Error in example script surface_timeseries_.py

GenevieveBuckley commented 1 year ago

🐛 Bug

The napari example script surface_timeseries_.py returns an error when it is run.

To Reproduce

Steps to reproduce the behavior:

Install napari, and nilearn into your python environment
Download and run the napari example surface_timeseries_.py from here: python surface_timeseries_.py
An error message is produced: "ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (2,) + inhomogeneous part"

Error message:

``` python surface_timeseries_.py Dataset created in /Users/genevieb/nilearn_data/nki_enhanced_surface Downloading data from https://www.nitrc.org/frs/download.php/8470/pheno_nki_nilearn.csv ... ...done. (1 seconds, 0 min) Downloading data from https://www.nitrc.org/frs/download.php/8261/A00028185_rh_preprocessed_fsaverage5_fwhm6.gii ... Downloaded 42377216 of 42412120 bytes (99.9%, 0.1s remaining) ...done. (100 seconds, 1 min) Downloading data from https://www.nitrc.org/frs/download.php/8260/A00028185_lh_preprocessed_fsaverage5_fwhm6.gii ... Downloaded 41697280 of 42402060 bytes (98.3%, 1.3s remaining) ...done. (81 seconds, 1 min) Traceback (most recent call last): File "/Users/genevieb/Documents/GitHub/napari/examples/surface_timeseries_.py", line 24, in brain_vertices, brain_faces = surface.load_surf_data(fsaverage['pial_left']) File "/Users/genevieb/mambaforge/envs/napari-dev/lib/python3.10/site-packages/nilearn/surface/surface.py", line 733, in load_surf_data data_part = _gifti_img_to_data(gii) File "/Users/genevieb/mambaforge/envs/napari-dev/lib/python3.10/site-packages/nilearn/surface/surface.py", line 684, in _gifti_img_to_data return np.asarray([arr.data for arr in gifti_img.darrays]).T.squeeze() ValueError: setting an array element with a sequence. The requested array has an inhomogeneous shape after 1 dimensions. The detected shape was (2,) + inhomogeneous part. ```

Expected behavior

I expected the example script to run without an error, and to display some timeseries surface data in napari.

Environment

Please copy and paste the information at napari info option in help menubar here:

napari: 0.4.13.dev917+g54c1dbf8 Platform: macOS-12.3.1-arm64-arm-64bit System: MacOS 12.3.1 Python: 3.10.8 | packaged by conda-forge | (main, Nov 22 2022, 08:25:29) [Clang 14.0.6 ] Qt: 5.15.6 PyQt5: 5.15.7 NumPy: 1.24.2 SciPy: 1.10.1 Dask: 2023.1.0 VisPy: 0.12.1 magicgui: 0.6.1 superqt: unknown in-n-out: 0.1.6 app-model: 0.1.1 npe2: 0.6.2

OpenGL:

GL version: 2.1 Metal - 76.3
MAX_TEXTURE_SIZE: 16384

Screens:

screen 1: resolution 1440x900, scale 2.0
screen 2: resolution 3440x1440, scale 1.0

Settings path:

/Users/genevieb/Library/Application Support/napari/napari-dev_8a798ce8db3962d716d33109f1c23fa2f783f2dd/settings.yaml

Additional context

Juan and I were talking about how to colour the vertices of a mesh according to some property, and he suggested I look at this example. We were both surprised when it gave an error, and thought that was something the CI should have caught.

Czaki commented 1 year ago

@sofroniewn, as you are the author of this example, maybe it is possible to take a look at it?

psobolewskiPhD commented 1 year ago

We were both surprised when it gave an error, and thought that was something the CI should have caught.

I assume the check at the top for nilearn prevents this from running on CI? It doesn't show up in the logs.

Edit: looking at the workflow and stuff further—yes. This example is skipped: https://github.com/napari/napari/blob/3ec4be1ae8eee50ab4912ba87981261cc94c075f/napari/_tests/test_examples.py#L23-L33

GenevieveBuckley commented 1 year ago

Yes it's skipped on the CI, but my point is that the example fails when you do install nilearn and try to run it.

Czaki commented 1 year ago

The examples test was moved to separate job. So maybe someone could review list of skipped examples and check if some of them could be enabled.

@GenevieveBuckley the only solution is that some core dev will spend time to fix it.

dalthviz commented 1 year ago

Hi, just in case, I checked the example and seems like the error comes from nilearn being incompatible with numpy 1.24.x. If you run the example with numpy 1.23.x it works:

imagen

Also, with numpy 1.23.x a VisibleDeprecationWarning message appears which points to the line inside nilearn where the error occurs when using numpy 1.24.x :

\site-packages\nilearn\surface\surface.py:684: VisibleDeprecationWarning: Creating an ndarray from ragged nested sequences (which is a list-or-tuple of lists-or-tuples-or ndarrays with different lengths or shapes) is deprecated. If you meant to do this, you must specify 'dtype=object' when creating the ndarray.
  return np.asarray([arr.data for arr in gifti_img.darrays]).T.squeeze()

So I guess either being sure to use numpy < 1.24.x or adding the dtype=object kwarg at the line mentioned inside nilearn should fix the example

psobolewskiPhD commented 1 year ago

If I'm parsing this correctly, this is an upstream issue in nilearn and not a bug in napari/the example?

GenevieveBuckley commented 1 year ago

FYI nilearn does have a PR open that could fix this error: https://github.com/nilearn/nilearn/pull/3644

napari / napari