Registration and phase

Registration separates when a trajectory traverses a pattern from what spatial pattern is traversed.

If one participant inspects a source at 400 ms and another at 1,400 ms, blind registration can erase the experimental effect.

register_to_landmarks() therefore returns the registered trajectories, originals, warping functions, observed/reference landmarks, and phase summaries.

registered = register_to_landmarks(
    gaze,
    observed_landmarks=inspection_times,
    reference_landmarks=np.array([1.0]),
)

Recommended pattern

1. Fit the unregistered analysis.
2. Register only when scientifically justified.
3. Plot the warpings.
4. Analyze phase or landmark times when timing matters.
5. Report whether conclusions differ before and after registration.

Danger

Registered curves do not preserve latency. They intentionally alter timing to align selected features.

Analyze phase after registration

The estimated warping functions can be converted into a functional phase object with phase_trajectory_set() and summarized using fit_phase_fpca().

Use compare_registered_unregistered_fpca() when you need to document whether registration materially changed the dominant spatial covariance structure.

See Phase FPCA and registration sensitivity.