Clustering Visualizer

Revamp of a project I made during my college days! The one I created during my college days was an Express app, not deployed anywhere and was a total spaghetti code in my opinion :p.

Run locally

Install dependencies: npm install

Run development server: npm run dev

Algorithms:

K Means

1. Maintain current-clusters and previous-clusters
2. (Cluster Assigment) For each data point
   • Find closest centroid using Euclidean distance
   • Update in current-cluster
   • Redraw
   • Increment data index
3. Reset parameters after assigning each data to a centroid/cluster
4. (Update Centroid Location) For each centroid
   • Find center coordinates of the centroid's cluster
     • Move centroid towards the center of cluster in small increments
     • Redraw
   • Increment centroid index after centroid has reached the center
5. Compare previous-clusters against the current-clusters
   • If the current-clusters and previous-clusters are same, stop the algorithm
   • Else
     • Set previous-clusters to current-clusters
     • Reset current-clusters for next iteration
     • Reset parameters after updating each centroid's location
     • Repeat from step 2

K Medoids

Note: The final medoids may not be the optimal medoids (As every combination of data may not get evaluated)

1. Set Total Cost(TC) to Infinity
2. Maintain distance matrix using Manhattan distance
3. (Cost Evaluation) Find cost of current combination of medoids
   • Set medoidsCost to 0
   • For each data point
     • Find closest medoid using distance matrix
     • Add that distance to medoidsCost
     • Update current-clusters
4. Check medoidsCost against TC
   • If medoidsCost >= TC, stop the algorithm
   • Else
     • Set TC equal to medoidsCost
     • Redraw
     • From each cluster, find a point whose average distance from all other points in the cluster and its medoid is the least. Use it as the cluster's new medoid
     • Repeat from step 3

DBSCAN

Note: Stars are cores, all data points whose center lie on or inside a neighbour circle are boundary points, rest are noise

1. Set the epsilon (neighbourhood radius) and the minimum points (minPoints)
2. For each data point that is not a core
   • Find its neighbours (N)
     • Maintain a neighbours array
     • Find Euclidean distance between current data point and every other data point
     • If distance <= epsilon, add that point as neighbour
   • If |N| + 1 >= minPoints
     • Set that data point as core
     • Create cluster
       • Find cores in the neighbour array
       • If a core is found, add its neighbours (which are not cores) to the neighbour array
       • Repeat and iterate till end of neighbour array
3. Redraw