The end of summer (June 30th maybe..)
There are many possible worst programming memories you might have, for some those are learning a new coding language, debugging a project at night, or Leet code. For most of you this might be the worst moments from your memories, aimlessly starring at a screen with some weird syntax and tags wondering why somethings work and some don't, well, you might've not experienced this but here's what happened:
June 20th maybe..
Logging onto devpost I was just scrolling through the ongoing hackathons, that's when I saw a niche little topic to work on for me at the very least, Hack4Health: A hackathon in which the participants had to build a prototype for use in healthcare (oversimplified). So, I had just the idea to build, was working on improving my python skills and also had the perfect cause behind the work as well. Now for those who might be curious, I was planning on building a webpage to identify types of strabismus into 5 categories
- Normal, 2. Hypertropia, 3. Hypotropia, 4. Esotropia, 5. Exotropia And for each of the category when identified to pull up appropriate eye movements through interactive activities which pushed a person to move their eyes in complex manners and perform all the eye movements throughout. I got a teammate onboard and began writing the code.
By the end of the competition, we had a working prototype to present but a few casualties were eminent along the way. one of our greatest achievements while writing the program was to allow for image as well as video inputs as per the user choice and would also support an inbuilt camera.
Everything worked just as I had imagined and well managed thanks to my teammate who was plenty help. But like I said there were casualties, one of the first and largely the most important part of any ML or DL based classifier is that it should do its job with high accuracy, something missing in our model. The model had achieved nearly 86-87% accuracy on a dataset via dropouts, early stopping, augmentations, lr-reductions etc. but at some point, during a larger 'panic' had been replaced by a less optimal 70-72% model without any backups. Take it as you might, this might have just been the inexperience when working on a larger project or late-night coding, I can't say for sure.
Now, i'll skip over some trivial issues like frontend-backend communications, learning keras and tensorflow, using dlib, which leads us to the topic of this post: Stack overflow is the greatest forum site for programmers from my experience (given it's not that much) is truly helpful when one faces a rare/uncommon issue which even our AI friends can't provide solutions to.
Now the larger panic I mentioned early on was that while importing all the necessary libraries for our project I had not thought to check compatibilities..(rookie mistake, I know) and wouldn't you know it 60% the way their and somethings just stop working, the frontend doesn't respond to button calls, the animations don't work with the correct category of output, no input is taken in, huh.
This sole issue plagued my code and my mind for 32 hrs working uptill 3am trying to solve this plague, AI just told me to change my code which was never the plan, I wanted to build on my pythonic skills i.e. learn. The headache was unbearable and the deadline nearly there so I couldn't scrape my work now! That's when it appeared, some user on stack had mentioned this issue persisting for NumPy 2+ but was solid for 1.23.xx versions, and it was the greatest assist I could've ever gotten. The problem occurred due to how NumPy handled array datapoints which was the key input for dlib.
Now given the state of our project I was proud for our achievement, yes proud yes, confident no. So, it wasn't a big surprise when we saw we didn't win but hey the more you know, I guess.
So, this was my first, yes first ever hackathon experience, if you want to check out the page: https://github.com/Unknownerror-404/Hack4Health/
(the code maybe broken as I haven't updated it in a while)
Top comments (0)