Import error npyx 4.0.6, python3.9 NameError: name 'KronDecomposed' is not defined`

[Antorithms]

Dear Maxime,
Thanks for developing this nice package!
With npyx version 4.0.6 I get the error below after installing in a conda environment with python==3.9 and trying to import from npyx.spk_wvf import get_peak_chan. Can you reproduce the issue?
Thanks for your help.
Anto

In [2]: from npyx.spk_wvf import get_peak_chan
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
Cell In[2], line 1
----> 1 from npyx.spk_wvf import get_peak_chan

File -\lib\site-packages\npyx\__init__.py:7
      1 # -*- coding: utf-8 -*-
      3 from . import utils, inout, gl, spk_t, spk_wvf, corr, stats, plot,\
      4               behav, merger, circuitProphyler, feat, metrics,\
      5               info, model, h5, testing, ml, datasets
----> 7 from .c4 import dataset_init, misc
      9 try:
     10     import torch

File -\lib\site-packages\npyx\c4\__init__.py:5
      2 import torch
      4 from .acg_augmentations import *
----> 5 from .acg_vs_firing_rate import *
      6 from .dl_transforms import *
      7 from .dl_utils import *

File -\lib\site-packages\npyx\c4\acg_vs_firing_rate.py:24
     18 from .dataset_init import (
     19     extract_and_check,
     20     extract_and_merge_datasets,
     21     get_paths_from_dir,
     22 )
     23 from .monkey_dataset_init import MONKEY_CENTRAL_RANGE, get_lisberger_dataset
---> 24 from .predict_cell_types import get_n_cores, redirect_stdout_fd
     26 BIN_SIZE = 1
     27 WIN_SIZE = 200

File -\lib\site-packages\npyx\c4\predict_cell_types.py:42
     36 from .misc import require_advanced_deps
     37 from .plots_functions import (
     38     C4_COLORS,
     39     plot_features_1cell_vertical,
     40     plot_survival_confidence,
     41 )
---> 42 from .run_deep_classifier import (
     43     CustomDataset,
     44     encode_layer_info,
     45     ensemble_predict,
     46     load_ensemble,
     47     prepare_classification_dataset,
     48 )
     50 MODELS_URL_DICT = {
     51     "base": "https://figshare.com/ndownloader/files/42117042?private_link=2530fd0da03e18296d51",
     52     "mli_clustering": "https://figshare.com/ndownloader/files/42129447?private_link=d508ebc51d544ed8cd4c",
     53     "layer_information": "https://figshare.com/ndownloader/files/42853297?private_link=6531855c261b7bad032d",
     54     "layer_information_mli_clustering": "https://figshare.com/ndownloader/files/42853708?private_link=3a81e48aff77d844a402",
     55 }
     57 HESSIANS_URL_DICT = {
     58     "base": "https://figshare.com/ndownloader/files/42117033?private_link=2530fd0da03e18296d51",
     59     "mli_clustering": "https://figshare.com/ndownloader/files/42129435?private_link=d508ebc51d544ed8cd4c",
     60     "layer_information": "https://figshare.com/ndownloader/files/42853141?private_link=6531855c261b7bad032d",
     61     "layer_information_mli_clustering": "https://figshare.com/ndownloader/files/42853567?private_link=3a81e48aff77d844a402",
     62 }

File -\lib\site-packages\npyx\c4\run_deep_classifier.py:354
    348         new_probs = new_probs.numpy()
    350     return new_probs
    353 @require_advanced_deps("torch", "torchvision", "laplace")
--> 354 def get_kronecker_hessian_attributes(*kronecker_hessians: KronDecomposed):
    355     hessians = []
    356     for h in kronecker_hessians:

NameError: name 'KronDecomposed' is not defined

[m-beau]

Hi there!

It seems like you are trying to import a dependency that you shouldn't need when using the boilerplate version of npyx: would you mind uninstalling it fully and reinstalling it from pip with 'pip install npyx', and let me how that goes?

Hope you find it useful :)

[Antorithms]

Tried again but the error persists.
Thanks for your help.
Anto

[m-beau]

All right - can you tell me what was your exact procedure to install npyx (did you use a fresh conda environment?), and also tell me what is the installed version (you can find this out by running 'pip list' and scrolling to npyx in your terminal)

[Antorithms]

version is 4.0.6
After a few tests I think the issue might be causewd by other packages installing packages that are only there in a dev npyx version (?) and tricking npyx in believing that and trying to load functions that are not there, could that be the case?
Thanks,
Anto

[m-beau]

Hi Anto,

Did you manage to get npyx to work for you? Indeed, it seems like you were importing packages that you shouldn't be importing when using the basic version - I would recommend uninstalling npyx (pip uninstall npyx) and reinstalling it from pip.

[Antorithms]

The thing is that in my case npyx needs to be part of a larger environment.
you can try to reproduce running this in a newly formed conda env
conda install -c conda-forge pip ffmpeg cupy cudatoolkit
pip install joblib jupyterlab ccf-streamlines npyx tdt cebra scipy numpy matplotlib nemos ffmpeg-python matplotlib-venn scikit-learn dlc2kinematics ipympl rastermap affinewarp moviepy

[m-beau]

Hmmm I cannot reproduce your error with the MWE that you provided.

It does sound like, because you have torch installed for cebra, that npyx believes that the 'developer' version is installed, but it should not pose any problem and I cannot reproduce your exact issue - this line should set KronDecomposed to None when laplace is not found:

try:
    from laplace import BaseLaplace, Laplace
    from laplace.utils import KronDecomposed
except ImportError:
    # Used for type hints, will be ignored if not installed
    BaseLaplace = Laplace = KronDecomposed = None

Can you confirm that this MWE does cause an error for you:

conda create -n test_env python==3.11
conda install -c conda-forge pip ffmpeg cupy cudatoolkit
pip install joblib jupyterlab ccf-streamlines npyx tdt cebra scipy numpy matplotlib nemos ffmpeg-python matplotlib-venn scikit-learn dlc2kinematics ipympl rastermap affinewarp moviepy
python

(now, in the python session)
from npyx.spk_wvf import get_peak_chan

[m-beau]

Hi Anto,

Any update?