Will Trax be required for all programming assignments for Courses 3 and 4?

Am I going to be forced to use Trax for all of my programming assignments for the third and fourth courses of the NLP specialization?

For most of them if TRAX has been already introduced, they will use it in one way or another, thats what I remember when I did the course.

Hi @ABCarter

Nobody is “forcing” you to do anything in this Course.

This Course is about learning Natural Language Processing and the library used (or programming language for that matter (Python vs. Javascript vs other)) is not the most important thing (even though it was a very good choice at the moment (2020) and still is one of the best readable code of all the Deep ML libraries today).

But to answer your question, yes you will be “forced” :slight_smile: to use trax for Course 3 and Course 4.


I will certainly continue with the course because of its excellent instruction in natural language processing. But it is disappointing that one of the things I will not be learning is a framework that will continue to use after the course. I was beginning to do this with TensorFlow when I completed the specialization in Deep Learning and was hoping to continue this with the natural language specialization.

I’m not doubting that there were some good features with Trax and the design goal of combining the speed of TensorFlwo with the ease of PyTorch is exemplary. But for whatever reason, Trax never took off. With only two commits this year and less that two dozen in 2023, it is no longer being actively developed (as of this writing there were more commits to TensorFlow in the past three hours than for Trax this year). I looked at Pluralsight and Udemy for training in Trax, and there is nothing. I google “google Trax” and restrict the search to last year and I get more hits with “Chevy Trax” than with this repository.

And the choice of library for teaching NLP is important. I’ve taken more courses in Coursera, and specifically DeepLearning.ai, than with Pluralsight or Udemy exactly because they are hands-on. At every step I am FORCED to write code. That is a good thing, an essential thing, because you can’t really learn anything in information technology without actually coding. Your courses teach more than theory and concepts, but introduces you to the languages and frameworks you will need to work with if you continue in the field.

This is why the choice of Trax is deeply disappointing.

I agree 100%. When I was taking a course I later completed all the Weeks with PyTorch and also checked most of calculations in Excel to get the deeper understanding of underlying processes and calculations. That is why I say that the framework in question should not be a source of disappointment.
If you complete all the Courses in parallel with TensorFlow or PyTorch (or pick your choice) you will get better at understanding whats going on underneath (compare for example, if the course was TensorFlow only where all the underlying architecture most of the time is hidden from you).

In other words I would still disagree that the trax library is burden, on the contrary, I would lean into numpy only direction rather than TensorFlow only (if that would have been a choice).

Just my take on the matter :slight_smile:

I agree. Since Lukasz left Google, it is not maintained anymore. Would be smarter to update the course and replace Trax with something that is maintained and widely used. For me it is not motivating to use a dead library.

Glad there is someone who agrees. I would be fine with using NumPy, which would give me a deeper understanding of the underlying processes in a package I’m thoroughly familiar with and forms the foundation of both TensorFlow and PyTorch. That would be worth the effort. What I’m not fine with, is implementing a neural network in a framework that I’m never going to see again.

Hi @ABCarter ,
Just in case, FYI the specialization has been updated and the Courses 3 and 4 are now in TensorFlow.