Repository navigation
make_heap code object inequality at import time #250
Description
Activity
I have never seen this before. I'll have to see if I can reproduce anything like it locally.
Reacted by Jesse MeyerA member of our group reports that a very similar setup with Python 3.10 works as expected (both on Native Windows and through WSL).
I have the same issue on my side with the following specifications:
- Linux
- Python 3.12
- scikit-learn=1.6.0
- scipy=1.15.0
- umap-learn=0.5.7
- pynndescent=0.5.13
- numpy=2.0.2
- numba=0.60.0
I installed everything via
mamba install -c conda-forge umap-learn.Maybe this helps in tracking down the issue.
I have this problem when I run a file using debugger (VS Code). Is there any recommended debugger?
Yeah, without being able to use a debugger, I can’t maintain anything …
This seems to be the asssertion that gets triggered: https://github.com/microsoft/debugpy/blob/39879bdeeae5908a8f3527c77ba252e0930afabc/src/debugpy/_vendored/pydevd/_pydevd_sys_monitoring/_pydevd_sys_monitoring_cython.pyx#L562
Workaround: microsoft/debugpy#1719
Current workaround is to downgrade either numba to 0.59.1 or python debugger to v2024.10.0
(I think using Python 3.11 also works)
Upstream issue: numba/numba#9760
In attempting to go through the introductory material for UMAP, I hit the following exception with
import umap:<code object <module> at ..., file "/path/to/keras/lib/python3.12/path/to/pynndescent/sparse_nndescent.py", line 1> != <code object make_heap at ..., file "/path/to/keras/lib/python3.12/path/to/pynndescent/utils.py", line 171>.I'm at a loss on how to debug this. Note that the venv here is called "keras".
WSL2
Python 3.12
scikit-learn 1.6
umap-learn 0.5.7
scipy 1.14.1
pynndescent 0.5.13
numpy 2.0.2
numba 0.60.0
keras 3.6.0