This second part of the ‘Python for Data Scientists’ post talks about the specifics of Python for data scientists. Part 1 of Python for Data Scientists talks about Python generally and can be found here.
Python for data scientists
Where you can use Python
Python is a general purpose programming language meaning that is has many use cases outside of data science. These include game development, graphics, web development, GIS, and control systems. A benefit of this is that data analytics undertaken with Python can easily be integrated with other tools. Python is an easy language to learn and the capability of Python for advanced analytics is quickly growing. Here’s a data visualisation video that uses Python:
Key packages for data science
The packages mentioned in Part 1 are all used extensively for data science. Additional useful packages are:
- Matplotlib - data visualizations
- ggplot - graphical plotting based on R’s ggplot2
- Seaborn - statistical model visualizations
- Bokeh - interactive visualizations
- Plotly - Web based toolbox for web visualizations
- SciKit-Learn - machine learning and image processing
- Theano - machine learning framework
- Keras - high-level neural networks
- NLTK - Natural Language processing
- Scrapy - web crawling framework
- Statsmodels - estimation of statistical models and conducting statistical tests and data exploration
Important things to learn
Learning to use the previously mentioned packages is essential for using Python for data science.
Python is a dynamically-typed language, this means that a user does not need to define what ‘type’ a variable is (e.g. numbers or text) when they create a new variable; this is in contrast to statically-typed languages such as Java. A dynamically-typed language is generally less verbose meaning it can be written, read and maintained in a shorter time frame. However errors can be created if Python assigns variables as the wrong ‘type’ and the user does not check this. In addition type errors are not checked at the start of execution (as they are for statically-typed languages) and so type errors in large programs are only found after wasting time running the first half of the code.
Iteration is an important feature in Python which contrasts to vectorisation in R (if you come from a R background). For example if you want to add the elements of two lists in Python together, you need to loop through the elements of each list and individually add them (iteration). In R you can simply add the two lists together using one line of code. The Python package NumPy supports vectorisation. The fundamental idea behind vectorisation is that operations apply at once to an entire set of values.
Using Python with virtual environments is another key concept to learn about. The main idea here is that the way that Python packages are stored on your computer can cause problems when working on multiple projects. A way to get around these problems is to use virtual environments. More information on this can be found here and here
Learning resources
- Python Data Science
Handbook -
this is one of the most popular books on Python for data science
- Data Camp - provides a free Python for Data Science introductory course
- Kaggle - join a high profile data science community, learn new skills and take part in competitions
- Pivigo - a data science community to learn new skills and network
- Python Regular Expressions - regular expressions are very useful for data cleansing