Machine Learning Nanodegree Experience
By — Divyesh Peshavaria
Motivation
I was always interested in Artificial Intelligence from an early age. It’s potential and applications filled me with an inspiring awe. Adaptation of Artificial Intelligence in Hollywood like Iron Man and I, Robot showed great promise about the wonders of AI to a me as a kid. But, I had no idea on how to get started in this field practically. I first came to know about Machine Learning when Google open sourced TensorFlow in November 2015. I tried to understand the framework myself but could not make much progress at first. I had previously done a few free courses from Udacity and enjoyed their teaching style and content quality. I had been coding in Python since 2015 and fortunately it turned out to be the language of choice for Data Scientists and Machine Learning Engineers. Therefore, I finally enrolled in Machine Learning Nanodegree in 2017 after self-learning the libraries NumPy, Pandas and Scikit-learn for a few months.
Components of the Udacity Nanodegree
- Video lectures and exercises from Udacity’s website
- Forums
- Slack Community (the best part)
- Mentoring up to 2nd project
- Projects on GitHub
The content of Udacity Nanodegree program is a perfect balance of equal parts theory and practical knowledge. The lecturers first explained theoretical concepts related to Maths and Algorithms, and then showed how to implement them in code.
The video lectures were intermittently followed by mini-exercises which served as points for revision and milestones in learning progress. If I ever got stuck on any topic, the Forums on Udacity and Slack team were always there to help. This was indeed the best part of Nanodegree experience for me. The Slack team has thousands of talented individuals who were either completing the Nanodegree or graduated from it. Therefore, someone was always there to help, no matter what time you ask a question. And instead of spilling the answer directly, they help you guide yourself towards it which is a great learning experience. Being a part of such an awesome community was a new and exhilarating experience for me.
To keep you on track, you are assigned a mentor until you complete your project. The mentor is available for 2–3 days/week. They are people with sound knowledge in Machine Learning and help you in every aspect to progress forward. My mentor guided me a lot when it came to complete my project. The initial boost gained by them helped me to complete the Nanodegree even after their unavailability.
The submission system for MLND allowed for two options: -
- A zip file containing all the required files
- A GitHub repository containing the same
I always preferred submission through a GitHub repository and I urge anyone reading this to do the same because it shows all your work (including your mistakes and how you fixed them) to others and also builds your portfolio.
For each project, there is a rubric stating the requirements clearly. The submission format was of Jupyter Notebooks. Earlier I was accustomed to using Python scripts for Machine Learning through the course of Nanodegree, I fell in love with Jupyter Notebooks which provide a perfect way to blend Code, Visualization and Documentation in a single file. I now use them on a regular basis and prefer them over plain scripts whenever needed.
Reviewers — The stars of the Nanodegree
Going through my project reviews felt simply awesome. Whenever, I failed to meet any requirements, the reviewer mentioned it clearly and showed possible workarounds. The project review was completely mapped to the rubric and hence it was super easy to zero in on the errors. Also, even when I had completed a requirement correctly, the reviewer discussed other possible methods to achieve it, which helped to expand my knowledge a lot.
Curriculum
The MLND curriculum consisted of the following sections when I graduated from it: -
- Introduction to Machine Learning — Titanic Survival project
- Model Evaluation and Validation
- Supervised Learning
- Unsupervised Learning
- Reinforcement Learning
- Deep Learning
- Capstone Project
One doesn’t need to complete them in the order specified, but this order would be beneficial if you are completely unfamiliar to Machine Learning. In fact, I attempted to complete Deep Learning section after Model Evaluation and Validation but after a failed try I returned to the original order. This only proved useful because by the time I sequentially reached Deep learning, it was more comprehendible. I got hands on experience of Keras library which was previously not a part of the project.
The Capstone project is a nice inclusion to the curriculum. It gave me a brief idea about the entire process which Machine Learning Engineers follow while solving a problem. It allows you to choose a problem/dataset of your choice and complete the following steps: -
- Problem scoping
- Data Collection
- Data Visualisation
- Data Preprocessing
- Feature Engineering
- Model Selection
- Model Hyper-parameter Tuning
- Model Testing and Deployment
And the best part is that even though your idea is not ground breaking, or a convention defying research you still get acquainted with the entire Machine Learning process, which is a crucial factor for preparing you for a job in Machine Learning.
My Key Takeaways from the Nanodegree
Photo by Aaron Burden on Unsplash
- A solid theoretical of Machine Learning, its types and their applications
- Practical knowledge of implementing Machine Learning models in Python and best practices in Machine learning
- An opportunity to be part of an amazing community of learners from around the world
- First job out of college as a Data Scientist
- A passion for Lifelong learning
Enroll into Machine Learning Nanodegree here
About the Author | Divyesh Peshavaria
Pythonista. Love AI and Computer Vision. Passionate about using Technology for solving problems.
