usaddress

Why we included this project

Anyone cleaning messy customer or public records eventually hits free-form US addresses that won't fit a predictable pattern. usaddress handles those with a conditional-random-field model instead of brittle rules, so a single line like '123 Main St. Suite 100 Chicago, IL' splits into labeled parts such as street name, occupancy type, place name, and state, even when the formatting varies. The parse method returns a simple component list, while tag merges consecutive parts and strips punctuation for a cleaner result. It is honest about its limits: it makes educated guesses, and it does not validate or normalize addresses, so think of it as the parsing step inside a larger data pipeline rather than a complete verification tool. If you are not on Python, a Parserator API wraps the same model with a free first 1,000 parses.