Numpy dependency >=2.0

[ethanhung11]

Hi there, this is a great tool. I was wondering if there were any plans to update the pyproject.toml to allow numpy versions >=2.0? I have a few packages within the same environment that require numpy>=2.0.0, which conflict with pybanksy when solving for a compatible environment.

[cafferychen777]

I'm also encountering this dependency conflict. I've done some testing that might be helpful for the maintainers:

Testing Results

I tested pybanksy 1.3.4 with NumPy 2.2.6 on Python 3.12 (macOS ARM64), and it appears to work correctly:

NumPy version: 2.2.6
BANKSY imports: OK

# Full test with spatial transcriptomics data (1932 spots × 33,538 genes)
- initialize_banksy(): OK
- generate_banksy_matrix(): OK  
- Leiden clustering on BANKSY features: OK
- Result: Successfully identified 5 spatial domains

The only warning I observed was unrelated to NumPy:

RuntimeWarning: invalid value encountered in divide (in banksy/main.py:302)

Suggested Fix

Based on my testing, it seems the numpy<2.0 constraint may be overly conservative. A simple update to pyproject.toml might resolve this:

# Current
"numpy<2.0,>=1.21"

# Suggested
"numpy>=1.21"

I'd be happy to submit a PR with this change if the maintainers confirm compatibility. Are there any specific NumPy 2.0 breaking changes that the team is aware of that might affect BANKSY?

Thanks for the great tool!

[zhou-1314]

Hi, I've submitted PR #38 to address this.

The fix is a one-line change in pyproject.toml: removing the <2.0 upper bound (numpy>=1.21,<2.0 → numpy>=1.21).

Before opening the PR I did a full audit of all source files in src/banksy/ and src/banksy_utils/ — no deprecated or removed numpy APIs are used anywhere (no np.bool/np.int/np.float aliases, no removed functions like np.in1d, etc.), so the constraint was purely precautionary.

I also ran an end-to-end test with the bundled STARmap dataset under numpy 2.3.5 / scipy 1.16.3: data loading → spatial graph construction → PCA → Leiden clustering → cluster refinement all pass cleanly (ARI = 0.714, refined ARI = 0.723).

Hope this helps!

[vipulsinghal02]

Thank you for the PR, and hang in there! I will test it this weekend, and if all looks good, merge.