How do you sort a dictionary based on values?

To sort a dictionary by its values, you first extract its key-value pairs as an iterable of tuples, then sort this iterable, and finally, if desired, reconstruct a new dictionary from the sorted pairs. The most common Pythonic approach involves using the sorted() function with a custom key.

Mechanics and Why

Python dictionaries are inherently unordered before version 3.7, and even after, they maintain insertion order, not value-based sorted order. To sort by values, you must convert the dictionary's items into a sortable sequence. d.items() returns a view of key-value pairs as tuples (e.g., ('a', 10)). The sorted() function then takes this iterable and sorts it. The key=lambda x: x[1] argument specifies that the sorting should be based on the second element of each tuple (the value). Alternatively, operator.itemgetter(1) can be used for slightly better performance and readability in some contexts. To sort in descending order, add reverse=True.

The output of sorted() is always a list of tuples. If you need a dictionary where the insertion order reflects this sort, you can pass this list back to the dict() constructor.

data = {'apple': 3, 'banana': 1, 'cherry': 2}
sorted_items = sorted(data.items(), key=lambda item: item[1])
# sorted_items will be [('banana', 1), ('cherry', 2), ('apple', 3)]
# To get a new dictionary with insertion order reflecting the sort:
sorted_dict = dict(sorted_items)
# sorted_dict will be {'banana': 1, 'cherry': 2, 'apple': 3}

Production Considerations and Cost

This operation has a time complexity of O(n log n) due to the sorting algorithm. For large dictionaries, this approach loads all items into memory, which can be a bottleneck. In data engineering contexts, sorting large datasets often involves distributed processing frameworks like Apache Spark. Spark's sortByKey or orderBy operations, for instance, trigger a "shuffle" phase, which redistributes data across partitions to ensure global ordering, a much more resource-intensive process than in-memory Python sorting.

In the interview, also mention that the output is fundamentally a list of tuples representing the sorted order, as dictionaries themselves don't maintain a value-based sorted state.