Upload to OpenML

[PGijsbers]

Supersedes #236.

Adds a script that allows you to upload the results from the results directory to OpenML.

New runbenchmark.py arguments:

• --openml-test-server: connect to the test server instead of the production server.
• --openml-run-tag: add a tag to the OpenML runs (the tag amlb is also added automatically).

After running the benchmark, you can now invoke upload_results.py which can:

• check that all required information is there to upload a run to OpenML, in particular if results for all folds of the task(s) are available.
• upload the runs to OpenML

Example usage:

python runbenchmark.py randomforest openml/t/115 1h8c --openml-test-server --openml-run-tag pr523
python upload_results.py  -m check
python upload_results.py  -m upload

Uploaded example: https://www.openml.org/r/10593764 (note: this is a different example than the example usage above, as there were test server issues so I had to use the production server)

[PGijsbers]

Looking for feedback from anyone interested to review this, feel free to forward this to users that have some familiarity with the benchmark and/or openml. @mfeurer I would really appreciate it if you could have a quick look, at least as far as the usage and openml upload script goes.

[PGijsbers]

closing/re-opening to trigger CI on new master