Additional features for plotly exporters

[victoriaono]

Description

This PR contains changes to the following viewers, adding features to plotly exporters not previously rendered:

Scatter 2D

• connecting lines
• unfilled/filled circles
• error bars
• vectors
• point size scaling
• background and axis color propagation
• logarithmic scales
• non-Cartesian projections

Unresolved issues:

• Does not reflect accurate vector arrow size and placement of the arrow
• Does not render graph with appropriate axis ranges for larger data
• Labels are not shown around non-Cartesian projection graphs

Scatter 3D

• background and axis color propagation
• error bars
• vectors
• point size scaling

  Unresolved issues:

• Error bars do not extend outside of the plot box
• Shared axes cannot be made visible
• Does not render graph with appropriate axis ranges for larger data
• Vectors can only be shown as cones

New viewers

Histogram

Unresolved issues:

• The last bin does not appear if the upper limit is the same as the maximum data value
• Does not render graph with appropriate axis ranges for larger data
• Data cannot be binned logarithmically

Image

Unresolved issues:

• For larger data, the exporter file is too large to open
• Certain colorscales cannot be exported

[Carifio24]

Edit: As issues are resolved, I'll cross them out here.

@victoriaono Sorry it took me a while to look this over. Overall things are looking good, but I have some comments on how things stand right now:

General

• There are some code style issues that our checker picked up, so these should be fixed. To run the codestyle tests, you can do the following in your environment
  • pip install tox
  • tox -e codestyle

2D Scatter Viewer

Filled/Unfilled Circles

• No issues, looks great

Background/foreground color propagation

• No issues, looks great

Lines

Solid Lines

• The default gray color for a layer in glue is represented as the string '0.35', so we need to be able to handle that. Currently the exporter gives an error if this is the case

Colormaps

• Overall I like the idea of this implementation, this is a nice workaround to the lack of plotly support. I think it might be nice to have smoother color transitions, since the midpoint "seam" is often pretty obvious. This would require more than 1 line per point.

  • One way to do this is potentially to use more more divisions and take an appropriate weighted average of the color (see the first comment here: https://stackoverflow.com/a/29576746)

Edit: This isn't an issue because it actually matches the glue implementation! For some reason I had thought the glue colored lines were more continuous.

Point size scaling

• Everything looks good here, but I'm a bit curious why the sizes are restricted to be between 3 and 30. Is these bounds that we use somewhere in glue?

Error bars

• If the colormap is linear, the points currently show regardless of whether the points are visible in glue or not

Vectors

• These have the same issue as the lines do when '0.35' is the color

• If a layer has vectors displayed, subsequent layers will not appear in the exported plotly graph

• I think we discussed this a couple of weeks ago, but the vector arrows and positioning need some tweaking. It would be nice to support glue's vector origin property. One way to do this would be to apply the appropriate translation to the points used to create the vectors

Full-sphere projections

• Currently, every layer gets colored with the color of the first layer

Error Reporting

• In addition to flagging when there are incompatibilities between glue and plotly, we also want to notify the user when something goes wrong. It's annoying as a user to get a "silent failure" - in our case, not alerting the user that something went wrong, but their output plot isn't correct/there at all

3D scatter viewer

Point size scaling

• Same question as in the 2D case

Image

• Currently the exported image seems to only contain the first image layer, which is obviously a problem

  • The heatmap idea that you have here is clever, but I'm wondering if it might be easier to use the plotly Image class. Per the docs here, the image class takes a PIL image as the source property. Using edit 2 from here, I was able to convert the viewer figure into a PIL image. (You can get the figure from the viewer as fig = viewer.figure). Note that PIL is already a dependency.

• We should have some sort of warning message if the selected colormap isn't supported by plotly

• The contrast and bias settings in glue aren't used in the exporter

Histogram

• The difference between the plotly and glue binning conventions is unfortunate. I'm not sure yet what the best way to handle that is

Let me know if you have any questions about any of these. Also, if I remember correctly you've got some other stuff lined up for the rest of the summer. If you don't think you have more time to work on this, just let me know and I can take things from here.

[Carifio24]

The updates here were repackaged into #18, #19, #20, #28, so this can be closed.