search

josephsdavid / N2D

Library implementation of https://arxiv.org/abs/1908.05968v5
https://n2d.readthedocs.io/en/latest/
MIT License
23 stars 6 forks source link

Fixing the assignment of the embedded manifold (hle). #2

Closed rymc closed 5 years ago

rymc commented 5 years ago

Previously hle was being set to the autoencoder hidden layer, rather than the UMAP manifold. The only impact of this was in terms of visualization, as the objects reference to hle was only used for visualization - for clustering the local (correct) hle reference was used.

josephsdavid commented 5 years ago

Makes sense, merging!

rymc commented 5 years ago

Thanks for merging. It may be a good idea to update the README with the new visualizations too!

Edit: I see that the images come directly from the repo and there is no need to update the README. My bad.

josephsdavid commented 5 years ago

Will do! Currently updating tensorflow which has turned into rebuilding it, so ill fix it when that is finished, and get the proper code on Pypi :)

josephsdavid commented 5 years ago

Actually ill rebuild in the morning this is much more enjoyable

rymc commented 5 years ago

I'm not sure what happened but it looks like this commit [1] brought in the old incorrect visualisations rather than the visualisations from this pull request.

[1] https://github.com/josephsdavid/N2D-OOP/commit/6205cd71db9f023f1f359218ab91b569c67b5793

josephsdavid commented 5 years ago

yep, just noticed my stupid mistake

josephsdavid commented 5 years ago

Oh man yeah this is worlds better!