Specialized container datatype 'named tuple' in Python

Dr. Rahul Remanan

​CEO, Moad Computer

In addition to the general purpose container datatypes: dict, list, set and tuple; Python also implements specialized container datatypes. One of the commonly used specialized container datatype in Python is the 'named tuple'. 

​A 'named tuple' can be created in Python using the collections module factory function namedtuple(). This function returns a tuple like object whose elements can be accessed using the corresponding attribute name. Hence the name of this tuple like object: 'named tuple'.

The attribute names or field_names for the 'named tuple' can be either a sequence of strings (Eg. ['x', 'y']) or a single string with each fieldname separated by whitespace and/or commas (Eg. 'x y' or 'x, y').

Of note here, while assigning a fieldname; they should not start with a digit or an underscore. However, both digits and underscores are allowed elsewhere in the fieldname.

A common application for the 'named tuple' is to create more readable vectors such as those representing the Cartesian coordinate space or for representing the latitude-longitude data from a GPS receiver.

Creating and working with 'named tuples' in Python

Import libraries

Create a 'named tuple'

Python 'for-loop' to manipulate the 'named tuple'

Tensorflow 'while_loop' to manipulate the 'named tuple'

The same operations performed in the Python 'for-loop'  is replicated using Tensorflow 'while_loop'. The slicing of tuple elements is applicable for the 'named tuples'. Therefore, a 'named tuple' provides the convenience of accessing its contents using both field_names as well as their position index.

Performance optimization of the 'named tuple' operations in Tensorflow 'while_loop' 

Described below is the a quick code-snippet on performance optimization for the 'named tuple' operations. In this example, a 'named tuple' is updated using the Tensorflow 'while_loop' by wrapping the operation as a decorated function.

References

1.  Python docs on Collections module
2.  Tensorflow docs on working with while_loop
3.  Stackoverflow post on Tensorflow while_loop performance optimization

​