Gradient Descent Dungeon

Playground

Mix algorithms and dataset shapes to observe model assumptions, decision surfaces, clustering behavior, and generalization.

This is not a pass/fail level. It is a sandbox for comparing what each algorithm can represent, where it struggles, and how dataset geometry changes the result.

Back to lessons

Playground Setup

Choose an algorithm and dataset

Current: K-Means

K-Means Arena

Centroid assignment map

K-Means alternates between assigning each point to the nearest centroid and moving centroids toward the mean position of their assigned points.

Current dataset: Clean BlobsTrain K-Means to reveal centroid assignments.

K

Inertia

Shift

Dataset Variant

Choose the data condition

Clean Blobs

main

Three compact groups. K-Means should work well with K = 3.

samples: 180noise: 0seed: 931natural groups: 3

Training Console

Tune the model

Choose K, run assignment/update steps, and watch centroids move toward compact clusters.

K-Means idea

K-Means alternates between two simple steps: assign each point to the nearest centroid, then move each centroid to the average position of its assigned points.

Clustering Report

Train K-Means

Choose K, set the number of iterations, and train the model to inspect inertia, silhouette score, centroid movement, and cluster sizes.