.2f}" +
          ("  ✓" if abs(summed - float(total)) < 0.01 else "  ← worth a human look"))