Portfolio
Back to portfolio

TSP Visualizer

The traveling salesman problem asks for the shortest closed tour through every city. It is NP-hard, so past a handful of cities nobody checks every tour. Instead you pick a strategy for searching well. Below are four strategies from two of my papers, running live in your browser. Pick a method, press Run, and watch how it searches.

Method

Objective

Speed

Total length

588.0

Longest edge

88.4

Generations

0

Best so far

588.0

GA + Wisdom of Crowds: Population evolution with crowd consensus

The CSE 545 hybrid. Alongside ordinary crossover, each generation counts how often every edge appears among the top-performing tours and builds extra offspring that greedily follow the highest-agreement edges, using nearest neighbor to complete the tour when the crowd offers nothing. Offspring inherit from the whole population, not just two parents.

What to look for

  • Nearest Neighbor finishes instantly but leaves one long edge stranded at the end, because it never reconsiders an early choice.
  • Simulated Annealing looks chaotic at first. That is deliberate: a high temperature accepts worse tours so it can cross valleys, and the tour only settles as it cools.
  • The plain Genetic Algorithm improves in visible jumps as a better tour takes over the population, then stalls once diversity is gone.
  • GA + Wisdom of Crowds usually locks in the obvious edges sooner, because offspring inherit whatever the whole leading group agrees on rather than just two parents.
  • Switch the objective to the longest edge and the winning tour changes shape, happily adding total distance to shorten its single worst hop.

This is a teaching model, not the code from the papers. It runs entirely client-side at a smaller scale, so the tuning is lighter than the experiments the papers report.