Need for a combination of parallel coordinates and parallel categories

Dash Python
1

I am trying to visualize the influence of some discrete hyperparameters on some continuous metrics.
I think there is the need for a widget which allows both discrete and continous hyperparamters. So far the best I can get is to use a parallel categories plot and discretize the continuous values into some bin (or approximate them to the closest integer to reduce the number of classes).

Here is an overview of what it is available, to the best of my knowledge, with the current APIs.

  • The first plot is the parallel categories plot that I am talking about, rounded to the closest integer.
  • The second plot shows only the continuous metrics
  • The third plot is an attempt to plot both continuous and categorical values with a parallel coordinates plot, the result is bad since one can’t distinguish proportions for the categorical coordinates

Screen Shot 2020-08-13 at 16.04.43
Screen Shot 2020-08-13 at 16.04.431595×795 287 KB

Screen Shot 2020-08-13 at 16.04.27
Screen Shot 2020-08-13 at 16.04.271597×782 138 KB

Screen Shot 2020-08-13 at 16.04.53
Screen Shot 2020-08-13 at 16.04.531574×784 152 KB

The first plot is using parallel_categories, the second parallel_coordinates, the third Parcoords.

2

I have just come across this issue too. It would be ideal for both the parallel_categories and parallel_coordinates to be able to handle the opposite style of data. I’ve also resorted to binning the numerical data into categories - which works fine, but you do lose something.