Train Shake

Machine learning

Turn Audio on. Hold each teach button while you perform the gesture (3+ seconds). Record still, sway, and shake (3+ each), then Listen.

Compare k-NN vs a tiny 2-layer net — all on your phone, no server.

Feature map — each dot is one example (mag vs jitter). White ring = live reading while listening.

Confusion matrix — rows = true label, columns = predicted; off-diagonal = mislabels on your training set.

Ethics — data stays on device until you export JSON. Different grips and phone models change readings; discuss bias before trusting the sound. Export only when the class agrees.

Still: 0 · Sway: 0 · Shake: 0

Teach · feature map
still sway shake
Confusion matrix (leave-one-out)
Ethics · on-device