XpyriMentor and Suggestors

[SRFU-NN]

XpyriMentor, Suggestor, LHSSuggestor, RandomSuggestor, InitialPointSuggestor, DefaultSuggestor and suggestor_factory implemented.

Small fixes in old PO code:
ProcessOptimizer.Optimizer.update_next() recognizes changes in number of told points
ProcessOptimizer.utils.is_listlike returns true on np.ndarray
ProcessOptimizer.utils.is_2Dlistlike returns true on 2D np.ndarray

Also removes support for Python 3.8, which is now in end-of-life and no longer gets security updates.

[CoRiis]

I only went through the XpyriMentor code, and I should probably have started with the suggestor.py script instead of going through the files from the top.

[SRFU-NN]

We could change "name" to "suggestor_name" in suggestor definition dicts.

On the one hand, in means that only suggestor and things starting with suggestor_ is disallowed as inputs in Suggestor.__init__()'s, and it makes for more readable suggestor definitions.

On the other hand, it makes for more typing to make suggestor definitions.

Thoughts? If nobody complains, I'll change it.

Old style:

suggestor_definition = {
    "name": "Sequential",
    "suggestors": [
        {"suggestor_budget": 7, "name": "LHS"},
        {
            "suggestor_budget": 20,
            "name": "Random",
            "suggestors": [
                {"suggestor_usage_ratio": 80, "name": "PO", "acq_func_kwargs": {"xi": 10}},
                {"suggestor_usage_ratio": 20, "name": "PO", "acq_func_kwargs": {"xi": 0.00001}},
            ],
        },
        {"suggestor_budget": float("inf"), "name": "PO", "acq_func_kwargs": {"xi": 0.00001}},
    ]
}

New style:

suggestor_definition = {
    "suggestor_name": "Sequential",
    "suggestors": [
        {"suggestor_budget": 7, "suggestor_name": "LHS"},
        {
            "suggestor_budget": 20,
            "suggestor_name": "Random",
            "suggestors": [
                {"suggestor_usage_ratio": 80, "suggestor_name": "PO", "acq_func_kwargs": {"xi": 10}},
                {"suggestor_usage_ratio": 20, "suggestor_name": "PO", "acq_func_kwargs": {"xi": 0.00001}},
            ],
        },
        {"suggestor_budget": float("inf"), "suggestor_name": "PO", "acq_func_kwargs": {"xi": 0.00001}},
    ]
}

[RuneChristensen-NN]

"skipping" can be hard to interpret. I get why (starting with n data points it jumps to n+1) but maybe make it another sentence.

" If n data points are supplied, the strategizer will behave as if these points are generated by the suggestors and start from n+1 evaluations in the sequential list of suggestors."

Not great but at least more descriptive.

[RuneChristensen-NN]

Example in docstring would be nice

[SRFU-NN]

I would prefer to provide examples in the format the factory accepts. In general, the factory input is designed to be readable and modifiable, while the init() inputs for suggestors are designed to be easy to work with.

Does that make sense? Should we write that down somewhere in the documentation? Should we have more examples in the factory docstring, or in the suggestor docstrings?

[RuneChristensen-NN]

... used as budget for the last suggestor.

[SRFU-NN]

Done, much better phrasing.

[RuneChristensen-NN]

I would probably say sharp smaller than

[SRFU-NN]

Makes sense, I've written negative all over. A budget of 0 will still work, it will just mean the suggestor is never used.

[RuneChristensen-NN]

Change Error msg: Only the last suggestor can have infinite budget

[SRFU-NN]

Done, much better phrasing.

[RuneChristensen-NN]

Something seems a bit off here. I will take a deeper look later.

What if you have 3 suggestors, 5 points budget each and load in 7 data points.
How would that work. I guess it is the tracking of what has already happened and what budget each suggestor has that is a bit confusing.

[SRFU-NN]

I'll move the testing whether we are done to a separate if clause at the end, and moved updating suggestions and number_left_to_find out of the main if loop. That makes the flow slightly clearer.

If we ask for 4 points, we will start out with number_left_to_skip = 7 and number_left_to_find =4

For the first suggestor, we have to skip it completely (number_left_to_skip >= budget; 7>=5)

We then have number_left_to_skip = 2 and number_left_to_find =4

For the second suggestor, we will use it, but need more points than the suggestor can provide (number_left_to_skip + number_left_to_find >= budget; 2 + 4 > 5). We use that suggestor to find 3 points (number_from_this_suggestor = budget - number_left_to_skip = 5 - 2 = 3).

We then have number_left_to_skip = 0 and number_left_to_find =1

We use the final suggestor to find one point (else).