Data Science Portfolio

If you have read my beginners guide to building a data science portfolio then you already know how important it is to find a way to demonstrate your skills, especially if you don’t have any work experience yet. The three most popular platforms for building a portfolio are Github, Kaggle and a personal website or blog.

At the end of this blog post you should be able to pick one (or all three) of these platforms to host your portfolio.

In my opinion, all three of these aspects are required to build a good portfolio that will get you hired even without any work experience. Work on projects (either your own project or from a competition), push them to a Github repository and then write a clear, easy to follow blog post that tells a story or teaches someone something.

You can even combine your Github repo with your blog by building a static site. There are a number of great articles out there that show you how to do this (this is one of my favourites).

Github

Github is one of the most popular platforms to store code and build a portfolio and it is used by developers all over the world. In addition, git is a must-have skill for all data science and developer roles.

Pro’s:

• It is pretty easy to set up your projects on Github and most people are already familiar with the platform so potential employers will be comfortable with navigating around your repositories
• Public repositories are free
• Editing code is simple – commits are cheap
• Github lets people see what you have built and how you have built it.

Con’s:

• Since Github is so popular, chances are that there will be other candidates who will also be using the platform. This means it can be a bit more difficult to differentiate yourself from the crowd.
• Github will show anyone viewing your profile all of your contribution activity. This means you must remain active on the platform because it will be off-putting for potential employers if they see you haven’t actually done anything in a while.

Lastly, the README file is probably the most important part of your Github profile. Github automatically parses the Markdown format of your README and renders it on the front page of your repositories. It is the first thing anyone will see when they land on your Github profile so you need to make sure it stands out.

Blog

Hiring managers who land on your personal website or blog will see you going the extra mile and this can be eye-catching and impressive.

Pro’s:

• A blog or personal website can help you to stand out and help to get your foot in the door for a first interview
• In addition to portfolio projects, you can write detailed blog posts, deepening your own knowledge while also helping others to understand complex topics
• A blog allows you to create a data-based story or interactive project that can make quite the impression.

Con’s:

• Building a personal website or blog takes a lot more time and skill than simply pushing code to a Github repo.

Kaggle

Kaggle is one of the most popular data science competition platforms. Not only can you compete in competitions with fellow data scientists but you can also publish Kernels on the platform and engage with other users on the forums.

Pro’s:

• Kaggle competitions can be a lot easier than a ‘real-world’ projects since they take care of coming up with a clearly defined goal, get the data for you, and even clean it into some usable form.
• A lot of new and experienced data scientists post notebooks and kernels on Kaggle, describing their process and you can learn a lot from them
• Actively contributing in the forums can help you to make a name for yourself which can potentially bring new opportunities

Con’s:

• Having the goals and data handed to you is also a con since you would not have demonstrated your expertise in defining a problem and cleaning data, which are some of the most important skills to have in data science

Remember that one repository, one project, or one competition does not make you a data scientist. You need to need to show that you are dedicated and passionate and that will require you to do multiple projects, continuously improving and iterating on them over time.

The projects in your data science portfolio should show that you have mastery in a particular area within data science. At the end of this post you should know how many projects to include as well as the types of projects that should make up your portfolio.

You can demonstrate mastery by following these two guidelines:

• Choose a specialty. By specialising in a specific area, you will be able to obtain mastery much faster and you will get noticed as something of an expert.
• Demonstrate domain knowledge. Are you looking for a position in marketing, finance, health, etc? Tailor your projects and data around the areas that your job or client are expecting you to have knowledge in.

Once you know what your specialty is and what area you will be applying your skills to, you then need to know how many projects you need to have in your portfolio.

For this, I recommend that you focus on two types of projects: pillar projects and supporting projects

Pillar Projects

A pillar project should be able to clearly demonstrate one or more core skills from your specialty and domain. These projects tend to be larger and more in-depth than a usual project.

A portfolio should typically have 2 pillar projects.

All aspects of this project will depend on your chosen specialty and domain and can vary greatly from person to person.

There is no one-size-fits-all here because some domain areas lean heavily on a few specific techniques which you should demonstrate thoroughly in your pillar projects.

While the types of projects can vary a lot, it is recommended that you produce some kind of interactive experience for one of your pillar projects. You can differentiate yourself from the crowd immediately by building an interactive interface that your potential client or employer can view on their PC or mobile either in their own time or during the interview.

