HSNE embedding should have area of influence as 3rd dimensions (in addition to x and y coordinate)

[thoellt]

It should be possible to visualize the area of influence, e.g., as size in a scatterplot, to give an impression on how important landmarks are.

Additionally, as a 4th dimension the number of represented points in the point-based hierarchy would be useful.

[JulianThijssen]

What do you exactly mean by the area of influence here? Is that not similar to the number of points represented?

[thoellt]

Sorry, area of influence is probably a confusing term here. it should be the landmark weight based on the Aoi, probabilisitic hierarchy. The weibght is the fractional version of the number of points represented. In the point-based hierarchy this is replaced, such that every point on a level is represented by exactly one landmark on the next higher scale. As such you can just count the number of points every landmark represents. In the actual HSNE hierarchy a landmark represents landmarks on the lower level with a certain probability. What we have used in the past to indicate the importance/coverage of a landmark is the sum of those probabilities. It can be retrieved from the hdi::dr::HierarchicalSNE<float, hsneMatrix_t> by hierarchy()[scale]._landmark_weight).