This is regarding week 2 in course 2.
I started the mobilenet assignment with the latest version of tensorflow i.e. 2.6.
There were a lot of hickups in passing the test cases.
Creating a new conda environment with tensorflow==2.3.0 and jupyterlab didn’t help either.
So, I pasted the code in coursera jupyter environment. All tests passed.
Could you please provide an environment file for coursera environment?
Here’s one way to share your environment.
The course intends for students to use the Coursera platform.
Yes, there are no instructions for how to set up your own local environment to run the notebooks. Why is it not good enough to run them on the website as intended?
But if you want to play these games and know how to work conda, you can execute the conda sharing commands in the Coursera environment. Either use “bang escapes” to run linux shell commands in a cell in the notebook or click “File → Open” and then “New → Terminal” to get a linux shell in the Docker image that is running the notebook.
Here’s an earlier thread with an environment file from another student and some discussion on these points.
@paulinpaloalto Thank you for the pointers. I’ll check the environments you pointed out.
That said, it would be nice to make these tests pass on tensorflow 2.6. Course 4 Week 2 Assignment 2 (MobileNet) is the first assignment in the specialization that doesn’t work locally.
If I wanted to try something out after my subscription ends, isn’t doing it locally the only option?
You don’t have to run locally. You could use Colab, which supports Jupyter Notebooks as their fundamental interface. That said, I don’t know how they cope with the “versionitis” issue.
But the other point is that you can’t expect the course staff to spend their days updating everything to the very latest version of all possible packages. Things change all the time. It took them > 2 years to get around to switching from TF1 to TF2 (which just happened in April 2021), so I don’t think you can expect another version upgrade anytime soon. Well, maybe that’s not quite a fair comparison in that I’m sure the change from TF 2.3 to TF 2.6 is not as big a deal, but the general point stands: they’ve got more useful things to do with their time than reissue their software everytime TF (or matplotlib or numpy or scipy or yadda yadda yadda) changes version. If you’re going to play this game, you need to learn how to use Anaconda. Fixing “versionitis” like this is exactly what it’s designed for. That said, I’m not sure whether it covers TF. I guess you’ll find out .
What’s the procedure for me to fix source code for them to review and accept it in the course?
Sorry, but there is no official procedure for students to submit code fixes. The best you can do is to describe the problem and your proposed fix in a Forum post here. If your post is convincing enough, then we can call it to the attention of the course staff or file a bug on your behalf.
Alright. Thanks for letting me know.