From Data to Decisions

Navya Sharma — Thu, 10 Sep 2026 12:20:30 +0000

Optimizing Sales Territory Allocation with Predictive and Prescriptive Analytics

My Lessons from a 7-Model Ablation Study on Multimodal Models

Navya Sharma — Thu, 10 Sep 2026 12:18:25 +0000

Comparing gated approaches for hate speech

Investigating whether a learned per-sample gating mechanism that dynamically weights visual and textual features can outperform static fusion strategies for multimodal classification.

Research Question
When fusing image and text for hate detection in memes, does a model that learns how to combine modalities per-sample outperform fixed fusion? And what does the model learn about when each modality matters?

Dataset
Facebook Hateful Memes (Kiela et al., NeurIPS 2020) -- 10,000 memes with binary labels. Designed with benign confounders: neither modality alone is sufficient.

I Turned Pathfinding Algorithms into an Interactive Game

Navya Sharma — Wed, 15 Jul 2026 16:19:20 +0000

🧩 MazeBench: Benchmarking and Visualizing Pathfinding Algorithms

What is MazeBench ❓

An interactive framework for visualizing and benchmarking BFS, DFS, A*, and Ant Colony Optimization (ACO) on randomly generated mazes.

Why I Built MazeBench 🧐

Initially, my goal was to create an interactive framework where I could visualize each algorithm step by step and compare their search patterns. To make the project more engaging, I used Pygame to give it the look and feel of a real-time maze-solving game.

As I explored a few research and review papers, I noticed that different pathfinding algorithms were often evaluated under varying conditions, such as obstacle density, single-exit, and multiple-exit mazes.

Inspired by this, I decided to build a benchmarking system of my own. I extended the project by adding an automated script that runs BFS, DFS, A*, and ACO across 100 randomly generated mazes and compares their performance using metrics such as runtime, nodes expanded, path length, search efficiency, and success rate.

Features ✨

Interactive Maze Visualization – Watch each algorithm explore the maze in real time.
Four Pathfinding Algorithms– BFS, DFS, A*, and Ant Colony Optimization (ACO).
Step-by-Step Animation – Observe how different algorithms search and make decisions.
Random Maze Generation – Generate a new solvable maze with every run.
Multiple Exit Support – Evaluate algorithms in more realistic maze configurations.
Automated Benchmarking – Run all algorithms across 100 randomly generated mazes.
Interactive Performance Dashboard – Compare runtime, path length, nodes expanded, search efficiency, and success rate.
Exportable Results – Generate benchmark reports and CSV files for further analysis.

Architecture 🏗️

The project follows a modular architecture, separating maze generation, search algorithms, visualization, and benchmarking into independent components. The diagram below illustrates only the interactive visualization framework; the benchmarking pipeline is intentionally omitted for clarity.

Benchmark Setup 📊

Rather than testing the algorithms on a single maze, I wanted a fair comparison across different scenarios. Each algorithm was evaluated on 100 randomly generated mazes under identical conditions.

Metrics collected:

Runtime
Nodes Expanded
Path Length
Search Efficiency
Success Rate

Results 📈

One of the biggest takeaways was that no single algorithm dominated every metric.

DFS achieved the fastest runtime but often produced longer paths.
BFS and A* consistently found optimal paths.
A* explored the fewest nodes while maintaining optimality.
ACO produced competitive solutions but required significantly more computation.

The dashboard below summarizes the complete benchmark across all 100 runs.

🔗 Interactive Benchmark Dashboard: View Live Dashboard

What I Learned 💡

This project reinforced a simple but important lesson:

'The fastest algorithm isn't always the best algorithm'

Visualizing and benchmarking these algorithms made their trade-offs much clearer than simply studying their implementations. Choosing the right algorithm ultimately depends on the problem you're trying to solve.

Thanks For Reading So Far! 😊

💻 GitHub Repository: MazeBench

⭐ If you found this project interesting, feel free to leave a star on GitHub!

Which pathfinding algorithm or feature would you add next to improve MazeBench? I'd love to hear your thoughts in the comments.

DEV Community: Navya Sharma

From Data to Decisions

My Lessons from a 7-Model Ablation Study on Multimodal Models

I Turned Pathfinding Algorithms into an Interactive Game