Before Python and C, I learned to code on a Casio graphing calculator. This is a nostalgic look at what survived fifteen years in its memory.
If you’re anything like me, you’ve trained, tuned, and tweaked tree models just about every possible way to squeeze out that last half-percent of accuracy. But what happens when we leave the forest behind? This time, we’re venturing into the world of neural networks by building a simple PyTorch MLP, with a few bells and whistles added for good measure.
How do you do justice to two weeks of work in half an hour? Visual design, pacing, last-minute edits, rehearsals, handling criticism... Here's a detailed behind-the-scenes look at how I built and gave a presentation to defend a final project report.
Overfitting is everywhere in machine learning, easy to define, but harder to conceptualize. So we cook up tiny datasets, throw a lot of models at them, animate, and toss in an interactive plot for good measure. All in the name of a single goal: get a feel for what a well fitted model looks like.
What do you get when you mix a century-old shipwreck, a Kaggle leaderboard, and a stubborn data analyst? A machine learning journey that begins with raw CSVs and climbs to 0.82 accuracy, passing through group survival logic, spatial deck patterns, careful hyperparameter tuning, and a detour into SHAP values and model interpretation. And at the end of it all: a quiet tribute to the 2,208 passengers who did not ask to become training data.
I assumed Python's range was a function, until a friend's correction launched me down a rabbit hole. Join me as we dissect iterables, low level C implementations, and how being wrong about range taught me more than being right ever could. Spoiler: It ends lazy evaluation, and newfound awe for Python's greatest magic trick : making hard things appear simple.
When a "simple" binary fought back. This seven years old CTF writeup, preserved for posterity, captures the struggle of a low-level beginner: humbled by Linux, lost in assembly, saved by decompilers, and almost outwitted by self-deleting code. And beyond the flag, a timeless lesson about knowledge.