Fast AI Deep Learning Course – Raw Notes

Session-1

There were several cool things at the FastAI deep learning course. Many of the ideas are worth checking out and digging deeper.

A few thoughts:

  1. One of the easy to copy ideas is the fastai/cups. I am going to copy that right away.
  2. It is amazing how many uses they found for Jupyter notebooks. I have known it as a great teaching tool, but the idea of writing a book or creating a blog post blows my mind.
  3. I got some cool references (books) to teaching techniques. Can’t wait to learn them.

Notes

  • FastAI is built on top of PyTorch
  • The mouse click image generator is cool
  • The slides are in Jupyter Note books using a JN extension called REST #todo: Check out REST
  • There are several notebook servers including the one from Kaggle
  • Jerome prefers functional style of Python programming (time to revisit sicpy)
  • JN as a blogging tool (fastpages). Need to checkit out (perhaps for pyskills)
  • The diagrams on how ML works (attributed to Arthur Samuel) are a clever hack using GraphViz and JN). Need to check it ou.
  • The entire course is on github. So I can go there and fill up these notes

Applications

  • Tabular Analysis caught my attention. There were many applications including a whole set of NLP Tasks

Concepts

  • Data blocks,
  • Data loaders
  • Learners (cobines models and data)