Mismatching behavior between segment and section taper rates

[eleftherioszisis]

Segment taper rates calculate the absolute slope of the segment diameters:

NeuroM/neurom/features/neurite.py

Lines 230 to 243 in a9e3020

	def segment_taper_rates(neurite):
	"""Diameters taper rates of the segments.
	The taper rate is defined as the absolute radii differences divided by length of the section
	"""
	def _seg_taper_rates(sec):
	"""Vectorized taper rates."""
	pts = sec.points[:, COLS.XYZR]
	diff = np.diff(pts, axis=0)
	distance = np.linalg.norm(diff[:, COLS.XYZ], axis=1)
	return np.divide(2 * np.abs(diff[:, COLS.R]), distance)
	return _map_segments(_seg_taper_rates, neurite)

np.abs(diff[:, COLS.R])

whereas the section taper rates calculate the signed slope from the linear fit to the diameters:

NeuroM/neurom/features/neurite.py

Lines 247 to 259 in a9e3020

	def section_taper_rates(neurite):
	"""Diameter taper rates of the sections from root to tip.
	Taper rate is defined here as the linear fit along a section.
	It is expected to be negative for morphologies.
	"""
	def _sec_taper_rate(sec):
	"""Taper rate from fit along a section."""
	path_distances = np.cumsum(interval_lengths(sec.points, prepend_zero=True))
	return np.polynomial.polynomial.polyfit(path_distances, 2 * sec.points[:, COLS.R], 1)[1]
	return _map_sections(_sec_taper_rate, neurite)

Is this to be expected? I would expect that in both cases the signed slope would be desired.

from io import StringIO

import numpy as np
from numpy import testing as npt

from neurom.features import neurite
from neurom import load_morphology

neu = load_morphology(StringIO(u"""((CellBody) (0 0 0 2))
                                    ((Dendrite)
                                     (0 0 0 2)
                                     (1 0 0 1)
                                     (2 0 0 0))"""), reader='asc')


npt.assert_almost_equal(neurite.section_taper_rates(neu.neurites[0]), [-1])
npt.assert_almost_equal(neurite.segment_taper_rates(neu.neurites[0]), [-1., -1])

The second test fails with [1, 1]

@lidakanari @arnaudon @adrien-berchet

[arnaudon]

You're right, abs should not be. I think I added section taper rate for diameter synthesis a while ago, probably I didn't pay attention to the segment one, which is very noise in general