Street Address

Purpose

Use the address detector for Czech street-style addresses.

This detector reports accuracy above 85% and relies on component scoring rather than a single rigid format.

It is useful for full addresses and shorter street-plus-number forms.

Detector Name

address

Supported Formats

• Street Number
• Street Number/Flat
• Street, City
• Full address forms that include postal code and city

Examples:

• Dlouhá 15, Praha
• Vinohradská 1523/45, 120 00 Praha

Validation Logic

The detector tries two address patterns:

1. Full address pattern with street, number, postal code, and city
2. Simpler street-plus-number pattern

Component Scoring System

FastPII assigns a score from 0 to 100:

• Street component present and plausible: +30
• House number present and valid: +30
• Postal code present in Czech form: +20
• Known Czech city present: +20

The minimum accepted address score is 60.

Additional Validation Rules

• Street names shorter than 3 characters are rejected.
• Common non-address context words are filtered out.
• Full-address matches require a valid Czech city signal.
• House numbers must fit the detector's numeric pattern.

There is no checksum algorithm for this detector.

Python Examples

Detect an address

from fastpii import PrivacyGuard

guard = PrivacyGuard(regions=["cz"])
result = guard.detect("Bydlí na Dlouhé 15, Praha", detector_names=["address"])

for finding in result.findings:
    print(finding.value, finding.metadata)

Validate an address-like string

from fastpii import PrivacyGuard

guard = PrivacyGuard(regions=["cz"])
result = guard.validate("Dlouhá 15, Praha", "address")

print(result.is_valid)
print(result.metadata)

Expected Metadata

This detector documents these metadata fields:

• street
• house_number
• city
• postal_code
• score

Example Output

{
    "street": "Dlouhé",
    "house_number": "15",
    "city": "Praha",
    "postal_code": "",
    "score": 80,
}

Limitations

• City recognition depends on the built-in Czech city list.
• Short street-plus-number matches can pass with less context than full addresses.
• The score is heuristic, not a postal authority validation.

Notes

Verified detection example from the codebase:

guard.detect("Bydlí na Dlouhé 15, Praha")