Explore alternative frameworks to speed up scatter plots

[astrofrog]

We should explore the use of VisPy and PyQtGraph to make scatter plots. This issue is to keep track of findings related to this.

• Vispy: currently the axes are not very esthetically pleasing, though they were only implemented recently. I opened an issue here to ask if we can customize the appearance: Feature request: more control over axes appearance vispy/vispy#1057
• PyQtGraph: fast for line plots, but super slow for scatter plots. I've opened an issue here to find out why: Scatter plots are very slow pyqtgraph/pyqtgraph#217

[astrofrog]

cc @nmearl @stscieisenhamer @ibusko, in case you have any insight relating to PyQtGraph and performance

[astrofrog]

Just for info, PyQtGraph, while fast for lines, is not going to work for scatter plots: pyqtgraph/pyqtgraph#217

[nmearl]

Indeed. While fast scatter plots are possible in PyQtGraph using OpenGL (specifically, the GLScatterPlotItem widget), the author has indicated that the pipeline will be replaced with VisPy.

The 3D graphics system in pyqtgraph is composed of a view widget and several graphics items (all subclasses of GLGraphicsItem) which can be added to a view widget.

Note 1: pyqtgraph.opengl is based on the deprecated OpenGL fixed-function pipeline. Although it is currently a functioning system, it is likely to be superceded in the future by VisPy.

Note 2: use of this system requires python-opengl bindings. Linux users should install the python-opengl packages from their distribution. Windows/OSX users can download from http://pyopengl.sourceforge.net.

[nmearl]

As a side note, you can get some pretty good performance gains by enabling down sampling and view clipping.

 import pyqtgraph as pg
 import numpy as np

 pw = pg.plot(title="Testing")
 pw.setDownsampling(ds=True, auto=True, mode='peak')
 pw.setClipToView(True)
 pw.plot(np.arange(1e5), np.random.sample(1e5), symbol='o')

[astrofrog]

Also requested in #403, and the official issue to keep track of this is glue-viz/glue-vispy-viewers#138.

[astrofrog]

This isn't so much of a problem anymore now with the scatter density plots, so closing