How To Not Do fast.ai?

The Seemingly Impossible & Perfect Study Plan

For each lesson:

  1. Head to the course website and begin from there.
    1. Read the overview of the lesson.
    2. Try to understand it.
    3. Google for unknown terms, just to get an idea of “What Is It?”.
    4. Create a mental model of what the lesson might be about.
  2. Head to the lecture video.
    1. Watch it first time.
      1. In an organic fashion, without stopping, not worrying much if you don’t understand.
      2. Connect how it matches with the lesson overview you read earlier.
      3. Know what is being said in the lesson, just the topics.
      4. Perhaps, make an outline.
      5. You should know “What Is There In This Lesson?”
    2. Watch it second time.
      1. Pause and watch.
      2. Open a copy of NB and run each cell with Jeremy.
      3. Take Notes.
      4. Search Google for unknown stuff.
      5. Ask questions on forum.
    3. Watch it third time.
      1. With your notes in front of you watch the lecture one last time to make sure you have not missed any words Jeremy said.
  3. Head to the main NBs, Jeremy wrote.
    1. Read the pros and understand them.
    2. Read them again.
  4. Head to the documentation
    1. Read about the functions used in the lesson.
    2. Try to tweak.
  5. Head to the forum
    1. Read everything others have asked.
    2. Take notes.
    3. Try to answer if you can.
  6. Head to the textbook. Read the lesson 3 times, similar to how you watched the lecture.
  7. Spend alone time the NB.
    1. Experiment & Tweak stuff.
    2. Try new datasets.
    3. Re-write from scratch.
  8. Go Beyond
    1. Go through all the sources(books, papers, etc.) mentioned in the lesson and within the chapter.
    2. Click every resource link and go through it!
  9. Read the heavy source code behind the documentation.
  10. Do your own project/Participate In Kaggle/Start A Company/Join OpenAI/Write a Paper

After 100 years you will know everything you could have ever imagined?!

Did anyone try this approach?

6 Likes

Yep :slight_smile: My 10 was then take all those notes and re-teach them with Walk with fastai

1 Like

Oh Gosh!!! Thanks for saying this. I thought I was a bit crazy.

It’s certainly a longer (and in some ways harder) route, but if you stick to it you will come out well :slight_smile:

1 Like

Did you miss out ‘join a study group’? :grinning:
It can replace at least one of those ‘read it again’ sessions, it can be good for review, more fun and generally helps keep commitment to a regular schedule.

3 Likes

Though it is okay not to, study groups aren’t for everyone :wink: (But do try it though! More often than not they help!)

2 Likes

This does look good!

Some things I would tweak (based from my experience) :slightly_smiling_face: :

  • don’t watch the video lectures for the third time, except for certain portions which you feel like you don’t understand. The same goes for the book chapters. I say this so that you spend more time doing, and less time on the fine details.

  • don’t focus on taking notes. The video lectures and lesson resources will always exist, so you can always refer back to them. Take notes only to capture the main ideas, not the fine details. Again, this is so that you spend more time doing. That said, do summarize what you have learned, as this is a great way to check if you know/understand what you’ve learned.

  • don’t get bogged down in the documentation. This is not to say you shouldn’t read the code behind certain functions, classes, and method or that you shouldn’t experiment with them. But again, it’s so that you spend more time doing. You’ll gain an intuition/idea of certain functions/classes/methods just by using them, or by gradually reading more and more of the documentation as you do more projects or encounter problems.

6 Likes

what’s the meaning of this? Is this post meant to satirize that learning fastai takes too long or is not very realistic? I am a beginner and am trying to learn deep learning through the course offered by fast.ai. Can anyone help to answer, thanks a lot.
(没懂作者发这个帖子的意思)

It was meant to satirize that learning fast.ai the way I had mentioned above in a very exhaustive way might possibly take too long.

Hope that clears it up!

Check the following get some tips as a beginner:

1 Like

I would do everything but in reverse from 10 to 1

1 Like

That would be the ULTIMATE top-down approach! :sweat_smile:

1 Like

Personally i don’t see the point of doing it any other way. Fastai is just a tool. Some people fuss over tools. I prefer to use them to build stuff

1 Like

Thanks for your reply. I got it. I think the material is really great. :slight_smile:

1 Like