Overview

As discussed briefly in #3194 with @philippjfr , this is an issue to catalog some of the ways that the plotly backend could be enhanced by taking advantage of existing plotly.py/plotly.js features. Feedback appreciated 🙂

Established Elements

The following elements are supported by matplotlib and/or bokeh and could also be supported with plotly.

• Text, Arrow, Labels (See https://plot.ly/python/text-and-annotations/)
• VLine, HLine, Box, Ellipse (See https://plot.ly/python/shapes/)
• Spikes (scatter line trace with gaps)
• Bars (See https://plot.ly/python/bar-charts/)
• Contours (with scatter lines traces if lines are precomputed)
• Area, Spread (See https://plot.ly/python/filled-area-plots/)
• RGB
• HSV
• Violin (This is a first-class trace with nice hover info, see https://plot.ly/python/violin/)
• Vectorfield (See https://plot.ly/python/quiver-plots/)
• Bivariate (See https://plot.ly/python/2d-histogram-contour/)
• Scatter with polar projection (First class trace type with polar specific zoom and rotation, see https://plot.ly/python/polar-chart/). This would also support Curve, Area, and Spread with polar projection.
• Radialheatmap could be implemented with barpolar trace (See https://plot.ly/python/wind-rose-charts/)
• Sankey (There is a nice first-class Sankey trace that support node/link tooltips and drag to reorder nodes, see https://plot.ly/python/sankey-diagram/)
• Hextiles (Not built in, but it's been implemented from primitive shapes before, see https://plot.ly/~empet/13706/hexbin-plot/#/)
• Chord (Not built in, but it's been worked through before, see https://plot.ly/python/filled-chord-diagram/)
• Graph. Could be done with scatter for node/edge tooltips. The highlighting of connected edges on hover would require custom JS though. See https://plot.ly/python/network-graphs/#create-network-graph

Potential new elements

The following are potential new elements that I don't think are currently supported by matplotlib/bokeh (but if I'm wrong we can just bump them up to the previous section)

• Bar with polar projection: First class trace with zoom, rotation, etc. Also supports selection events.
• Parallel Coordinates: For multi-dimensional continuous datasets. WebGL accelerated, supports drag to reorder dimensions, and click-drag on axes to filter data.
• Parallel Categories: For multidimensional categorical datasets. Interactive drag to reorder categories and dimensions and supports click/hover events (See brushing examples in the docs)
• Carpet plot, Carpet scatter plot for representing 2 continuous independent variables and 1 continuous dependent variable. (also see https://en.wikipedia.org/wiki/Carpet_plot)
• Curve, Scatter, Points with ternary projection for 3 continuous dimensions constrained such that the 3-coordinates of each observation sum to 1.
• Dendrogram
• OHLC and Candlestick finance charts
• 3D Cone/Vectorfield: For representing a 3D vector field at discrete locations.
• 3D Streamtube for representing contunuous flow in 3D from particular starting points.

General Improvements / Fixes:

• Convert (Item)Table from figure factor to dedicated table trace type (https://plot.ly/python/table/). This trace type supports scrolling and can handle much larger tables.
• Move from plotly.tools.make_subplots to using layout.grid for laying out subplots. This
  way all trace types are supported (the make_subplots approach only really supports 2D and 3D subplots, not tables, polar, etc.). Replaced by new subplot foundations in New subplot foundation for Plotly backend #3255.

Streams

Plotly.js emits the events needed to detect selecting, clicking, and hovering over points in a trace.
It also emits events in response to zoom and pan. It does not currently emit events for clicking
and hovering the plot background (only when clicking or hover on particular points).

So I think it would be possible to support the, Selection1D, Range(X/Y), and Bounds(X/Y) streams across the elements based on scatter, scatter polar, bar, bar polar, ternary, and heatmap traces.

• Selection1D
• Range(X/Y)
• Bounds(X/Y)

There might be a bit of a mismatch with the click/hover events. since plotly.js associates these with points and it looks like Holoviews wants positions.

Datashader

With the RangeXY stream and ability to display RGB images I think it should be possible to support datashader. Here's an example of wireing up datashader in plotly.py directly with FigureWidget callbacks https://plot.ly/python/change-callbacks-datashader/.

Performance

All of plotly's 3D trace types are WebGL accelerated. The 2D scatter trace (which would be used for the scatter, line, area, and spread elements) and scatter polar trace have both D3+SVG and WebGL implementations. The WebGL implementations support a million or so points and are pretty much feature complete so we would want to default to these, with the option of falling back to non-WebGL for generating static images in vector format.

Static image export

plotly.py recently got (standalone offline) static image export support, see https://plot.ly/python/static-image-export/. This is electron-based and so it was possible to package everything using conda without a dependency on a system web browser or webdriver. It supports raster (png, jpg, webp) and vector (svg, pdf, eps) image formats, and all formats support figures with multiple subplots. Note: All plot types can be output to vector formats, but stuff that's rendered with webgl will be embedded in raster form.

LaTeX

Plotly supports rendering labels and annotations with LaTeX using MathJax. This is now supported in the Jupyter notebook, JupyterLab, in standalone html files, and in static image export (raster and vector). So this would make the plotly backend capable of supporting both interactive exploration and the generation of publication quality vector images with LaTeX typesetting.

Themes

Plotly recently got theming support (See https://medium.com/@plotlygraphs/introducing-plotly-py-theming-b644109ac9c7). It's possible to configure a theme to be used by default, so it might not be necessary for Holoviews to be aware of them. But it might also be nice to be able to configure the current theme as an option.