This past week, I started my Machine Learning journey, and more specifically, my Deep Learning journey. I started by watching the 3Blue 1Brown videos about Neural Networks which I found to be a very good introduction to Deep Learning. Then I remembered finding, about a year ago, the Fast.ai website. I went to revisit the website and found the course Deep Learning for Coders (part 1 and 2). Intrigued, I looked at the course introduction, and the learning method seemed interesting: a top to bottom approach, meaning that you start by training models and learn as you go. You don’t have to follow math courses for 2 years before starting. You only need a basic knowledge of coding (Python preferably).
I’m game! 🙂
This doesn’t mean that there is no math. It doesn't mean that the coding is basic. It doesn’t mean that it will be easy. But it will be more applied than other courses. Here is what I learned from lesson 0, which explains how to use the course, and from visiting the website.
How to finish the course
It might seem like a weird advice. Why start a course about something you want to learn if you don’t plan to finish it. But the subject is challenging, and many people just drop out of it.
- It is recommended to make yourself accountable by telling people about taking the course and about your projects.
- Have a plan: when will you watch the lectures? Do the assignments?
- Join the forum so you have a community helping you when you’re stuck.
- If you need more coding knowledge, consider watching the CS50 and the Missing Semester videos on YouTube.
- It will take tenacity: you may meet some obstacles, you may have to stop for a while, but make sure to come back as soon as possible.
How to do well in the course
This is a very challenging course. One lecture may take more than one week. Remember that your ultimate goal isn't to finish the course in record time, it's to get as much knowledge as you can from it!
- Watch the lecture and/or read the corresponding chapter of the book.
- Run the Jupiter notebook and play with it (the course explains the different platforms you can use to run the notebooks including free ones). Experiment with the code: try to change values or the order of the lines, for example.
- Then reproduce the code on a new notebook, from scratch.
Note: You may have to go through these three steps more than once.
- Repeat with a different set of data.
- Ask questions. You may think that your question is stupid, but it will surely be useful to many people including you, of course.
- Start a great project and publish it! Now, it doesn’t have to save the world to be great. It can be something pretty small. Or something more important. Your choice.
Connect with others
I've already talked about the Fast.ai forum which I believe is an amazing tool. Being part of a community, especialy when studying something as challenging as Machine Learning, is vital. Here is a list of other places and ideas to meet fellow data scientists, stay motivated, learn and have fun:
- Twitter, of course. Tech Twitter is my favorite place! There's a lot of great people wanting to connect and help. You can find me there as MarieTKD.
- Kaggle: It is highly recommended to participate in the competitions, even though you will do terrible at first. 😆 You can make a team and work your way up the leadership. Kaggle also has a forum where you can ask questions and network. Reading posts from top Kagglers will help a lot.
- Blogging, start a podcast, make videos, go live on Twitch... Teaching what you just learned will not only help others understand the concepts, but it will help you having a deeper understanding as well. Teaching is the best way to learn. It's also important when job hunting. Companies are looking to hire people with good communication skills! If you're not convinced, I recommend this article by Rachel Thomas, co-founder of Fast.ai.
I truly hope this article will motivate you to start your own journey.
I also truly hope mine will go well. 😟 The only way to know, is to start! Will you join me?