CS 224D course was really wonderful. I wish I had access to all the video lectures and also Q/A during office hours. But, whatever was made public was also huge. A big thanks to Stanford (and CS 224D instructor and TAs) to make this public.
I was finally able to complete the assignments for CS 224D. (Except assignment 3 2nd part for which I was not able to find the zip file.) Pushing the initial commit for now (may be in future would refine it).
After doing this course, I got a basic understanding of DNN in NLP. To make my understadning a bit concrete I need to impliment few of thee DNN structures being talked in the course. Would do that separately, while mastering TensorFlow.
TensorFlow has gone through quite some changes from the time when the course was created. However, enough material is present on net to incorporate the changes needed.
- The forward and backprop equations have been derived at: https://aknirala.github.io/CS224D/FowdBkws.html
- Assignment1 folder contains a pdf where for problem 4, cost function and it's derivation, as needed, has been calculated. It contains the images.
- I have not searched for optimal parameters, nor I have used dropout everywhere (I just needed to get to know what DNN in NLP were all about).
- A lot of code for assignment 2 is inspired/taken from: https://github.com/vijayvee/CS224d_Assignment_2_Solutions