search

so-wise / weddell_gyre_clusters

Unsupervised classification of Weddell Gyre profiles
MIT License
2 stars 1 forks source link

Quantify how different the classes are from eahc other #74

Closed DaniJonesOcean closed 2 years ago

DaniJonesOcean commented 2 years ago

Another good question from Erin Thomas at OSM22:

Is there a way to quantify how similar/different the classes are from each other? Since the classification is multivariate, is there a way to determine if the separation is dominated by salinity or temperature?

I suggested that we could look at the tSNE diagram. Instead of the label as the color, we could plot Tmax, surface properties, etc.

We could also do the "shadow score" and other standard metrics of separation.

DaniJonesOcean commented 2 years ago

The tSNE plots are good enough for this. Because we don't necessarily expect them to be extremely different, not by the usual metrics. (Our dataset is highly correlated in space and time! That's just ocean-atmosphere data for you.)