[WIP] Add datasets from "Encoding high-cardinality string categorical variables" paper

[TwsThomas]

Adding the 24 datasets from "Encoding high-cardinality string categorical variables" paper.

Datasets active on OpenML:

• building_permits
• beer_reviews
• colleges
• crime_data (dead link :/)
• drug_directory
• employe_salaries
• federal_election
• journal_influence
• kickstarter_projects
• medical_charge
• met_objects
• midwest_survey
• open_payment (dead link :/)
• public_procurement
• road_safety (dead link :/)
• traffic_violations
• vancouver_employee
• wine_reviews
• Adult (from UCI, https://www.openml.org/d/1590)
• cacao_flavor
• california_housing (https://www.openml.org/d/537)
• dating_profile
• house_prices
• house_sales
• intrusion_detection (https://www.openml.org/d/1113)

I've drop few rows (in colleges and la_crime) when label was missing. (It seems it was required from openml).
In public_procurement, I replace nan by -1 in the label award_value_euro

[GaelVaroquaux]

Would the code here be simplified once those datasets make it to openML?

[TwsThomas]

Would the code here be simplified once those datasets make it to openML?

Yes!
This code is just to record the upload process.
Once in openML, this will be a one line like:
sklearn.datasets.fetch_openml(data_id=xxx)

[GaelVaroquaux]

This code is just to record the upload process.

Of course! Great, super useful!!

[GaelVaroquaux]

Should we merge this with what you have, and we open another PR?

[TwsThomas]

Should we merge this with what you have, and we open another PR?

Yes. I'll open a new one to upload the last datasets.