top of page

No Sleep, Just Models: My First ML Project Breakdown

Man of the year: When those in power use compliments to gaslight you while constantly moving the goalpost
Man of the year: When those in power use compliments to gaslight you while constantly moving the goalpost

The Beginning: Choosing an ML Classification Project


I barely survived statistics, and suddenly it was time for the next phase: machine learning. When I opened the course notes two days before class and immediately closed them again, I literally saw stars. For the rest of the week, I was an anxious mess, worried I wouldn't grasp what was being taught and would subsequently fail. My ADHD had also decided to make an appearance, adding another layer of chaos to the mix.


Once I finally grasped the concept of machine learning, though, it became genuinely interesting and opened up opportunities for fascinating projects. This phase was fortunately an individual project; we were tasked with identifying a real-world problem that we wanted to solve. The idea I chose was based on something my best friend and I (and probably many of you) struggle with: content fatigue. My idea was to create a model that could analyze movie summaries and classify films based on the emotions they evoke. I decided to focus on the main five emotions: Happy, Sad, Scared, Anger, and Disgust.


When I made this decision, I thought it would hit that sweet spot; not too hard, not too easy. A chill project, if you will. Boy, was I wrong!


The Challenges: What Made Me Question My Life Choices


I quickly discovered I needed an NLP (Natural Language Processing) library, which I assumed would make the process easier. Dead wrong again. In addition to NLP, I needed libraries and code that could split movie summaries into individual words and count recurring keywords tied to our five target emotions.


I should mention that I discovered this requirement during the model development phase; a classic case of learning as you go. I ended up needing TextBlob, regex (re), and text feature extraction libraries. The hardest part of this entire process was deciphering the endless stream of errors I encountered. Shout out to Google for being that reliable friend who's always there when you need answers.


It's worth noting that I sacrificed significant amounts of sleep for this project, especially while debugging those cryptic error messages that seemed designed to break my spirit.


The Focus: Why Queer Cinema Mattered


Despite the challenges, I think my project turned out quite interesting. I focused primarily on queer movies since it was Pride month, but also because I believe queer films are often misclassified. A movie might be labeled "comedy," but the actual comedy amounts to maybe 10 minutes, while the rest is soul-crushing (yet relatable) trauma, which isn't always what audiences want or expect.


Diversifying storytelling is crucial so that queer narratives aren't reduced to a monolith or seen through just one lens. But I digress, let's break down what I enjoyed versus what nearly drove me to insanity.


The Breakthrough Moments


My favorite part was actually getting the model to work. Initially, I had to write code that would break movie summaries into individual keywords, then classify films based on certain emotional indicators.


The first iteration was a disaster. I thought I didn't need to separate words properly, and while the code technically ran, it was wildly inaccurate and severely underfitted. The model also struggled with emotional nuance; it would confuse "intense" with "happy," creating a beautiful mess of misclassified films.


Shout out to classmates who struggle alongside you; they're especially valuable when they've solved similar problems and can share working code. This collaboration helped me improve my model significantly, transforming it from emotionally confused to somewhat emotionally literate. This was honestly a huge improvement, especially considering I was working with concepts I didn't fully understand yet.


The Final Sprint: ADHD Superpowers Activated


I was thrilled with how everything came together: the code, the presentation, and the write-up. I had activated my ADHD superpower, which helped ensure everything met my standards. I found the perfect PowerPoint template and images to use. Watching all the documents come together was genuinely beautiful, and it became my favorite moment because I could finally rest and fix my destroyed sleep schedule.


This marked the end of four days of sleepless nights and multiple delirious episodes.

The scariest moment was when I thought I had developed a crack in my vision, literally seeing a zigzag divide, like looking through a phone with a cracked screen. It was terrifying, though thankfully it returned to normal after a few hours.


The Results: When Success Feels Like Failure


After submission, we received results faster than usual. I'll preface this by saying my grades weren't bad, some might even call them almost amazing. However, they failed to impress enough to earn full points. As someone with ADHD, failure sticks with us regardless of how minor it might seem. For me, this manifested as hyper fixating on why I fell short of perfection.


Honestly, after learning the reason, I secretly wished I could return to the blissful ignorance of not knowing. The feedback was more infuriating than the actual grade: I was told that because I was one of the best-performing students, they expected more from me. I was being unfairly held to higher expectations.


This genuinely pissed me off. While my friends and therapist validated my anger, they also reminded me not to lose focus. We all agreed that shifting the goalposts was incredibly frustrating. In my very first project, I lost marks for not adding a conclusion to my coding notebook. Now I was being denied points for what felt like arbitrary reasons.

With my support system's help, I managed to regulate my emotions until I discovered that most of what I'd implemented in my project would be part of the next phase's coursework. This sent me spiraling into rage. I had essentially gone above and beyond, but apparently, this still wasn't enough for full marks.


Moving Forward: Channeling Beyoncé Energy


To say I'm less angry now would be a lie; I'm still properly livid. But after being talked down from the ledge by my incredible support system and chosen family, I've decided to heed Queen Bee's wisdom: "Always stay gracious; the best revenge is your paper."


So I'll continue doing my best and let karma handle the rest.


What I Learned: The Unexpected Wins


This might sound nerdy, but I think the greatest thing I learnt was improving my understanding of data visualization graphs. Learning new code to achieve my desired results was simultaneously fun, frustrating, and stressful, but stressful in an oddly enjoyable way.


I'm actually excited to explore more ways to improve the model. I believe it could genuinely benefit the community and, most importantly, help my friends and me avoid movie-induced trauma. I also want to refine the keyword selection to see if it improves accuracy.


This project taught me that sometimes the most valuable learning happens in the messy middle, when you're debugging at 3 AM, questioning your life choices, but ultimately pushing through to create something meaningful. Even when the grading system seems rigged against you, the knowledge and skills you gain are entirely your own.


And next time? I'm managing my sleep schedule better. That vision crack was not it.



Comments


© 2024 by Raheli.M.

bottom of page