Lesson 1 of Fast.AI 2022 Edition
The first lesson is very much about “playing the whole game” (insert link): Without going into any details, we are creating a computer vision classifier - isn’t that cool!
My learning approach
After watching the video,
- My first exercise was to play with Jeremy’a notebook “Is it a bird?”. After running it on the Kaggle platform, I also ran it on my local computer, here is the resulting notebook
- To go beyond the “Is it a bird?”-notebook, I decided to build a model which can identify if a picture shows a cat or a dog. Somehow I felt, that is a mandatory exercise for a ML student ;). Here is the resulting notebook: “Is it a cat or a dog?”.
Anything beyond that, I postponed to later, knowing that the course will dig deeper in the next lessons anyway.
Lessons learned
The actual implementation in analogy was not too hard, but there were some challenges on the way. Here are some of my learnings:
- “Cat or Dog” is more difficult than “Bird or Forest”: I had to experiment a bit with the dataset until it really produced good results, i.e. to get the test images to be classified with enough confidence.
- Using my local machine for development and training is a challenge of its own. No wonder that Jeremy had discouraged students from working locally in the previous version of the course.
While doing the “Is it a bird?-exercise, somehow I had the idea that I needed to update pytorch, which I seem to have messed up big time (not knowing exactly why…). Anyway, after a bit of trial and error, I decided to re-install my python installation. Sounds difficult?
Actually not! The Live Coding Video 1, the forum notes, and FastSetup made this really easy. Since I had watched the video before, I was up and running again less than 30 minutes.
- I work with WSL, but the git repos still reside in Windows on OneDrive. This has so far proven to be very useful
- I like to write markdown in VS Code, because it also has a spellchecker.
- When my WSL instance was in trouble, I never worried about the git repos.