Make learnerND datastructures immutable where possible

At the moment there are a few datastructures that are described as "sets". where possible we should make these frozenset to remove the possibility of them being modified.

Wauw, according to this thread , a frozenset is hashable and can therefore be used as a dictionary key. Maybe this is a clean way to store simplices (rather than having a sorted tuple) in the triangulation/learner.

any opinions?

In [1]: import sys

In [2]: sys.getsizeof((1, 2, 3))
Out[2]: 72

In [3]: sys.getsizeof(frozenset((1, 2, 3)))
Out[3]: 224

just saying

ohw, that is a shame

because a frozenset is a hashmap, whereas a tuple is an array

added LearnerND label

Will this ever be implemented? If not, can we close it?

mentioned in issue #130

It's priority: low, so please let's not close it. @jbweston's illustration shows clearly that there's a clear benefit in implementing this issue.

@jbweston's illustration shows clearly that there's a clear benefit in implementing this issue.

No, I was just saying that we should prefer tuples over sets where possible, but that some of the structures that currently are sets should be frozenset if they can be

Indeed, I misunderstood.

Looking at the documentation of LearnerND I'm not sure what data structures would benefit from conversion.

On the other hand, it does seem like we could make a bunch of methods private and therefore make the API simpler.

mentioned in commit 7c340516

mentioned in merge request !142 (merged)

Like this @anton-akhmerov? !142 (merged)

closed via commit 7c340516

closed via merge request !142 (merged)