Trame support and new Jupyter backend #3385

banesullivan · 2022-09-29T18:25:21Z

This adds streamlined support for using PyVista with Trame and adds a new Trame-based Jupyter backend.

The new Trame-based backend is the backend that we will try to centralize around per #3690 and with it's addition comes an immediate deprecation of the ipyvtklink backend.

Trame is a new framework for building lightweight, reactive web applications in a stateful manner. It has great support for VTK and thus PyVista. These changes add a trame module to PyVista that streamlines usage of the trame-vtk interface for use with a PyVista Plotter object.

While the bulk of this PR is actually focused on the new Jupyter backend, this is also provided streamlined support for using PyVista and Trame to make reactive web apps. There are a few examples web apps included under: examples_trame/

Basic Trame Web App

Screen.Recording.2022-09-29.at.12.01.39.PM.mov

Advanced Trame Web App

Screen.Recording.2023-01-21.at.12.12.57.PM.mov

New Jupyter Backend

Screen.Recording.2023-01-21.at.12.16.05.PM.mov

Jupyter Backend

See usage instructions in docs but simply

import pyvista as pv
from pyvista import examples

await pv.set_jupyter_backend('trame')

mesh = examples.load_random_hills()
arrows = mesh.glyph(scale="Normals", orient="Normals", tolerance=0.05)

p = pv.Plotter()
p.add_mesh(mesh, scalars="Elevation", cmap="terrain", smooth_shading=True)
p.add_mesh(arrows, color="black")
p.show()

Additional Changes

This PR required a significant amount of work to PyVista and sneaks in a few changes that really could've been their own PRs. These include:

An an internal refactor of ren_win to render_window as a @property
The addition of a new suppress_rendering @property on the BasePlotter that will suppress calls to vtkRenderWindow.Render() for guarantee use of the PyVista plotter without a virtual frame buffer
A refactor/change to the _id_name attribute of the plotter so that it is a safe string to use as a variable name (added a P_ prefix)
Addition of an actors @property on BasePlotter to get all actors on the Plotter
A new mechanism for adding on-render callbacks. See add_on_render_callback. This has two modes (to register with actual VTK render events or be called explicitly by PyVista's BasePlotter.render() method (often to be paired with suppress_rendering)
Removal of the already deprecated use_ipyvtk argument and PYVISTA_USE_IPYVTK environment variable

Resolves and Closes

This PR renders many issue non-actionable as the new Trame backend handles many issue with the other existing backends

Closes pyvista (pythreejs or ipygany backend) incompatible with ipywidgets>7.7.2. #3274: since ipyvtklink is deprecated, this is no longer relevant or actionable
Closes Display grid with pythreejs backend #3091: use the new trame backend for grid lines
Resolves "pythreejs" jupyter backend requires virtual framebuffer, contrary to documentation #3221: suppress_rendering should fix this when combined with the pythreejs backend
Closes JupyterLab : The widget returned by plotter.show(return_viewer=True) isn't updated when the plotter is #2940: this new Trame backend is the only backend that supports dynamic updating
Closes Coincident topology issues with pythreejs jupyter_backend #3482: Recommend using the new trame backend to resolve coincident topology issues
Resolves Scalar bar color cannot be changed in trame local view #3771

We will continue to address the remaining issues listed in #3690

Upstream work

A lot of upstream work was necessary for this new Jupyter backend and much thanks to @jourdain! Here are some (not all) of the relevant work:

Future Work

Screenshot button for client-side rendering (currently only works for server-side rendering)
Support for picking
Volume rendering support on the client-side viewer (ref vtkSmartVolumeMapper serializer fails Kitware/trame-vtk#14)

pyvista/trame/__init__.py

codecov · 2022-10-05T03:10:03Z

Codecov Report

Merging #3385 (ff085d5) into main (bf670b6) will increase coverage by 0.05%.
The diff coverage is 92.29%.

@@            Coverage Diff             @@
##             main    #3385      +/-   ##
==========================================
+ Coverage   94.13%   94.19%   +0.05%     
==========================================
  Files          87       91       +4     
  Lines       19111    19581     +470     
==========================================
+ Hits        17991    18444     +453     
- Misses       1120     1137      +17

pyvista/plotting/plotting.py

banesullivan · 2022-10-13T23:14:38Z

MNE integration test is failing as:

Traceback (most recent call last):
  File "/opt/hostedtoolcache/Python/3.10.7/x64/lib/python3.10/site-packages/pyvistaqt/plotting.py", line 309, in _render
    return BasePlotter.render(self, *args, **kwargs)
  File "/home/runner/work/pyvista/pyvista/pyvista/plotting/plotting.py", line 1606, in render
    if self.render_window is not None and not self._first_time:
  File "/opt/hostedtoolcache/Python/3.10.7/x64/lib/python3.10/site-packages/pyvistaqt/rwi.py", line 4[26](https://github.com/pyvista/pyvista/actions/runs/3246255962/jobs/5324798559#step:13:27), in __getattr__
    raise AttributeError(self.__class__.__name__ +
AttributeError: BackgroundPlotter has no attribute named render_window. Did you mean: '_RenderWindow'?
Fatal Python error: Aborted

Trying to decide if I should implement the render_window property in BackgroundPlotter or if I should just tag this onto BasePlotter....

