How to Parse the ER Challenge

The model we use live postal is the great open source project. We've been making contributions to it now for some time. That model is trained on open street maps and other sources so that you can give it a messy address and will parse the pieces into the right buckets. Then you score the address, then you insert that into the system. So comparators like our name comparators have been statistically trained on 850 million names.