Add matplotlib based drawer #304

mtreinish · 2021-04-06T21:23:52Z

This commit adds a a new python visualization module which contains a
matplotlib drawer for retworkx graphs. This first iteration of the
visualization function is porting networkx's equivalent functionality.
For now the api is basically identical to networkx's functionality
except it takes a retworkx graph object instead of a networkx graph
object. However, in follow-on commits to this branch the API will be
adapted to be more retworkx native (i.e. using callbacks to generate
labels from nodes and edges, etc).

Fixes #298

TODO:

This commit adds a a new python visualization module which contains a matplotlib drawer for retworkx graphs. This first iteration of the visualization function is porting networkx's equivalent functionality. For now the api is basically identical to networkx's functionality except it takes a retworkx graph object instead of a networkx graph object. However, in follow-on commits to this branch the API will be adapted to be more retworkx native (i.e. using callbacks to generate labels from nodes and edges, etc). To implement this a basic random_layout function is added to retworkx and a corresponding 2D layout type Pos2DMapping which is used to return the layouts efficiently (instead of creating a dict). This is still a rough draft and before this merges into master things will be updated. Fixes Qiskit#298

mtreinish · 2021-04-06T21:27:52Z

Just to show the basics here, running:

import retworkx
from retworkx import visualization
import matplotlib.pyplot as plt

g = retworkx.directed_gnp_random_graph(10, .9)
fig = plt.figure()
ax = fig.gca()
visualization.mpl_draw(g, ax=ax)
fig.savefig('test.png', dpi=600)

yields:

There is still a lot of work to do here, but the basics work for now.

coveralls · 2021-04-06T21:36:15Z

Pull Request Test Coverage Report for Build 853089054

0 of 0 changed or added relevant lines in 0 files are covered.
No unchanged relevant lines lost coverage.
Overall coverage increased (+0.03%) to 96.646%

Totals
Change from base Build 851076522:	0.03%
Covered Lines:	7060
Relevant Lines:	7305

💛 - Coveralls

This commit adds a new function random_layout() which is used to generate a random layout for a graph that can be used in visualization. This is necessary for building a matplotlib drawer (issue Qiskit#298 and a first draft of the implementation Qiskit#304). To make the function more efficient it also adds a new custom return type Pos2DMapping which is used to build an imutable readonly dict compatible result container for the output type from this function. Related to Qiskit#280

mtreinish · 2021-04-07T12:30:41Z

I've split out the layout functionality into #305

This commit adds a new function random_layout() which is used to generate a random layout for a graph that can be used in visualization. This is necessary for building a matplotlib drawer (issue Qiskit#298 and a first draft of the implementation Qiskit#304). To make the function more efficient it also adds a new custom return type Pos2DMapping which is used to build an imutable readonly dict compatible result container for the output type from this function. Related to Qiskit#280

This commit starts the process of reorganizing the matplotlib drawer into the final form. It flattens the api so there is a single public api entrypoint, mpl_draw() which exposes all the drawer options and returns a figure (if not in interactive mode). This is the first module and function in the visualization subpackage. In the future we will add some other ones built around graph.to_dot() that use pydot (so we avoid the boilerplate drawer code everywhere).

mtreinish · 2021-04-08T20:10:09Z

I think this is ready for review now. But this is on hold until #305 (which an early draft of is included inline for now) merges because for the mpl_draw() function to work we need a layout function available. Ideally the default will be #306 but #305 will probably be ready first.

* Add random_layout function and Pos2DMapping This commit adds a new function random_layout() which is used to generate a random layout for a graph that can be used in visualization. This is necessary for building a matplotlib drawer (issue #298 and a first draft of the implementation #304). To make the function more efficient it also adds a new custom return type Pos2DMapping which is used to build an imutable readonly dict compatible result container for the output type from this function. Related to #280 * Add release note * Update doc organization * Apply suggestions from code review Co-authored-by: Jielun (Chris) Chen <31495624+Chriscrosser3310@users.noreply.github.com> * Update releasenotes/notes/add-random-layout-c1c2751be971e5d0.yaml Co-authored-by: Nahum Rosa Cruz Sa <49600259+nahumsa@users.noreply.github.com> * Update releasenotes/notes/add-random-layout-c1c2751be971e5d0.yaml Co-authored-by: georgios-ts <45130028+georgios-ts@users.noreply.github.com> * Use [f64; 2] for center type This commit changes the input type of the center option to be a fixed 2 element array instead of a tuple. This still works as a tuple for the python interface but has a defined contiguous memory layout in rust. It also is consistent with the type for elements in the position mapping. Co-authored-by: Jielun (Chris) Chen <31495624+Chriscrosser3310@users.noreply.github.com> Co-authored-by: Nahum Rosa Cruz Sa <49600259+nahumsa@users.noreply.github.com> Co-authored-by: georgios-ts <45130028+georgios-ts@users.noreply.github.com>

