group_params.py

from collections import defaultdict
from six import iteritems


def group_params(items, key_name=lambda x: True):
    """bin by sub-dicts where values are compared by id if not hashable
    
    >>> a = ('boo', 6)
    >>> b = ('boo', 6)
    >>> id(a) == id(b)
    False
    >>> import numpy
    >>> c = numpy.array([1, 2, 3])
    >>> items = [{'p': a, 'q': 1}, {'p': b, 'q': 2}, {'p': c, 'q': 3}, {'p': c, 'q': 4}]
    >>> groups = list(group_params(items, lambda x: x == "p"))
    >>> len(groups)
    2
    >>> groups
    >>> sorted([[x['q'] for x in gitems] for g, gitems in groups if g['p'] == a][0])
    [1, 2]
    >>> sorted([[x['q'] for x in gitems] for g, gitems in groups if g['p'] is c][0])
    [3, 4]
    """
    # can reduce work if input is sorted tuples rather than dict

    groups = defaultdict(list)
    canonical = {}  # maps hashable x to a canonical instance of x
    id_to_obj = {}

    for params in items:
        group = []
        for k, v in iteritems(params):
            if key_name(k):
                try:
                    v_id = id(canonical.setdefault(v, v))
                except TypeError:
                    v_id = id(v)
                id_to_obj[v_id] = v
                group.append((k, v_id))
        groups[tuple(sorted(group))].append(params)

    return [({k: id_to_obj[v_id] for k, v_id in group}, group_items)
            for group, group_items in iteritems(groups)]