Stream: jupyterlab-hub

Topic: Error displaying widget: model not found

Brian Bonnlander (Apr 28 2023 at 00:22):

Hi everyone,

I'm using a self-created conda environment that explicitly installs dask_jobqueue and ipywidgets. It also includes jupyterlab and a bunch of extensions, but I'm not installing these packages explicitly; they are somehow being installed as a dependencies to other packages.

When I am using this conda environment in jupyterhub, I am unable to see the distributed client widget after running this code; instead I get the error message "Error displaying widget: model not found":

from dask_jobqueue import PBSCluster

num_jobs = 20
walltime = '0:20:00'
memory='10GB'

cluster = PBSCluster(cores=1, processes=1, walltime=walltime, memory=memory, queue='casper',
                     resource_spec='select=1:ncpus=1:mem=10GB',)
cluster.scale(jobs=num_jobs)


from distributed import Client
client = Client(cluster)
cluster

Error displaying widget: model not found

Has anyone seen this message before, and knows how to get around it?

Deepak Cherian (Apr 28 2023 at 14:14):

I've seen it and suspect its a version mismatch between your env and the hub. The client repr works so i didn't look for a workaround

Brian Bonnlander (Apr 28 2023 at 17:00):

OK, thanks! I have not been using the hub much over the past couple of years and always used my custom environment before that. If there is some special package that I want to use not available from those provided, is it up to me to provide a complete env that does not conflict with the hub's jupyter settings?

Brian Bonnlander (Apr 28 2023 at 20:49):

After some trial and error, it looks like if the custom environment includes the ipywidgets package, then jupyterlab and some associated packages also get installed.

Having jupyterlab installed in a custom conda environment appears to create conflicts with jupyterhub settings, which is probably to be expected.

If jupyterlab and other jupyter-related packages are not in the custom environment, then the Client widget is displayed properly for me.

Dhamma Kimpara (he/him) (Jul 11 2023 at 22:19):

I've seen this previous post https://zulip2.cloud.ucar.edu/#narrow/stream/16-jupyterlab-hub/topic/Error.20displaying.20widget.3A.20model.20not.20found/near/80053

but I tried making a fresh environment making sure no jupyterlab/notebook dependencies were installed. I still get Error displaying widget: model not found

Here is the code I'm testing the packaging with from this stackx:

import matplotlib.pyplot as plt
import asyncio
%matplotlib widget
fig = plt.figure()


async def update():
    for i in range(5):
        print(i)
        x = list(range(i + 2))
        xx = [x**2 for x in x]
        plt.clf()
        plt.plot(x, xx)
        fig.canvas.draw()
        await asyncio.sleep(1)


loop = asyncio.get_event_loop()
loop.create_task(update());

Brian Bonnlander (Jul 12 2023 at 15:38):

When you try "conda activate" and "conda list | grep jupyter", do you see any jupyter-related packages?

Dhamma Kimpara (he/him) (Jul 12 2023 at 17:18):

i see:

jupyter_client
jupyter_core
jupyterlab_widgets

Brian Bonnlander (Jul 12 2023 at 17:30):

I don't have much experience with updating an interactive figure in a loop. I do see some ESDS blog posts on the topic, maybe it is worth trying a different example using holoviews or hvplot. See for example: https://ncar.github.io/esds/posts/2021/intake-esm-holoviews-diagnostics/

Last updated: May 16 2025 at 17:14 UTC