nahumsa · 2021-05-15T20:17:46Z

tests/visualization/test_mpl.py

+    import matplotlib.pyplot as plt
+
+    mpl.use("PS")
+    plt.rcParams["text.usetex"] = False


Why it is needed to disable usetex?

It's mostly for reproducibility in the tests, matplotlib's usetex mode relies on external latex installations to run: https://matplotlib.org/stable/tutorials/text/usetex.html and even if present can cause different behavior on different platforms or latex distributions. So just simplicity it makes sense to disable it here so things run consistently between different systems (although I don't think it should come up in any of the tests).

That being said this line was just borrowed from networkx's equivalent drawer test module: https://github.com/networkx/networkx/blob/ead0e65bda59862e329f2e6f1da47919c6b07ca9/networkx/drawing/tests/test_pylab.py#L9

nahumsa · 2021-05-16T22:29:14Z

retworkx/visualization/matplotlib.py

+    # This is the only case handled differently from matplotlib
+    if (
+        np.iterable(edge_color)
+        and (len(edge_color) == len(edge_pos))


Could it be possible for edge_color to be a dict?
My reasoning is that we could have a default color and just change some of the nodes that we want to change using a dictionary with edge_pos as key values. Is that reasonable?

Well edge_pos isn't user facing (it's just internal array of node positions for the nodes of an edge), but I think I see what your getting at here. The only way we can do different colors for a subset of edges now is multiple calls to mpl_draw with a subset of nodes and edges (the tests show examples of this). In networkx this kind of thing is handled normally by calling draw_edges() separately, but for this PR I opted for a single function for simplicity. This however causes some tension for more custom things like different colors for specific edges or nodes. I think there are 3 ways to address this:

Make the internal draw_nodes, draw_edges, etc functions part of the user facing api like networkx does.

Add support for dicts/mappings as an input to edge_colors, node_colors, node_shape (and maybe node_size) so that these properties can be set per node and per edge.

Add callback function support for edge_colors, node_colors, etc so that when we evaluate colors we can just call the user provided function to get a color string (or whatever other property.

None of these options are mutually exclusive so we can do any combination of them. How do you think we should proceed here?

The most natural solution, at least for me, is the first one. However, exposing those functions may lead to some trouble. For the long term, I think the second solution would be better. What you think about doing both 1 and 2?

I was on the fence about doing 1, but I agree it seems like it makes the most sense for user expectations. I think it'll be good as a followup (either before or after release). Same for 2, we can do that in a follow up. I'm going to merge this now (with your review) I'll open an issue about this after merging.

nahumsa

LGTM. Overall it looks great, and the tests are excellent.
The only comment is to make the drawer function more customizable regarding the coloring of the graph, but this could be a future improvement.

Since Qiskit#304 merged building the retworkx documentation has required that matplotlib be installed for generating the matplotlib visualizations in the documentation of `mpl_draw()`. However that PR neglected to add it to the read the docs conda environment configuration. This commit corrects this oversight to hopefully fix the documentation builds on read the docs.

