Visualization Plots

[DreRnc]

Here I will post information or good articles I have gathered on strategies for latent space visualization.

[DreRnc]

Umap vs TSNE: https://pair-code.github.io/understanding-umap/

We have great plots and good explanations: great for learning.

[DreRnc]

Umap things to remember:

1. Hyperparameters really matter
2. Cluster sizes in a UMAP plot mean nothing
3. Distances between clusters might not mean anything
4. Random noise doesn’t always look random (especially at low values of n_neighbours)
5. You may need more than one plot

Points 2 and 3 are because of the fact that a high dimensional graph with local connections is being constructed, there are no distance considerations when projecting.

[DreRnc]

Leaving here some more material I'll be looking into:

How to (mis)read a TSNE: https://distill.pub/2016/misread-tsne/