Recipe Clusters & Ingredients

612 recipes grouped into 4 natural families using k-means on ingredient ratios. Silhouette score: 0.138, a measure from -1 to 1 of how cleanly the groups separate (higher = better-separated clusters).

In plain terms: we let the math sort the recipes by their ratios alone (no labels, no titles) and distinct cookie philosophies fall out anyway (a buttery-rich family, a cakey family, a chocolate-maximalist family). The 2D map below squashes all the ratio dimensions into two axes using PCA (principal component analysis, which keeps the two directions the recipes vary along most) just for plotting, so distance between dots ≈ how differently two cookies are built. Cookies blend into each other, so a modest silhouette score is expected: the families are real tendencies, not hard borders.