Damien

I concur. I think there's need to update the repo to newer library versions if possible: quite a few mismatches in the environment yaml file. e.g: dgl needs the dglteam...

Here is a script to reproduce. This issue is quite pressing: it affects basically all easy workarounds for using Graph Neural Networks (PyG, DGL, etc) with Skorch. And the assertion...

I think this issue can be closed with a simple update in documentation. The behaviour described can be circumvented by using a torch Dataset object as X with some collate...

I can reproduce it, yes. And indeed the dumped version does die also. I am confused as the dumped model _did work_ for me at one point, but the cloned...

okay, after some checks: ```python from sklearn.datasets import make_regression from sklearn.base import clone import numpy as np import torch import skorch import dill import pickle dill.__version__ # 0.3.6 X, y...

we could print a trace of execution with the final fit and diff across, maybe?

I was thinking pdb might be of some help here, will report if I manage anything. In the meantime, I have found that dumping byref with dill solves the fail:...

I'll try and take a look. If you manage to track down the cause yourself, feel free to submit a pull request! (I'm working full time in industry now, with...

Hi Mersad, _"solving this bugs should not be something hard"_ -> It kinda is: I have not used this software in a while (since my PhD) and again, I am...

> > > Pip installing autografs (AuToGraFS) has requirement for numpy 1.14.6 (at least for linux 3.10.0-957.12.2.el7.x86_64) which does not find the "numpy.core._multiarray_umath" module. > > Subsequently upgrading to numpy...