Possible bug: Datageneration from custom function inspects function signature.

[elvisbr0wn]

When playing around with this example in the documentation: https://f3dasm.readthedocs.io/en/latest/auto_examples/003_datageneration/001_own_datagenerator.html#car-stopping-distance-problem

If you change the custom function to

def y(x2):
    z = norm.rvs(1.5, 0.5, size=1)
    y = z*x2 + 0.1*x2**2
    return y

You get that the experiment sample have the status of error after executing experiment_data.evaluate(data_generator=y, output_names=['y']), getting

File "/Users/eaguerov/mambaforge/envs/3dasm/lib/python3.10/site-packages/f3dasm/_src/experimentdata/experimentsample.py", line 161, in get
    value, from_disk = self._load_from_experimentdata(item)
TypeError: cannot unpack non-iterable NoneType object

This is because we changed the signature of y(x) to y(x2), which does not coincide with the input name on the domain of the experimentdata object.

I personally could understand why this behavior might be desirable, but I believe that the consequences of the current implementation are that you cannot generate basically anything that was not typed out manually by the user.

If you train a model, and want to test the predictions of the model on some input data, you simply cannot do data_generator=my_model, unless the my_model variable has the exact signature of the input features that the experimentData Domain has. This is cumbersome for problems with 10+ input features.

I do not have an explicit solution in mind, I just wanted to share my thoughts with the developers so that a scalable solution is devised.

[elvisbr0wn]

Just for the record, I noticed that this is also discussed in #268. I will keep this issue because I believe this topic needs more discussion