Collections and Defaultdict in Python

NSA whistleblower in exile, Edward Snowden, talks about how FBI could have reviewed 650K emails in less than 8 days!

snowden_tweet

Snowden says the FBI could have used hashing to identify emails that were not copies of ones they had already seen. Few things capture people’s interest like alleged conspiracies and political intrigue, yes? I’m no different. But what interests more is hashing. Touted by many as the “greatest idea in programming,” hashing, which involves the hash function, helps you find, say A, stored somewhere, say B. For example, the organizing and accessing of names and numbers in your “can’t bear to be parted from" smartphone.

Hashing is a technique where a data-structure called the “hash map” is implemented. This structure is an associative array where specific keys are mapped to specific values. A hash function is then used to compute an index into an array of buckets or slots from which the desired value can be found. The result is that (key, value) lookups are extremely fast and more efficient than searches based on popular trees like BST. To get in-depth knowledge about hashing, I recommend that you can go through our “Basics of Hash Tables” in our practice section.

Almost all modern languages have hashing implemented at the language level. In Python, hashing is implemented using the dictionary data structure, which is one of the basic data structures a beginner in Python learns. If you have only been using the dict module implementation in your code, I suggest you look at other implementations like defaultdicts and ordereddicts and use them more frequently in your code. Here, we will look more closely into the defaultdict module.

Defaultdicts come in the Collections internal library. Collections contains alternatives to the general purpose Python containers like dict, set, list, and tuple. Kind of like the Dark Knight is the more interesting “implementation” of Bruce Wayne.

burger

Defaultdict is subclassed from the built-in dict module. You may have encountered the following common uses cases for which you have been using the default container.

Building nested dicts or JSON type constructs:

JSON is a very popular data structure. One of the major use cases for a JSON is creating web APIs. JSON also neatly corresponds to our dict object. A sample JSON object could look like this.

Source:http://json.org/example.html.

We cannot create a json file by using the following command; it will throw a KeyError.

So, we will have to write complicated error handling code to handle this KeyError.

This way of writing is considered un-Pythonic. In its place, try out the following construct.

A defaultdict is initialized with a function (“default factory”) that takes no arguments and provides the default value for a non-existent key. A defaultdict will never raise a KeyError. Any key that does not exist gets the value returned by the default factory.

Please ensure that you pass function objects to defaultdict. Do not call the function, that is, defaultdict(func), not defaultdict(func()).

Let’s check out how it works.

Having cool default values:

Another fast and flexible use case is to use itertools.repeat() which can supply any constant value.

This should print out “John ran to.” As you can observe, the “object” variable gracefully defaulted to an empty string.

Performance:

Like you see in this stackoverflow post, we tried to do a similar benchmarking only between dicts(setdefault) and defaultdict. You can see it  here: https://github.com/infinite-Joy/hacks/blob/master/defaultdict_benchmarking.ipynb

Below is the output that I got on my machine using Anaconda.

Following are  the broad inferences that can be made from the data:

1. defaultdict is faster and simpler with small data sets.
2. defaultdict is faster for larger data sets with more homogenous key sets.
3. setdefault has an advantage over defaultdict if we consider more heterogeneous key sets.

Note: The results have been taken by running it on my machine with Python 3.5 implementation of Anaconda. I strongly recommend you to not follow these blindly. Do your own benchmarking tests with your own data before implementing your algorithm.

Now that we have discussed the DefaultDict module, I hope that you are already thinking of using it more and also refactoring your code base to implement this module more. Next, I’ll be coming up with a detailed discussion on the Counter module.

References:
stackoverflow, How are Python's Built In Dictionaries Implemented
stackoverflow, Is a Python dictionary an example of a hash table?e
python.org, Dictionary in Python
python.org, Python3 docs, collections — Container datatypes
python.org, Python2 docs, collections — Container datatypes
accelebrate, Using defaultdict in Python
wingware, Python Library Reference

About the Author

Joydeep
Joydeep Bhattacharjee is a Category Head (Python) at HackerEarth. He likes to dabble in all things tech and is passionate about open source.
Share