Question

def greatest(values):
    value_generator = (v for k,v in values)
    max_value = max(value_generator)
    return (k for k,v in values if v == max_value)

sample_data = ( ('id1', 3), ('id2', 5), ('id3', 5) )
items = list( greatest(sample_data) ) # Should produce ['id2', 'id3']

MapReduce anyone?

Answer 1

In fact, according to my tests, your version of greatest is faster — by a bit, anyway:

>>> def greatest_orig(values):
...     value_generator = (v for k,v in values)
...     max_value = max(value_generator)
...     return (k for k,v in values if v == max_value)
... 
>>> def greatest_max_key(values):
...     max_value = max(values, key=itemgetter(1))[1]
...     return (k for k,v in values if v == max_value)
... 
>>> sample_data = tuple(('id' + str(i), random.randrange(0, 1000)) for i in range(10000))
>>> list(greatest_orig(sample_data)) == list(greatest_max_key(sample_data))
True
>>> %timeit list(greatest_orig(sample_data))
1000 loops, best of 3: 1.67 ms per loop
>>> %timeit list(greatest_max_key(sample_data))
1000 loops, best of 3: 1.74 ms per loop

Of course, if you don’t like assigning your generator to a name, you can just pass the generator to max directly — way more readable than max(values, key=itemgetter(1))[1], IMHO:

>>> def greatest_max_iter(values):
...     max_value = max((v for k, v in values))
...     return (k for k, v in values if v == max_value)
...                                                
>>> list(greatest_orig(sample_data)) == list(greatest_max_iter(sample_data))
True
>>> %timeit list(greatest_max_iter(sample_data))
1000 loops, best of 3: 1.67 ms per loop

Python allows you to omit the outer parens when doing things like this:

>>> def greatest_max_iter(values):
...     max_value = max(v for k, v in values)
...     return (k for k, v in values if v == max_value)
... 

But for reasons I don’t understand, doing it that way is slightly slower:

>>> %timeit list(greatest_max_iter(sample_data))
1000 loops, best of 3: 1.69 ms per loop

These are true micro-optimizations, unlikely to matter much. But I think that readability favors max(v for k, v in values) or max((v for k, v in values)) over max(values, key=itemgetter(1))[1].