Technical meaning of axes

[avielbr]

I’d love to understand on a technical level what the three axes of the visualizer represent. I get what they mean on an intuitive level, and obviously one can’t cleanly visualize more than 3D. Perhaps EI selects the top three most varying features and uses those to approximate overall separation?

My lack of knowledge in ML theory is showing Please educate me!

[aurel]

Hi @avielbr,

Here’s some good explanation from Jan about the UMAP technique used for dimension reduction in the features explorer:

Generating Features

@Terell_Dale, and in addition to @dhruvsheth if you have high dimensional data (like audio or images) it’s impractical to plot every single axis (and doesn’t really show you much). In that case we use UMAP as a dimensionality reduction algorithm to reduce the axes to 3 (“Visualization layer 1/2/3”). These axes on itself don’t have any meaning, but serve the same purpose as on low-dimensional data: seeing how well your dataset separates.

Aurelien

[avielbr]

Exactly what I was after. Thank you.