Lecture 5 with PyTorch only

Recently I’ve completed Lecture 5 on embeddings, and I really wanted to try use PyTorch directly, without fast.ai wrappers. I’ve written a small post about my attempts. Probably someone will find the posted article interesting when carrying on their own experiments with PyTorch.



can you comment your code fully as that will help to better understand the code?

Hi @vibhorsood! Not a problem. Probably I was a bit too concise in my commentaries =) I’ll add more information then to the notebook.

1 Like

thanks…please let me know once you have finish commenting especially the Cyclical Learning Rate (CLR) part…greatwork!


Thank you for the appreciation! I’ve updated the notebook, probably now it is more explicit. I’ve updated sections about network construction, learning rates and training loop.