Make optional arguments to `create_study` actually optional

[PGijsbers]

From the docs create_study and create_benchmark_suite contain some optional parameters, e.g. alias, which don't have defaults which actually makes them required.

1. For the alias parameter a None default seems sensible.
2. For a study the suite's id is also optional, is it by design that we want studies without suites?
3. I would argue that for a study, the run_ids should also be optional as they could be created later (e.g. in a setup where you define a study, and upload results to it as they come in).

[joaquinvanschoren]

1. None is a good default indeed
2. Yes, we can create a collection of runs that is not directly tied to a suite. E.g. one could create a collection containing all runs of a specific algorithm/flow.
3. Agreed!

[PGijsbers]

I made some progress but ran into the following two "issues", which makes me want to double check before making the changes (in particular with @mfeurer for the first issue).

First, downloading studies (and suites) without any attached data is currently explicitly not possible. I think we should adapt the code, as the server should decide what is and is not acceptable. Wdyt?

Second, uploading an explicitly empty study (i.e. having a data field but only with an empty sequence) is not allowed (xsd). This is not necessarily an issue (we can deal with this in openml-python internally), but perhaps we should make it also explicit on the server side.

XML for an implicitly empty study (uploads fine):

<oml:study xmlns:oml="http://openml.org/openml">
	<oml:main_entity_type>run</oml:main_entity_type>
	<oml:name>empty-study-implicit</oml:name>
	<oml:description>a study with no runs attached implicitly</oml:description>
</oml:study>

XML for an explicity empty study:

<oml:study xmlns:oml="http://openml.org/openml">
	<oml:main_entity_type>run</oml:main_entity_type>
	<oml:name>empty-study-explicit</oml:name>
	<oml:description>a study with no runs attached explicitly</oml:description>
	<oml:runs>
	</oml:runs>
</oml:study>

returns the following error:

<oml:error xmlns:oml="http://openml.org/openml">
	<oml:code>1032</oml:code>
	<oml:message>Description file does not correspond to XSD schema</oml:message>
	<oml:additional_information>
	    XML does not correspond to XSD schema. Error Element '{http://openml.org/openml}runs': 
            Missing child element(s). Expected is ( {http://openml.org/openml}run_id ). on line 5 column 0. 
        </oml:additional_information>
</oml:error>

If we are not adding allowing explicitly empty studies server-side, I will just make sure they get created implicitly (first xml). But if we do change this, then I can immediately make sure both are supported.

[mfeurer]

This all sound reasonable and I agree with the proposed changes. I just don't see what I can say about the XSD not allowing empty studies

[PGijsbers]

Thanks! I just wanted an OK on the openml-python side of things because at some point code was added to explicitly prevent this behavior. So I wasn't sure if it was there for a particular reason and should remain (which would be incompatible with the empty suite/study feature).

[mfeurer]

So I wasn't sure if it was there for a particular reason and should remain (which would be incompatible with the empty suite/study feature).

I don't recall such reason and also can't think of one at the moment.