Should landmark registration throw error if landmarks is an empty list?

[hovinh]

Hi team,

I have used landmark registration for my trace alignment and noticed a strange behaviour. If I accidentally assign the argument landmarks = [[]] in registration.landmark_registration_warping, it still processed normally. The trace I provided is an anomaly trace, hence the defined landmarks were unable to extract. Based on the output trace and warping plot (a straight diagonal line), I believe it simply do no warping at all.

Do you think this is intended? Personally, I would like to have a checking if landmarks have the same number of elements with location argument and should contain at least 1 element, or a warning for users. And could you please share how this alignment method behaves when no landmarks provided, would it take the first and last element as landmarks?

Thank you.

Sincerely,
Vinh

[vnmabus]

Sorry but in the develop version there is a check for that (and the function is called landmark_elastic_registration_warping). Also you didn't provide any trace.

May you check if that works in develop, and show an example that fails if that does not work?

[hovinh]

Hi, I checked the develop branch:

• It's true that if landmarks is empty and location is non-empty, there is error raised.

Error: ValueError: Number of landmark locations should be equal than the number of landmarks (2) != (0)

• If both landmarks and location are empty, then no error thrown. This is actually my case. Put it simply, location are determined as median from landmarks, hence if one is empty, the other follows suite. I designed it that way to do something similar to a groupwise landmark registration.
  Example:

import matplotlib.pyplot as plt
import numpy as np

import skfda
fd = skfda.datasets.make_multimodal_samples(
    n_samples=4,
    n_modes=2,
    std=0.002,
    mode_std=0.005,
    random_state=1,
)
fd.plot()
plt.show()

landmarks=[[], [], [], []]
warping = skfda.preprocessing.registration.landmark_elastic_registration_warping(
    fd,
    landmarks,
    location=[],
)

# Plots warping
fig = warping.plot()