Repository navigation
get_sample_pts_euclidean: interpolation weights appear swapped #22
Description
Summary
When get_sample_pts_euclidean samples points along an SWC edge, the interpolation appears to place each point at the mirrored location on that edge. In a simple 3-node chain, this changes the actual distances from the root returned by the sampler, not just their order.
I’m including:
- sampling_with_demo_swc.py (repro using CAJAL
read_swc+get_sample_pts_euclidean) - sampling_potential_fix.py (standalone repro with the same arithmetic)
Minimal reproducible example
- Save the following as
demo.swc. The example neuron has 3 nodes on the x-axis atx = [0, 3, 10].
# id type x y z radius parent
1 1 0 0 0 1 -1
2 3 3 0 0 1 1
3 3 10 0 0 1 2
- Run:
python sampling_with_demo_swc.py- Observe output:
expected: [ 0. 2. 4. 6. 8. 10.]
got : [0. 1. 3. 5. 7. 9.]
Potential fix
In get_sample_pts_euclidean inside src/cajal/sample_swc.py, interpolation currently uses:
(root_triple * x) + (child_triple * (1 - x))
If sampled points are intended to be measured from the parent/root along the edge, the weights seem swapped. A possible change would be:
(root_triple * (1 - x)) + (child_triple * x)
sampling_potential_fix.pydemonstrates the output before and after swapping these weights, again using the 3-node example above.
Activity
pcamara commentedon Mar 19, 2026
Hi @tatsuookubo ,
Thanks so much for catching this. You are right that the interpolation was inverted and I agree with your fix. It should not affect morphological distances substantially, since GW does not require uniform sampling, but it is more accurate.
Please feel free to open a pull request, and I will merge it. Alternatively, if you prefer that we implement the fix on our end, just let me know.
tatsuookubo commentedon Mar 23, 2026
Thank you very much for taking a look!
In our hands, this had very little effect on the resulting intra-cellular distance matrices, but I just wanted to bring it up in case it helps make the implementation consistent with the documented behavior.
I've opened a pull request with a proposed fix: #23.
Overall, this CAJAL package is very inspiring and well implemented, and we have learned a great deal from it. Thank you to your group for this nice contribution to the field!