the-asir / mlt-x-fastai

Fast.ai study sessions organized by MLT.

MLT-x-fastai

Welcome to the MLTxfastai study sessions. This post is to help you get started with the fastai MOOC and prepare for our study sessions.

Study Sessions

The goal for these study sessions is to get help with the course materials and most importantly, focusing on building cool stuff with what we learn.

We will be reviewing some concepts from the lessons from time to time but it will mostly be a free format. You should plan on how you want to spend these sessions. We recommend you watch at least a video or two ahead of the sessions and start thinking about what you want to build to get the most out of your time.

Getting setup

There are quick guides on how to get started with the course materials on the course page. We recommend you use colab or kaggle kernels since they are free to use. Whatever platform you choose, its important you pick the one that's most comfortable or familiar to you. We don't want to be stuck setting up environments rather, use our time to get our hands dirty writing code

How to do the course

A few helpful guidelines on how to get the most out of the course:

• Watch the videos. But its important to understand that watching videos is not the same as practicing machine learning. Neither is pressing Shift+Enter at every line of code from the lesson notebooks. We suggest you watch a video and try to solve the problem from the lesson with recall. Its okay to take peek when you're doing this, eventually you won't need to.
• fast.ai teaches from a top-down approach which means you learn how to build something first and the specifics behind it later. This can be a bit frustrating at times as most of us come from a traditional bottom-up style (lots of theory first and practicing later). It’s okay to not know everything. No one does at first, you'll pick a lot of practical skills that will help you with theory later. And that's also why we have these study sessions!
• Have a project to work towards. It will make your learning engaging if you do so. A project can be anything you're interested in. If you're a domain expert, think about the problems in your are of expertise. How can deep learning help with solving those problems? It doesn't need to change the world, it just needs to be something you care enough about to want to write a little bit of code everyday.
• Join a community (MLT!!) and find people to collaborate with. Not only is there is a lot to learn by hearing someone else's take at a problem, it is also very motivational and will help you stick with your goals.

What can you build from lessons off a MOOC?

There is an entire thread in the forums dedicated to what students have created from the lessons. We'll highlight a few to give you an idea and some inspiration

• Classify spots in Tokyo
• Sign Language Predictor
• Kuzushiji Classifier (Advanced)
• Colorize old images (Advanced)

Advice from Jeremy Howard

Twitter thread by @math_rachel https://twitter.com/math_rachel/status/1104506606740430848

• Don’t try to stop & understand everything.
• Please run the code, really run the code.
• Pick one project. Do it really well. Make it fantastic.
• It’s okay to feel intimidated, there’s a lot, but just pick one piece & dig in.
• If you’re stuck, keep going.
• The answer to the question “Should I try blah?” is to try blah and see, that’s how you become a good practitioner.
• Don’t be intimidated. It’s meant to be intense.
• Don't waste time getting a ton of data. Start small.
• Perhaps the most important is to get together with others. Learning works a lot better if you have that social experience. Build things. It doesn’t have to be amazing. Just finish something. -- @jeremyphoward

Accompanying Thread from forums

Feedback Form:

Please leave your thoughts and comments by filling out this short survey