Hello!
Yes, but if you remember the XOR example, it is not like a linear interpolator: it allows non-linearities, complex shapes and contradictions.
More importantly, I can add points when I’m not happy with the training and continue the exploration, very much in line with Rebecca’s small data exploration paradigm.
The idea is to map 2 spaces that each make sense in themselves. In other word, the spaces UMAP has given us to play with (us = @balintlaczko @tutschku @tedmoore @jamesbradbury and you and myself) are losely interactively and playfully mapped to each other.
More soon-ish (I’m on the road this week but hotel rooms are there for that, right?)