SEM error due to numpy update to version 2.4 in late 2025

[qszhao]

Hi spreg development team,

In my recent spatial regression teaching session, my students encountered an error when running the SEM model over Airbnb datasets.

Error: TypeError: only 0-dimensional arrays can be converted to Python scalars.

After checking, I believe this relates to the recent numpy version update to 2.4.0 near the end of last year.

apple/coremltools#2633

https://github.com/numpy/numpy/releases/tag/v2.4.0

My suggestion to them now is to downgrade to numpy 2.3 or below, and it works fine.

It will be nice to adjust the code to accommodate this change in numpy, or specify the version in the environment yml file to avoid future issues.

I can provide examples/environment set up for you to replicate the issues.

Best,
Qunshan

[martinfleis]

Hey, could you provide reproducible example? The CI running with numpy 2.4.2 does not hit this so it might not be covered by current tests.

[qszhao]

Yes I can provide one for you. What format will be useful and would a notebook work?

[martinfleis]

Does not really matter if it is a code snippet in the body of the GitHub message or an attached notebook. Something that would allow us to run the code and hit the error :)

[knaaptime]

Hey, could you provide reproducible example? The CI running with numpy 2.4.2 does not hit this so it might not be covered by current tests.

it looks like the unittests last ran in january though so we need to get those turned back on

[martinfleis]

@knaaptime Yes though I did trigger a manual run earlier today to check the status.

[pedrovma]

Hi @qszhao ,

A reproducible example will be important here. I tried the code below, and it worked fine with Python v3.14.3, numpy v2.4.2, and spreg v1.8.5.

Best,

Pedro.

import libpysal
import spreg
import geopandas as gpd

libpysal.examples.load_example("AirBnB")

db = gpd.read_file(libpysal.examples.get_path("airbnb_Chicago 2015.shp"))

y_name = ["price_pp"]
x_names = ["income_pc", "crowded"]
db_clean = db.dropna(subset=y_name + x_names).copy()
y = db_clean[y_name]
x = db_clean[x_names]

w = libpysal.weights.Queen.from_dataframe(db_clean, use_index=True)
w_name = "airbnb_q"
w.transform = "r"

SEM_reg = spreg.GMM_Error(y, x, w=w)
print(SEM_reg.summary)

[qszhao]

Thanks Pedro - I tried the example shared from my student, and it works fine as well for me on my Mac. So I asked him to send me the notebook with the error and see if I can replicate that in Windows. I will update shortly.

[martinfleis]

It seems to be an environment question. In libpysal's reverse dependency testing, this error shows up in spreg's tests - https://github.com/pysal/libpysal/actions/runs/22981738753/job/66722885949

[qszhao]

SEMExample.zip

Hi all, see attached examples - my student also generated a html file to show the example. I think he set up the environment based on the conda-forge channel following my guidance here, so that's less likely the package compatibility reasons: https://github.com/qszhao/ATUA2026/tree/main/Spatial_Regression.

I have not tested in Windows, and I will probably leave it to you to check. @martinfleis @pedrovma

[pedrovma]

@martinfleis , it seems the failures started after pysal/libpysal#837 was merged, right?

[martinfleis]

That has nothing to do with it. The difference is that it is passing on March 2 with numpy 2.3.5 and failing on March 4 with numpy 2.4.2 enabled in the CI by release of numba 0.64. However, the same combination of deps is in CI here (https://github.com/pysal/spreg/actions/runs/22955117161/job/66629647189) where everything is green, so I am not fully sure how should the environment look like to trigger the issue. It is certainly not a Windows issue as libpysal CI is running Ubuntu.