Starting new project - recommended way to include the latest FastAI from the course?

I am starting a new work project tomorrow for some classification work, but what is the best way to leverage/use all the code from the part 2 course going forward?

Do we just use the notebook type includes and go with that as it’s own codebase (i.e. future upgrades will be propogated through those notebook exports), or has the code from the course already been rolled into the current FastAI github library and I should just run with main FastAI now (dev update?) and we can thus leverage things like XResNet there?

Any advice here would be appreciated so I can get this project off to a proper start, and ensure compat as FastAI progresses.

1 Like

Looking at your questions, my advice is to wait a few weeks for Fastai v2 roadmap .


Be aware that the fastai library frequently breaks backward compatibility which leaves you with two options:

  1. Choose a version (eg. I think the current release is 1.0.52) and stick with it through the lifecycle of your application.
  2. Be willing to fix breaks when they occur between different versions.


1 Like

Thanks very much - this advise was extremely helpful! (@fabris and @JoshVarty)

I’m in the same boat. Would love to start a new project ASAP but am a bit reluctant to use the old v1.0 style fastai style after doing the 2019 course…

Would anyone venture a guess if v2 is weeks or months into the future?