Trax is not bad though nowhere as nice as keras or pytorch specially for those with a decent programming background. Even the basic CPU version is a nightmare to install/upgrade. Combine that with lack of documentation and community support (which takes a while), this is a really bad choice for an online course. It is not sufficient for a library to be new or cool, there are many other things to consider while choosing a library. Coursera has clearly missed the mark here and made the learning experience unnecessarily bad.
Thank you for your feedback, @Abhijat.Vatsyayan. It is appreciated. I shall take this back to the team.
I completely agree. For me Trax is nowhere as user-friendly or intuitive as the other two (Keras, PyTorch) and the reduction of boilerplate from Pytorch gets replaced with succinct but cryptic syntax which is neither well-documented in the official documentation nor explained as part of these videos. Overall I find it inappropriate to push an experimental and relatively new tool such as Trax part of a MOOC.
I don’t think you should pin this on Coursera; deeplearning.ai owns this decision and my assumption/explanation is as simple as ‘The Google force is strong with this one.’
I’m more annoyed by the fact that trax isn’t widely supported by hardware/OS platforms than the syntax or usage of the library itself. Means I can’t run any of the exercises from the class at home without rewriting everything. PITA.
1 Like
Agreed, I did not take into account how Coursera works. Deeplearning.ai has completely missed the mark here.
I totally agree and found Trax to be nowhere as user-friendly as Keras. TBH, I had never even heard of it before. Personally, I prefer the functional API in the latter, which allows for a great deal of customization, especially in constructing complex model architectures.
Please consider this @Mubsi.
Agreed. Trax is no longer an active project. There were less than two dozen commits in 2022 and as of July 1, only two commits for all of 2023. I just finished the Deep Learning specialization and was beginning to get up to speed with Tensorflow. I was hoping to continue learning with this second specialization. I’m deeply frustrated to be forced to use a framework that I’m probably never going to see again. This is the first time I’ve been truly disappointed with deeplearning.ai.