[ENH] allowing old function API to be passed as experiment

[fkiraly]

From a user perspective, the ability to define ad-hoc functions via the old my_callable(opt) notation was quite nice, and imo should be retained for compabitility with user code, and due to quick specification.

This PR adds

• the ability to pass functions of the old my_callable(opt) signature to any optimizer
• a FunctionExperiment which adapts functions to the BaseExperiment API. This experiment is public, since it is also useful when a user already has a function or callable to adapt.

The feature also requires to relax the BaseExperiment contract, and allow passing of params with unknown keys, or arbitrary keys, or keys not known in advance.

[SimonBlanke]

These functions do not need to be top-level. Please move these test functions into the get_test_params-method.

[fkiraly]

hm, I thought they might be, otherwise the test sfor pickling might fail (when functions are in a local scope) but I will try

[SimonBlanke]

right. This can happen with some pickling packages. Still, there must be a better way, then putting test data in source code files at top level.

[fkiraly]

it's private... I can try to move it into get_test_params, but I suspect it will cause test failures

[fkiraly]

done

[SimonBlanke]

Tasks:

• Please cleanup the test-functions
• Add a basic example similar to this: https://github.com/SimonBlanke/Hyperactive/blob/master/examples/optimization_techniques/hill_climbing.py

[fkiraly]

I added the option as the "simplest" option in the hyperactive_intro notebook