As I explained in the previous post, translating a model from Tensorflow to Pytorch turned out to be a bit more complicated than I initially thought. So I decided to go back to the basics and follow from PyTorch tutorials for begineers.

After completing a few lessons I already feel like I have a better grasp on how PyTorch works and I’m starting to appreciate more its approach. I still think that Tensorflow has a lot of potential for experimentation given that you can create and train a model with three lines of code, though using PyTorch for the last two days forced me to think more about what I’m doing and learn exactly what every layer of my neural network is doing.

Lessons Learned

  • PyTorch has a steeper learning curve compared to other frameworks like keras and Tensorflow
  • There was a lot of logic I previously blindly copy-pasted into my Tensorflow model. Now with PyTorch I had to learn what every piece of the model does