Aug 27, 2020

FastAI Deep learning for coders:

Course Link: https://course.fast.ai/videos/?lesson1=

Note: I had started with Machine Learning course from FastAI Link: https://course18.fast.ai/ml.html

Lesson 1:

Summary:

• Myths and actual truth around practical Deeplearning.
• Top down approach for practical DL over bottom up approach followed in academia.
  • Base ball analogy for the same.
• Libraries covered Python -> Pytorch -> FastAI.
• Jupyter Notebooks:
  • REPL (Read Execute Print Look).
  • Similar to shell/terminal.
  • More interactive and fetures (GUI, rendering, etc).
  • Shifting from IDE to Jupyter env feels exactly like shifting from GUI to terminal.
  • We stick to terminal over GUI so stick with Jupyter.(Whole book, library, docs was written in Jupyter notebook).
  • Each instructions gets executed in server, changes server state and the output is displayed back. But most importantly displayed output shows the state at the time it was executed not the recent state.
• Traditional ML vs DeepLearning.

IMP:

• Each chapter is in each notebooks.
• Notebooks in the end has questions section instead of summary. The idea is to encourage us to revise and to be able to answer them. They have spent weeks in writing these question sections and assumption is you know them when you move into subsequent chapters.
• Links:
  • Course Home : https://course.fast.ai/
  • Course Content GitHub: https://github.com/fastai/course20/tree/master/
  • Course Book Chapters: https://github.com/fastai/fastbook
  • Forums for Q&A: https://forums.fast.ai/