Sumit Kumar - Projects

Most people's first (and only) encounter with matplotlib is a line chart in a Jupyter notebook. I wanted to see how far it could actually be pushed, so I set out to turn maze generation and pathfinding into a proper animated video, not just a static plot.

The final script renders a high-quality MP4 in two acts: first, Wilson's algorithm carves the maze out of raw grid space, one loop-erased random walk at a time, visualized as a swarm of orange particles eating through walls. Then Dijkstra's algorithm floods outward from the start in cyan, searching for the exit, before the final path lights up in a single bright red line.

Making a plotting library do things it wasn't built for

The whole animation is generated frame-by-frame - hundreds of individual matplotlib frames stitched into video. That alone creates a performance problem: naively, every particle you've ever drawn stays on screen and in memory forever, and the render grinds to a halt. The fix was giving each particle a limited lifetime, so it fades out and gets garbage-collected after a short window instead of accumulating indefinitely.

To make the visualization actually feel alive rather than just correct, I built a small particle system on top with trail effects and glowing pulses, and used Python's colorsys module to generate smooth color gradients that track the algorithm's progress - e.g., particles shift color based on distance explored from the start point, so you can see the search frontier expanding in real time.

What I actually learned

• matplotlib is a far more capable rendering engine than its reputation as "the charting library" suggests.
• Good visual effects are rarely one big idea - they're a lot of small parameters (particle lifetime, fade curves, color gradients) tuned until it feels right.
• Once you're generating hundreds of video frames, performance planning stops being optional and becomes the main design constraint.