Anjishnu I'm on Twitter :)
CS PhD student. Working on Fairness and Interpretability methods in NLP.

Week Nine

The beginning of the end.

Well, not in a dark way, of course. But yeah, this was the first week of the last month of the coding phase for GSoC. So, in that sense, yeah 😅

The week started with a reminder from Google that I needed to pack up things nicely and be able to present it by the end of this month, or as they term it, Work Product Submission. Now, in my case, I had been maintaining a personal repository with demonstrations, notebooks and all other code files that I have used throughout GSoC to produce the features and blogs. However, the repository was not very well-maintained, in the sense that it lacked clear READMEs for the different sections. So, my first task this week was to restructure the repository properly, so that I can present it directly as The “Work Product”. The image you see above is just a little something I added to the homepage README to make the repository feel a bit more personal (and cool)! Anyway, now that that is done, any feedback on the repository would be really welcome.

Apart from that, I opened PRs for the 3 features I had worked on during the second phase, namely Instance Norm, Pixel Shuffle and Spatial Dropout. I ran into some unexpected errors and issues while doing this, which took up a lot of time.The Gradient Test for Instance Norm fails. More precisely, the test crashes after throwing some memory error, which I am unable to understand. Apart from that, the Doxygen builds for Pixel Shuffle and Spatial Dropout fail, saying that there is no documentation for the Backward() method of those layers, when there clearly is! Maybe I am reading the build error message incorrectly. Would need to look at it more closely/ ask mentors about this. Anyway, these are the areas I need a little bit of help with. The rest of the proposed work for Phase 1 and Phase 2 is complete, at least in concept. Obviously, with all the major changes going on, with the Catch2 refactoring and all, I am having constant merge conflicts in all the branches, so those would need to be taken care of before any of these can be merged.

@birm merged one of my inital PRs on Soft Margin Loss a couple of weeks back. That was nice! I have been meaning to mention this in the blog since then, but I just keep forgetting.

In Week 10, I am mainly trying to understand the code that is present in the mlpack-TensorFlow-Translator, so that I can make the changes I proposed for it correctly, hopefully by next week.

Coming to fun stuff done this week, I learnt how to make pudding properly this week. I know it is probably very easy for someone who knows how to cook. But for me, this was big, mainly because the only other thing I know is how to boil eggs. So, yeah, it was a pretty steep learning curve, an enjoyable one nonetheless.

I watched the movie Mortal which was released this year. Its a pretty interesting take on Thor, the Norse God of Thunder. The scenery shots in it absolutely will remind you of the fact that you are stuck in home in the middle of an unending lockdown, and that beautiful places still exist outside of those 4 walls.