Web App: ShiftLens
Published:
Application: ShiftLens
Title: Interactive Visual Explorer for Domain Adaptation
Language: JavaScript / static HTML
License: MIT
Overview
ShiftLens is a lightweight teaching and inspection app for domain adaptation. It uses deterministic two-dimensional toy datasets and step-by-step canvas animations to show how adaptation methods transform a source distribution toward a target distribution.
Toolkit Role
ShiftLens is the intuition layer of the toolkit. It does not run real EEG benchmarks. Instead, it explains the geometry behind the domain-adaptation methods used by DA4BCI, CrossDA, and ShiftDx.
DA method idea -> ShiftLens animation -> visual intuition before real EEG experiments
Main Capabilities
- Method animations for SA, CORAL, RD, ART, PT, TCA, MIDA, GFK, OT, and M3D.
- Source-target color coding and class-shape coding.
- Compare mode for paired method playback.
- Live metrics for means, covariance, MMD, and class-centroid gap.
- Presets for covariate shift, conditional shift, label shift, nonlinear warp, and outlier stress tests.
- Custom CSV input, URL-shareable state, PNG export, and WebM recording.