Quick start

From install to a normalised batch in about five minutes.

Requirements

Python 3.9 or newer. No compiled extensions, no runtime dependencies. The rule tables are bundled in the wheel, so an offline install works as long as you have the .whl.

pip install petrock
python -c "import petrock; print(petrock.__version__)"

Parsing one address

from petrock import parse

a = parse("Nad Oborou 1149/7, 16200 Praha 6 - Brevnov", country="cz")
a.street      # 'Nad Oborou'
a.number      # '1149/7'
a.postcode    # '16200'
a.locality    # 'Praha 6'
a.district    # 'Brevnov'
a.confidence  # 0.97

Batch mode

For anything above a few thousand rows use parse_many. It compiles the rule table once and reuses it, which is roughly forty times faster than calling parse in a loop.

from petrock import parse_many

rows = [line.strip() for line in open("addresses.txt")]
for src, addr in parse_many(rows, country="cz", workers=4):
    if addr.confidence < 0.6:
        print("check:", src)

Confidence

Every result carries a confidence between 0 and 1. It is not a probability — it is the share of input tokens the parser could assign to a field. Anything above 0.85 is safe to use unattended; between 0.5 and 0.85 is usually a real address with an unusual suffix; below 0.5 means the parser mostly guessed.

Command line

petrock --country cz --in raw.csv --out clean.csv --column address
petrock --country de --stdin < list.txt

The CLI writes the same fields as the Python API plus a petrock_confidence column, and leaves every other column untouched.