Implement Jupyter notebook magicks and functions for visualization

[popescu-v]

Questions/Ideas

• Local visualization: Check that the binary exists (are there environment variables set for this?)
• Either open visualization in a new tab or as a IFrame in the notebook (both? two magicks)
• Support for visualization web service ? Different magick ?
• Expose functions and magicks
• Implement helper function to plot with matplotlib lift and REC curves
• The JS component currently does not load with no server
  • We can make a prototype by running a simple HTTP server that serves the page
• Talk with LAG about potential fixes of the kh-viz component so to not have the aforementioned problem.
• Can we have a REST API ?

[popescu-v]

Update:

• S.Gouache and G.Collot refactored the JavaScript visualization project. As a by-product there is the khiops-visualization-widget, a (packaged) jupyter widget allows the visualization in a notebook cell.
  • This package enables the resolution of this issue without too much effort
  • Implement the interface with pyKhiops: A helper function and estimator methods that use them.
  • Should check if the dependency is installed and no-op if it is not found.
  • Package with conda (change the name).

[popescu-v]

Goal (at the khiops-python level): implement a method on the estimators, show_report, which:

• checks whether pkvisualization.visualize is available and runs without raising any exception; if so, then this means we are in a JuPyter notebook and the function stops;
• else, falls back to opening the standalone "heavy" Electron-based application:
  • via xdg-open for Linux,
  • via different solutions for Windows and MacOS (investigate which ones);
  • else (if this fails), then print warning to the user that this functionality is not supported.