cma.utilities.utils.SolutionDict

dictionary with computation of an hash key.

The hash key is generated from the inserted solution and a stack of previously inserted same solutions is provided. Each entry is meant to store additional information related to the solution.

>>> import cma.utilities.utils as utils, numpy as np
>>> d = utils.SolutionDict()
>>> x = np.array([1,2,4])
>>> d[x] = {'f': sum(x**2), 'iteration': 1}
>>> assert d[x]['iteration'] == 1
>>> assert d.get(x) == (d[x] if d.key(x) in d.keys() else None)
>>> y = [1,2,4]
>>> d[y] = {'f': sum([n ** 2 for n in y]), 'iteration': 1}
>>> assert d[y]['iteration'] == 1
>>> assert d.get(y) == (d[y] if d.key(y) in d.keys() else None)
>>> d[2] = 3
>>> assert d[2] == 3

data_with_same_key behaves (in contrast to the order of keys()) like a stack (see setitem and delitem). Thereby the last assignment self[key] = ... determines, as to be expected, the value of self[key] even when self.data_with_same_key[key] is not empty. If the attribute behave_as_queue is set to True, it behaves like a queue (FIFO).

Technical detail: after a same-key item was removed, self[key] returns the stacked (or queued) previous value, however self._unhashed_keys[key] still returns the last unhashed key which could be a different object than the previous but with the same hash key. However, the object has most likely the correct content because the hash was the same. This happens because only the last unhashed key is kept. As long as the content of self._unhashed_keys[key] is not changed, this should be fine.

TODO: in .truncate, the iteration key is used to clean up without error management.