Discover-1912

Table of Contents

    New features

    [Back end]

    • Implemented the full functionality of NSGA-II, particularly the way in which the best designs from the previous and current generations are combined to create the sample pool for the next generation. This implementation no longer requires a separate 'elites' parameter which has been removed from the interface.
    • API tweaks to accommodate updated component behavior.

    [Front end]

    • Many design tweaks including the icons, buttons, and tooltip design.
    • Tweaking of animations to optimize responsiveness with large optimization runs.
    • Faster scrolling in scatter plot.
    • 'Hidden' designs during the 'isolate' commands are now more transparent with no strokes to make the isolated designs more visible.
    • 'Hidden' designs during the 'isolate' commands now do not trigger a mouseover to make it easier to select the isolated designs.

    [GH library]

    • Added an 'N' input to the input components so you can now supply an optional name to each input. Although this won't reveal itself in the interface since currently inputs are not selectable in the scatter plot (this is on the backlog), but can be useful when parsing the columns of the results.tsv data file.
    • The current Discover version is now displayed as a Message below the main Discover component to make it easier to ensure that the version of the components and front-end match.

    Bugs fixed 

    [Back end]

    • N/A

    [Front end]

    • Non-feasible designs are no longer considered during the 'isolate optimal' calculations. Only feasible designs are now ever considered optimal.
    • Auto-scaling added when switching between axes to ensure points stay within scatter plot.

    [GH library]

    • Each component generates a unique ID which it uses to register itself with the Discover server. Previously these ID's were re-generated each time the file was opened, causing issues when trying to regenerate designs from previous optimization runs. Now the ID's are sticky and saved with the file, so you should be able to reload a file that was previously optimized and regenerate designs from a previous optimization in Discover.

    Notes

    • Similar to the output components in the last update, the data structure changes necessary to make the ID's of the input component sticky will cause any input components in models using previous versions of the GH library to break. These components will need to be replaced by their versions from the new library.
    • Discover was tested with Python 3.8, which is now the newest version. Due to incompatibilities with the Flask and Werkzeug libraries this version is not supported at this time. Please continue to use Python 3.7.4 with Discover.


    Version: 1
    Updated on: Mar 28, 2021