Jimmy-KL · 2022-11-24T09:41:18Z

@banesullivan I tried this trame feature in pyvista and found a strange thing. I have a model with several data arrays, I want to want to change the data array and the data range of the model and scalar bar can be updated. First I remove the old scalar bar, then I add a new one. Here is the code for this part:

@state.change("data_array")
mesh = pv.read(os.path.join(file_path, file_anme))
mesh.set_active_scalars(mesh.array_names[0])
plotter = pv.Plotter(off_screen=True)
actor = plotter.add_mesh(mesh)
plotter.set_background('lightgrey')

def update_data_array(data_array=mesh.array_names[0], **kwargs):
    mesh.set_active_scalars(data_array)
    plotter.remove_scalar_bar()
    plotter.add_scalar_bar(title=data_array)
    min_value = mesh[data_array].min()
    max_value = mesh[data_array].max()
    plotter.update_scalar_bar_range([min_value, max_value])
    ctrl.view_update()

with SinglePageLayout(server) as layout:
    with layout.toolbar:
            vuetify.VSpacer()
            vuetify.VSelect(
                label="属性选择",
                v_model=("data_array", mesh.array_names[0]),
                items=("array_list", mesh.array_names),
                hide_details=True,
                dense=True,
                outlined=True,
                classes="pt-1 ml-2",
                style="max-width: 150px",
            )
    with layout.content:
            with vuetify.VContainer(
                fluid=True,
                classes="pa-0 fill-height",
            ):
                view = PyVistaLocalView(plotter)
                ctrl.view_update = view.update
                ctrl.view_reset_camera = view.reset_camera

This can work for a simple pyvista application and also for the remote view of trame. But if I use the local view, the old scalar bar cannot be removed unless I refresh the page.
The initial state:

After change the data array:

Refresh the page and everything is OK:

It maybe a problem of trame or I didn't use it properly. But I think I can give some feedback here. Or do you have a solution for this problem?

jourdain · 2022-11-29T01:17:16Z

This is most likely a little issue with the local view. I would think that if you update trame-vtk it may partially fix that problem. I'm still in the process on properly fixing that.

Jimmy-KL · 2022-12-01T09:13:22Z

This is most likely a little issue with the local view. I would think that if you update trame-vtk it may partially fix that problem. I'm still in the process on properly fixing that.

I was using trame-vtk == 2.0.8. I updated to the latest version 2.0.9, but the model disappeared, and nothing showed. However, for remote view, it's OK.

Co-authored-by: Eric Larson <larson.eric.d@gmail.com>

banesullivan · 2023-01-26T17:15:51Z

@larsoner, thanks for all of the feedback! I think I've addressed everything you brought up, but please double-check.

Regarding the ipyvtklink deprecation: Hopefully, this is okay now considering the trame conda-forge recipes should be up in no time (thanks to you! 😄 )

I really, really want to merge this today unless something significant comes up -- please let me know what you think.

FYI, I noticed the UI menu bar issue in your screenshot and will try to sneak in a fix for that

banesullivan · 2023-01-26T17:23:16Z

For the no-serializer issue, a better approach would be to notify the user (in the GUI) that "some content could not be available for local rendering" and then offer the option to switch to remote rendering or ignore.

@jourdain, I feel like that would be best implemented in trame-vtk. What do you think?

See Kitware/trame-vtk#15

akaszynski · 2023-01-26T17:34:12Z

Further, you can use the 'fixed_point' volume mapper, and all will work in client mode as is.

Should we just change the default mapper then if it detects that it's in a notebook with either trame or client enabled? We don't have to fix it, but I also don't like empty plots.

larsoner

Other than one remaining bug found via my minimal testing, LGTM!

pyvista/trame/jupyter.py

larsoner · 2023-01-26T17:34:38Z

pyvista/trame/jupyter.py

+        server_proxy_enabled=server_proxy_enabled,
+        server_proxy_prefix=server_proxy_prefix,
+        **kwargs,
+    )


... Add this

Suggested change

)

)

for ki, key in enumerate(('width', 'height')):

if key not in kwargs:

kwargs[key] = f'{plotter.window_size[ki]}px'

Or something similar somewhere. FWIW this code fixes the window sizing for MNE-Python

Ahh, nice idea!

