Bug in NAUTILUS I implementation

[light-weaver]

Technically, you don't need two extra elements in these arrays. Only one will do the trick if the bug is fixed.

desdeo-mcdm/desdeo_mcdm/interactive/Nautilus.py

Lines 472 to 477 in 442021c

	self._xs = [None] * (self._n_iterations + 2)
	self._fs = [None] * (self._n_iterations + 2)
	self._ds = [None] * (self._n_iterations + 2)
	self._zs = [None] * (self._n_iterations + 2)
	self._lower_bounds = [None] * (self._n_iterations + 2)
	self._upper_bounds = [None] * (self._n_iterations + 2)

Initially, the zeroth element should be set, not the first element as done here (first step number is 1):

desdeo-mcdm/desdeo_mcdm/interactive/Nautilus.py

Lines 493 to 494 in 442021c

	self._lower_bounds[self._step_number] = self._ideal
	self._upper_bounds[self._step_number] = self._nadir

In all subsequent steps, you should not add 1 to the step number:

desdeo-mcdm/desdeo_mcdm/interactive/Nautilus.py

Lines 527 to 528 in 442021c

	self._lower_bounds[self._step_number + 1] = new_lower_bounds
	self._upper_bounds[self._step_number + 1] = self._zs[self._step_number]

There are multiple instances of that error, that should be fixed.

Finally, changing the number of iterations does not work as intended:

desdeo-mcdm/desdeo_mcdm/interactive/Nautilus.py

Line 564 in 442021c

extra_space = [None] * (resp["n_iterations"] - self._n_iterations)

The above line implies that the user provided number is the total number (including iterations that have already taken place).

desdeo-mcdm/desdeo_mcdm/interactive/Nautilus.py

Line 576 in 442021c

self._n_iterations_left = resp["n_iterations"]

The line above implies that the user provided number is the new remaining steps.

The two lines are in conflict, leading to the algorithm never working if one takes a step back.

[light-weaver]

Accept the pull request if you agree with the changes. Check if similar changes need to be made to other NAUTILUS implementations.

I also did not reduce the number of elements in the arrays in case they were being used for some other reason. Though in my test, the last element is always left to be None.