Refactor of the fetching system

[LilianBoulard]

Follows #147

---

This is a refactor of the dataset fetching system.
Historically, datasets were fetched from various websites.

Our aim with this update is to only use OpenML.org's API, through Scikit-learn's fetch_openml() function.
This allows us to have a much more reliable and unified interface, and avoids losing access to datasets due to deletion, renaming, etc.

For instance, with the current system, 4 on the 7 datasets are unavailable (403 Access Denied, website down, etc.).

The user-interface stays the same with the functions fetch_*() (e.g fetch_open_payments()), still returning a similar dictionary.
The major difference is that this dictionary returns, among other information, a path, where a CSV file is located, and must be loaded (using for instance pandas' read_csv() function).

---

TL;DR of the previous thread:
the way fetch_openml() is used here makes pandas a requirement.

[LilianBoulard]

So, for this (I hope last) error, we got an issue with the JSON decoding.
It gets raised inside sklearn.datasets.fetch_openml, and unfortunately, this is linked to the version, as the tests pass with scikit-learn>=0.21.0.

Unless told to, I won't try to dig inside sklearn's code to figure out at which step it goes wrong, and perhaps monkeypatch a solution.
So meanwhile, we have two options:

• bump sklearn's minimal version for dirty_cat to 0.21.0
• raise a clean error inviting the user to upgrade his version if he tries to fetch a dataset

What do you suggest ?

[GaelVaroquaux]

Is this error happening for all the datasets, or only some?

[GaelVaroquaux]

By the way, I checked, sklearn 0.21.0 is a couple years old. It's fine to bump the requirement.

[LilianBoulard]

Is this error happening for all the datasets, or only some?

All of them, as the error is linked to fetch_openml and they all use it. See this screenshot:

[GaelVaroquaux]

I finished reviewing this and made all my comments. After these small things, this will be ready for merge.

[LilianBoulard]

These most recent changes improve the user interface.
Now, instead of reading the CSV from the disk, it is directly loaded in memory if load_dataframe=True.
It also returns a namedtuple instead of a dictionary, which has an object-oriented interface.

[alexis-cvetkov]

This PR looks good to me, I think it can be merged.
Thanks Lilian !

[GaelVaroquaux]

I was about to merge, but don't we need a an entry summarizing things in the changelog?

[alexis-cvetkov]

Yes you're right, I always forget that.

[GaelVaroquaux]

Yey! Merged