Hi, I took the Deep Learning Specialization courses twice and during that time versions of dependent libraries don’t seem to have changed much. For example, when I completed it August last year the source code was not compatible with Tensorflow major version upgrade. The same goes for a couple of other important libraries.
Is there a version of the source code with more recent versions of dependent libraries? At least compatible with major revisions. If not are there plans to make this available anytime soon, if so when?
I am not aware of any plans to upgrade any of the DLS code to match current versions of TF or other packages. The intent here was purely educational, not to be creating some kind of code base that you can use in real applications. Remember that the Terms of Use here are for your “personal use” only.
In the ML world, things change pretty quickly. The last major upgrade to DLS was released in April 2021, when they switched from TF1 to TF2. Since then things have continued to evolve in the world at large, but not here.
They have given you a consistent and functional environment here that serves their pedagogical purposes. If you want to repurpose the code, then you have some work to do to figure out how make that happen. You have two fundamental choices:
- Create an environment in which you can duplicate the versions of all the packages that are used in the courses. Anaconda or Conda are built for that purpose and there may be IDEs which also support that idea.
- Just work through the problems as you encounter them and figure out how to adapt the code to the current APIs.
Neither path is easy. Some tradeoffs that I can think of would be:
Path 1 is a lot of work up front, but Anaconda was originally created for an important reason: you can hit the “versionitis” problem anytime you import some complex application. Once you learn how to use one of these tools, then you have a general solution to versionitis.
Path 2 may be less work in aggregate, but it’s also unpredictable how much work it will be. It’s also not a permanent solution: you may get things working, but they upgrade TF next month in an incompatible way.
This general topic has come up many times in the past.
Here’s a thread which gives an introduction to path 1).
Here’s another thread about path 1) that mentions some other roadblocks if you are running on M1 Mac h/w. Don’t just read that one post, but follow the thread.
1 Like
No, there is not.