Ehh... I really want the default to be 100% rather than matching the plotter size. With these changes it makes the output not so elegant by default:

Where as with width: 100%; height: 600px; its much nicer no matter the window size:

I reverted this suggestion

But didn't it look that way with ipyvtklink already, suggesting that this maintains previous behavior?

Okay with me in principle for 100% -- 50% didn't seem to work all that well -- assuming there is some nice way in the public API to easily specify a fixed size. What is it with the current code? It seems like the current version completely ignores Plotter(..., window_size=...) which seems unexpected/not backward compatible...

One way out of this might be to have Plotter(..., window_size='auto') as a backward compat default which means "(600px, 600px) for all backends except trame, which will use ('100%', '600px')" or something.

@larsoner, are you okay with me merging as is and following up in a separate PR (before the release) to expose a better mechanism for controlling the widget size?

Tracking here: #3879

Will try to address ASAP

banesullivan · 2023-01-26T18:03:38Z

@akaszynski

Should we just change the default mapper then if it detects that it's in a notebook with either trame or client enabled? We don't have to fix it, but I also don't like empty plots.

We could... but I'd rather try to just fix this upstream and follow up with this sort of patch if necessary

Co-authored-by: Eric Larson <larson.eric.d@gmail.com>

This reverts commit e798b85.

akaszynski

Reapproving. Really appreciate everyone's feedback here, and if there are any issues post-merge we'll be quick to fix them in follow-up PRs.

akaszynski · 2023-01-26T20:49:50Z

@banesullivan and @larsoner, let's merge this (after 24 ish hours green) and then follow-up with smaller PRs. I've had my fair share of 2,000+ LOC PRs and I think it will be easier to track issues and fix them in individual PRs.

banesullivan · 2023-01-26T21:10:41Z

I think it will be easier to track issues and fix them in individual PRs.

Exactly. This PR is touching a bunch of code at this point and it'll be much easier to track and review further changes as new PRs

larsoner

Sounds good!

bilke · 2023-02-06T21:15:30Z

First thanks for this awesome new backend! But now a question: is the Trame backend available in the conda package? I get No module named 'trame' errros.

jourdain · 2023-02-06T21:19:41Z

It should but it could be a WIP. I know that @banesullivan is working toward streamlining some of those conda recipes.

banesullivan · 2023-02-06T21:28:10Z

The trame recipes just landed on conda-forge a few hours ago.

conda install -c conda-forge 'pyvista>=0.38.1' matplotlib ipywidgets trame jupyterlab

Should get you everything you need (just confirmed in a fresh env locally)

bilke · 2023-02-07T08:22:19Z

Dear @banesullivan, thanks installation worked so far but I always get a 404 on the trame widget, e.g.:

[W 2023-02-07 08:20:40.754 ServerApp] 404 GET /proxy/33025/index.html?ui=P_0x7f8b6409aa10_0&reconnect=auto (3dd9915b860a49639fc672524abd9e7d@141.65.34.101) 0.57ms referer=http://[IP]:8888/lab

I tried with and without PYVISTA_TRAME_SERVER_PROXY_ENABLED="True".

banesullivan · 2023-02-07T14:55:52Z

Would you please open a new issue to track this

Add streamlined support for trame

f073e8b

github-actions bot added the enhancement Changes that enhance the library label Sep 29, 2022

adeak reviewed Sep 29, 2022

View reviewed changes

pyvista/trame/__init__.py Outdated Show resolved Hide resolved

Merge branch 'main' into feat/trame

7ff80f8

banesullivan added 3 commits October 5, 2022 08:47

Add actors property

cb14c05

Add local view example

7732e41

Cleanup

60d156f

banesullivan commented Oct 11, 2022

View reviewed changes

pyvista/plotting/plotting.py Outdated Show resolved Hide resolved

banesullivan added 3 commits October 13, 2022 16:56

Refactor render_window access

eabf816

Further refactorin and cleanup of ren_win

8a2cca1

Fix del operation on ren_win in test

8a63d16

banesullivan and others added 8 commits October 13, 2022 19:06

Merge branch 'main' into feat/trame

5980913

Move render_window property

53bc134

Merge branch 'main' into feat/trame

a13f40f

add missing :

d75c5b8

Add trame jupyter backend

48de6fd

server_proxy default False

259c020

Use weakref

5e6c345

Update setup.py

b1613c0

banesullivan mentioned this pull request Oct 24, 2022

Coincident topology issues with pythreejs jupyter_backend #3482

Closed

banesullivan mentioned this pull request Nov 9, 2022

Support ipywidgets >=8 Kitware/ipyvtklink#36

Open

Merge branch 'main' into feat/trame

088c61b

Jimmy-KL mentioned this pull request Nov 24, 2022

Can not forward events without a current renderer on the interactor Kitware/trame#154

