Add inverted and shifted hartmann benchmark

[Hrovatin]

Add two new TL benchmarks:

• Inverted hartmann
• Hartmann shifted in one dimension

@kalama-ai

[kalama-ai]

Thanks, @Hrovatin for implementing the new benchmarks. Except some clean up, I am fine with the implementation.

[AVHopp]

Like the way this is going! Have some more ideas for further simplification

[AVHopp]

Some comments related to parts of the design that I still do not fully agree with resp. understand

[Scienfitz]

@Hrovatin any movement in this PR or should we close it?

[Hrovatin]

@Scienfitz still on my todo list to add your last two comments

[Scienfitz]

lgtm, last comments but otherwise ready for merge

[Scienfitz]

Suggested change

	# Shift the bounds
	# Extend the bounds

"shift" would only be accurate if you always change lower and upper end by same amount

Actually, this bounds = expression below still looks overly complex to me, the expression reads terribly. Naively one would shift both upper and lower bound by the same amount, that doesn't work?

Assuming there is a reason for that, I can still think of this:

bounds = [
    (low - abs(s), high + abs(s))
    for (low, high), s in zip(bounds_array, self.shift)
]

(or similar)
which is not shifting but widening the bounds (should also work for passing the validation and is a more readable expression).

[Hrovatin]

I needed to adapt it as follows

# Note: We can not only shift the upper and lower bounds as that leads to the
# optimal value (``_optimizer``) being excluded from bounds. As this is
# hard-coded in the Hartmann class init, we can not override it before it is
# evaluated.
shifted_bounds = bounds + np.array(self.shift)[:, None]
bounds_extended = list(
    map(
        tuple,
        np.stack(
            [
                np.minimum(bounds[:, 0], shifted_bounds[:, 0]),
                np.maximum(bounds[:, 1], shifted_bounds[:, 1]),
            ],
            axis=1,
        ),
    )
)

[Scienfitz]

does the benchmark work when running locally?

[Hrovatin]

For me it did with python -m benchmarks ...