Dimensionality Reduction Calculator

This calculator takes up to five eigenvalues, sorts them from largest to smallest regardless of which numbered slot you typed them into (line 12), and walks the running cumulative share of total variance until it clears your Target Variance Explained threshold (lines 26-31) — that crossing point is Components to Keep. The Total Variance field is not part of that arithmetic at all: the variance total the percentages are measured against is recomputed from the five eigenvalues themselves (line 15's `eigenSum`), so changing Total Variance alone moves nothing. A negative eigenvalue is not clamped to zero either — the collection loop only keeps entries where `v >= 0` (line 8), so a negative value is dropped from the count entirely rather than contributing a zero-variance component.

At the defaults (45, 25, 15, 10, 5), the first four eigenvalues already cover 95% of the total, so Components to Keep is 4 and Dimension Reduction reports 20% (1 of 5 dropped); the Kaiser criterion separately flags 2 components (those above the mean eigenvalue of 20). Raising Eigenvalue 5 — the smallest — actually raises Components to Keep, because it inflates the variance total every share is normalized against (line 15) enough to push the first four components' cumulative share just under the 95% line, forcing a fifth component in. What this does not account for: negative-eigenvalue inputs contributing anything to the total, or the Kaiser criterion changing which components it flags once a perturbation is large enough to cross the running average — this calculator's Target Variance field, unlike some others on the site, uses `??` rather than `||` (line 4), so a literal 0 there is respected and then clamped into the 1-100 range rather than silently replaced.