Closed

1 task

Merge branch 'main' into feat/trame

31e24fe

banesullivan self-assigned this Dec 1, 2022

banesullivan and others added 5 commits January 26, 2023 08:29

Use custom widget with viewer reference and support return_viewer

0322ce5

Update pyvista/jupyter/__init__.py

3eb4fc9

Co-authored-by: Eric Larson <larson.eric.d@gmail.com>

Formatting

06f3388

Disable render_window_serializer logger in Jupyter

c67a5ff

Fix test

d633ea1

larsoner previously approved these changes Jan 26, 2023

View reviewed changes

Update pyvista/trame/jupyter.py

d632eac

Co-authored-by: Eric Larson <larson.eric.d@gmail.com>

banesullivan dismissed larsoner’s stale review via d632eac January 26, 2023 18:42

banesullivan and others added 2 commits January 26, 2023 11:43

Update pyvista/trame/jupyter.py

e798b85

Co-authored-by: Eric Larson <larson.eric.d@gmail.com>

Revert "Update pyvista/trame/jupyter.py"

ff085d5

This reverts commit e798b85.

banesullivan mentioned this pull request Jan 26, 2023

Expose mechanism to easily control the Jupyter trame view size #3879

Closed

akaszynski approved these changes Jan 26, 2023

View reviewed changes

larsoner approved these changes Jan 26, 2023

View reviewed changes

banesullivan merged commit 36ab0e1 into main Jan 27, 2023

banesullivan deleted the feat/trame branch January 27, 2023 16:49

This was referenced Jan 31, 2023

Fix and improve chart interaction #3756

Merged

Release 0.38 #3935

Merged

dcbr mentioned this pull request Feb 2, 2023

Rework render event callbacks #3945

Draft

banesullivan mentioned this pull request Jun 2, 2023

Decouple core and plotting APIs to improve imports and tighten public namespace #4486

Merged

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Trame support and new Jupyter backend #3385

Trame support and new Jupyter backend #3385

banesullivan commented Sep 29, 2022 •

edited

codecov bot commented Oct 5, 2022 •

edited

banesullivan commented Oct 13, 2022

Jimmy-KL commented Nov 24, 2022

jourdain commented Nov 29, 2022

Jimmy-KL commented Dec 1, 2022

banesullivan commented Jan 26, 2023

banesullivan commented Jan 26, 2023 •

edited

akaszynski commented Jan 26, 2023

larsoner left a comment

larsoner Jan 26, 2023

banesullivan Jan 26, 2023

banesullivan Jan 26, 2023

banesullivan Jan 26, 2023

larsoner Jan 26, 2023

banesullivan Jan 26, 2023

larsoner Jan 26, 2023

banesullivan Jan 26, 2023

banesullivan commented Jan 26, 2023

akaszynski left a comment

akaszynski commented Jan 26, 2023

banesullivan commented Jan 26, 2023

larsoner left a comment

bilke commented Feb 6, 2023

jourdain commented Feb 6, 2023

banesullivan commented Feb 6, 2023

bilke commented Feb 7, 2023

banesullivan commented Feb 7, 2023

-    )
+    )
+    for ki, key in enumerate(('width', 'height')):
+        if key not in kwargs:
+            kwargs[key] = f'{plotter.window_size[ki]}px'

Trame support and new Jupyter backend #3385

Trame support and new Jupyter backend #3385

Conversation

banesullivan commented Sep 29, 2022 • edited

Basic Trame Web App

Advanced Trame Web App

New Jupyter Backend

Jupyter Backend

Additional Changes

Resolves and Closes

Upstream work

Future Work

codecov bot commented Oct 5, 2022 • edited

Codecov Report

banesullivan commented Oct 13, 2022

Jimmy-KL commented Nov 24, 2022

jourdain commented Nov 29, 2022

Jimmy-KL commented Dec 1, 2022

banesullivan commented Jan 26, 2023

banesullivan commented Jan 26, 2023 • edited

akaszynski commented Jan 26, 2023

larsoner left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

banesullivan commented Jan 26, 2023

akaszynski left a comment

Choose a reason for hiding this comment

akaszynski commented Jan 26, 2023

banesullivan commented Jan 26, 2023

larsoner left a comment

Choose a reason for hiding this comment

bilke commented Feb 6, 2023

jourdain commented Feb 6, 2023

banesullivan commented Feb 6, 2023

bilke commented Feb 7, 2023

banesullivan commented Feb 7, 2023

banesullivan commented Sep 29, 2022 •

edited

codecov bot commented Oct 5, 2022 •

edited

banesullivan commented Jan 26, 2023 •

edited