BF/OA: ND visualization of stream data, points, hypercuboids, clusters with spider graphs

[detlefarend]

Description/Motivation
N-dimensional data like stream instances, hyper-points and hypercuboids can be visualized using spider graphs. Each dimension is mapped to an axis of the graph. A hyper-point/stream instance has a single position at each axis and can be visualized as a number of lines connecting these positions axis by axis. A hypercuboid has min/max-borders on each dimension/axis and can be visualized accordingly.

Task list

• Class bf.plot.Plottable
  Alternatively to a matplotlib 2D/3D/ND plot, a new spider plot systematics shall be established. For this, a host plot with further high-level helper methods shall be implemented.
  • New view ND_SPIDER
  • New method _init_nd_spider()
    This method initializes all necessary data structures of the spider plot systematics.
  • New custom method _update_nd_spider()
• Class bf.math.geometry.Point
  • Method _update_plot_nd_spider()
• Class bf.math.geometry.Hypercuboid
  • Method _update_plot_nd_spider()
• Class bf.streams.tasks.StreamTask
  • Method _update_plot_nd_spider()
• Class oa.streams.tasks.clusteranalyzers.clusters.Centroid
  • Method _update_plot_nd_spider()

Cross references
https://stackoverflow.com/questions/52910187/how-to-make-a-polygon-radar-spider-chart-in-python
https://matplotlib.org/stable/gallery/specialty_plots/radar_chart.html

[detlefarend]

@steveyuwono @syamrajsatheesh @laxmikantbaheti What do you think of this alternative approach for nd visualization?

[syamrajsatheesh]

It is a nice way to visualize ND plots. But what about data with a lot of features or dimensions. For example, when there are 20 features, it would be hard to read the plot. Upto 8 features is OK I think.

[detlefarend]

It is a nice way to visualize ND plots. But what about data with a lot of features or dimensions. For example, when there are 20 features, it would be hard to read the plot. Upto 8 features is OK I think.

Hi Syam, you are totally right. This visualization has limits as well. On the other hand, our current nd visualization is also hard to read for a higher number of dimensions and it doesn't cover cluster visualization. Finally, 8 dims are better than nothing (cluster vis).