Unique class instances in Python3

Question

suppose I have a class definition like this

class structure:
    def __init__(self, handle):
        self.handle = handle

How can I use numpy.unique or another tool of Python3 to find unique elements in a list of instances of this class? The comparison should be done with respect to the value of the 'handle' field.

Answer 1

numpy.unique is not the best tool for custom classes. Make your instances hashable (implementing __hash__ and __eq__ ), then use a set to reduce a list of instances to unique values:

class structure:
    def __init__(self, handle):
        self.handle = handle

    def __hash__(self):
        return hash(self.handle)

    def __eq__(self, other):
        if not isinstance(other, structure):
            # only equality tests to other `structure` instances are supported
            return NotImplemented
        return self.handle == other.handle

Sets efficiently can detect duplicates via the hash, confirming that the objects with the same hash are also equal first.

To get the unique instances, simply call set() on a sequence of instances:

unique_structures = set(list_of_structures)

Demo:

>>> class structure:
...     def __init__(self, handle):
...         self.handle = handle
...     def __hash__(self):
...         return hash(self.handle)
...     def __eq__(self, other):
...         if not isinstance(other, structure):
...             # only equality tests to other `structure` instances are supported
...             return NotImplemented
...         return self.handle == other.handle
...     def __repr__(self):
...         return '<structure({!r})>'.format(self.handle)
...
>>> list_of_structures = [structure('foo'), structure('bar'), structure('foo'), structure('spam'), structure('spam')]
>>> set(list_of_structures)
{<structure('bar')>, <structure('foo')>, <structure('spam')>}

Do take into account that the hash of any structure instance stored in a set or used a dictionary key must not change ; not altering the handle attribute during the lifetime of an instance is the easiest way to ensure this.