pokemon-color-search/plan.md

64 Values

Every pixel set becomes 7 values (size, inertia, mu (3 components), nu (3 components)) Every image has 4 pixel sets (total + 3 clusters) All this is done in both spaces Minus the duplicated size of the total set Plus name

2 * 4 * 8 - 1 + 1 = 64 values to each image

Cleanup

• Properly refer to inertia and mu in ingest instead of XYZ
• Same fixes in nearest.js, make it easier to actually find the values needed

New values and logic

• Add cluster inertias
• Add nu metric/pointing vector to ingest
• Add cluster nus
• Record (non-outline) image size and raw cluster sizes instead of just percents
• Reorganize output, modify nearest.js reader to match
• Change logic of front-end
  • Now the process is, select cluster > metric > scale > sort
    • Clusters: Total, Cluster
      • Cluster choice is according to a separately-specifiable metric + scale + sort, compared per pokemon
    • Metrics:
      • RMS: sqrt(I - 2q . mu + q . q)
      • Mean of Angle: nu . q_hat
      • Angle of Mean: mu_hat . q_hat
      • Hue of Mean: mu_perp . q_perp
      • Euclidean to Mean: sqrt((mu - q) . (mu - q))
      • Chebyshev to Mean: max(abs(c) for c in (mu - q))
      • Inertia: I
      • Size: N
    • Scales: None (1), Proportion (N_k / N), Inverse Proportion (N / N_k)
      • hidden when not using clusters
    • Sort: max vs min
  • Now each tile has
    • Name
    • For each portion (whole + 3 clusters)
      • Mu (hex + vec, also sets tile color)
      • Inertia (I)
      • Size (N) (render as proportion for clusters)
      • Nu
      • RMS (sigma)
      • Whole angle (Theta)
      • Mean angle (theta)
      • Hue angle (phi)
      • E dist of mean (delta)
      • C dist of mean (C)
    • 4 * 10 = 40 results
    • Probably hide the metric results and focus on just mu and maybe sigma, but provide dropdowns

DB schema

          jab                                                    rgb
          total     cluster 1      cluster 2      cluster 3      total     cluster 1      cluster 2      cluster 3
name size I *mu *nu size I *mu *nu size I *mu *nu size I *mu *nu I *mu *nu size I *mu *nu size I *mu *nu size I *mu *nu