* Add matplotlib to read the docs builds Since #304 merged building the retworkx documentation has required that matplotlib be installed for generating the matplotlib visualizations in the documentation of `mpl_draw()`. However that PR neglected to add it to the read the docs conda environment configuration. This commit corrects this oversight to hopefully fix the documentation builds on read the docs. * Fix docs theme formatting This commit fixes some docs theme formatting issues to make sure the compiled HTML docs are correctly formatted. It mainly adds missing template files and configuration so the qiskit-sphinx-theme package has all the necessary details to correctly render. * Update copyright string

mtreinish mentioned this pull request Apr 6, 2021

Add spring_layout/fruchterman_reingold_layout function to retworkx #280

Closed

mtreinish mentioned this pull request Apr 7, 2021

Add optional keywords to GraphOptimizationApplication.draw qiskit-community/qiskit-optimization#52

Open

mtreinish mentioned this pull request Apr 7, 2021

Add random_layout function and Pos2DMapping #305

Merged

mtreinish added 8 commits April 8, 2021 09:35

Remove pointless retworkx_ prefix and add edge label options

f79d775

Add tests

1d5be2c

Relax matplotlib requirement

f3f8bb5

Fix mpl<3.4 compat

426e042

Fix test skip condition

b11af92

Add release notes

506f9ef

Merge branch 'master' into mpl-draw

8f649ae

mtreinish changed the title ~~[WIP] Add matplotlib based drawer~~ Add matplotlib based drawer Apr 8, 2021

mtreinish added the on hold label Apr 8, 2021

Merge remote-tracking branch 'origin/main' into mpl-draw

03ed74a

mtreinish added this to the 0.9.0 milestone Apr 23, 2021

mtreinish removed the on hold label Apr 23, 2021

mtreinish added 3 commits April 24, 2021 07:45

Merge branch 'main' into mpl-draw

0106125

Merge branch 'main' into mpl-draw

e7e0275

Run black

039f400

nahumsa reviewed May 15, 2021

View reviewed changes

Merge branch 'main' into mpl-draw

634c5f3

nahumsa reviewed May 16, 2021

View reviewed changes

nahumsa approved these changes May 17, 2021

View reviewed changes

mtreinish mentioned this pull request May 18, 2021

Add pydot based drawer #321

Closed

mtreinish added 5 commits May 18, 2021 07:07

Merge branch 'main' into mpl-draw

ae7ffbf

Merge remote-tracking branch 'origin/main' into mpl-draw

0d973b3

Update default layout method to spring layout

9906e80

Merge branch 'mpl-draw' of github.com:mtreinish/retworkx into mpl-draw

3db6187

Merge branch 'main' into mpl-draw

906a5d4

mtreinish merged commit 79c38f2 into Qiskit:main May 19, 2021

mtreinish deleted the mpl-draw branch May 19, 2021 18:25

mtreinish mentioned this pull request May 21, 2021

Add matplotlib to read the docs builds #329

Merged

mtreinish mentioned this pull request May 28, 2021

Rename edge_list arg with edgelist to be compatible with networkx #341

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Add matplotlib based drawer #304

Add matplotlib based drawer #304

mtreinish commented Apr 6, 2021 •

edited

Loading

mtreinish commented Apr 6, 2021 •

edited

Loading

coveralls commented Apr 6, 2021 •

edited

Loading

mtreinish commented Apr 7, 2021

mtreinish commented Apr 8, 2021

nahumsa May 15, 2021

mtreinish May 16, 2021

nahumsa May 16, 2021

mtreinish May 17, 2021

nahumsa May 17, 2021 •

edited

Loading

mtreinish May 18, 2021

nahumsa left a comment •

edited

Loading

Add matplotlib based drawer #304

Add matplotlib based drawer #304

Conversation

mtreinish commented Apr 6, 2021 • edited Loading

mtreinish commented Apr 6, 2021 • edited Loading

coveralls commented Apr 6, 2021 • edited Loading

Pull Request Test Coverage Report for Build 853089054

💛 - Coveralls

mtreinish commented Apr 7, 2021

mtreinish commented Apr 8, 2021

nahumsa May 15, 2021

Choose a reason for hiding this comment

mtreinish May 16, 2021

Choose a reason for hiding this comment

nahumsa May 16, 2021

Choose a reason for hiding this comment

mtreinish May 17, 2021

Choose a reason for hiding this comment

nahumsa May 17, 2021 • edited Loading

Choose a reason for hiding this comment

mtreinish May 18, 2021

Choose a reason for hiding this comment

nahumsa left a comment • edited Loading

Choose a reason for hiding this comment

mtreinish commented Apr 6, 2021 •

edited

Loading

mtreinish commented Apr 6, 2021 •

edited

Loading

coveralls commented Apr 6, 2021 •

edited

Loading

nahumsa May 17, 2021 •

edited

Loading

nahumsa left a comment •

edited

Loading