Measure scrubadub accuracy on real data

[thomasbird]

We can measure scrubadub accuracy with fake data (see #59), it would be good to find real data sources where this is also possible.

[TheBellman]

It would be useful to include that as a repeatable test

[thomasbird]

Yes agreed, we should do something similar to the fake data, where it is run in the CI process for speedy models (and offline for slower models):

https://travis-ci.org/github/LeapBeyond/scrubadub/jobs/741856233#L616

[thomasbird]

Thoughts:

• Wikipedia list of people
• Addresses/phone numbers of take away places

[aCampello]

This might be useful for the addresses: https://www.kaggle.com/openaddresses/openaddresses-europe. No UK data for this one, but perhaps there are some related dataset that have it.

[aCampello]

For the UK locale, for addresses and postcode, we can perhaps use the Companies House open database https://www.gov.uk/government/organisations/companies-house