Your project should be coupled with a compelling narrative that walks the reader through the project and specifically highlights the details of the core skill you are demonstrating. In addition to a write-up, you can also record a short video that summarises the key points of your project.

You should know your pillar projects very well and should be able to describe the overall goals and outcomes of the project in a few sentences.

Supporting Projects

A portfolio could have anywhere from 2 to 5 supporting projects.

Your supporting projects should demonstrate other skills that you may not have demonstrated specifically in your pillar projects such as:

• Data cleaning – in a data cleaning project you need to show your data munging, data preparation skills on a very dirty data set. The best way to put together a data set for this is to scrape it yourself – pick a website that interests you (such as a sports site, if you’re into that sort of thing), scrape it, clean it, and have some fun with it.
• Data visualisation – the purpose of a data visualisation project is to tell a story. You can do this in an R markdown or Jupyter notebook but I recommend that you try a platform like Tableau Public or Google Data Studio and design some dashbards that take the reader through the story in the data.
• Data modelling – this project would contain either a supervised or unsupervised model, depending on the focus of your pillar projects.

If you have little to no experience as a data scientist, then you have to find a way to demonstrate that you have the skills to do the job.

You can achieve this with a portfolio and this is arguably the most important part of your application.

However, working on projects and putting together a portfolio is about more than just the job application.

Through the process of working on data science projects you will learn and grow as a data scientist, exploring new and exciting methods and techniques and expose yourself to new ways of thinking about problems and how to solve them. You will get to know the common packages and functions as you use them and you will eventually be able to write code without needing to search on Google every few minutes (this one is particularly satisfying).

You will build confidence in yourself and your abilities and this will show when you are sitting down for an interview.

The key pillars of a data science portfolio

• Communication – you should be able to clearly communicate each aspect of your project and then present the final work in an appealing way that captures the readers attention and tells a story.
• Collaboration – you should participate in a group project at least once to show that you can work well in a team.
• Critical thinking – your projects should demonstrate your ability to reason with data, generate questions, and discover patterns in your data.
• Competence – your projects should demonstrate that you have the right skills for the job and that you can do them well.

The types of projects to include in your data science portfolio

Always aim to stand out in the crowd; be memorable.

Don’t re-use common data sets as all this shows is that you can follow a tutorial.

Find unique, interesting data sets for your projects that you are genuinely curious about: – Scrape it off the web (Wikipedia is a goldmine for this) – Conduct surveys with friends, family, students, or faculty members of your university (if you get creative here you can generate a pretty interesting data set) – Use real data rather than fake, simulated data sets and come to real conclusions about your data

See my post on finding data for a couple of great places to start in your search for data.

Your projects should show that you have mastery in a particular area within data science, so consider these two guidelines when choosing what projects to include in your portfolio:

• Specialty – what area of data science will you specialise in? By specialising in a specific area, you will be able to obtain mastery much faster and you will get noticed as something of an expert. For example, there is Natural Language Processing, financial forecasting, or visualisation and storytelling. Do your research and find the specialty that interests you and start there.
• Domain – what domain knowledge will potential employers expect you to have, even at a basic level? Are you looking for a position in marketing, finance, health, etc? Tailor your projects and data around these areas.

Where to publish your first data science portfolio

When you’re just getting started with your portfolio, I recommend that you create a Github account and publish all of your projects there. Check out this tutorial to master the basics of Github as a data scientist.

The most important part of each of your Github repositories is the README. This is what readers will see first and they will likely not stick around to sift through hundreds of lines of complicated code, trying to figure out what you were trying to accomplish.

A README is written in markdown. I recommend this guide for getting started with the markdown syntax.

Your README should clearly show these important aspects of your project:

• The goals and success measures of your project
• The techniques and methods you use
• The key findings or end result of your project
• How to reproduce the analysis

A portfolio should consist of around 5 core projects that you regularly go back to – making improvements, applying new methods, or scrapping the project entirely in favour of a bigger, more complex project to suit any new found skills.

Data science projects will undoubtedly enable you to grow as a data scientist both when you are on the job hunt and even while you work at your dream job.

Growth and learning does not stop when you get the job. As a data scientist, you will always be learning and data science will always be growing.