What does it mean when euclidean distance gives the best separation using t-sne (stochastic neighbor embedding function)?

Hello,

I have data which i have used PCA and t-SNE to cluster. Why does euclidean give me the best seperation?

Thanks

[the cyclist]

This is not really a MATLAB question, and is more of a theoretical statistics question.

I did some searching, and did not really find anything about why one should choose one distance metric over another.

[Neo]

@the cyclist

in the function description in matlab it gives comparison for each distances (attached). Does matlab not justify which one is better? or is it just based on the method itself you mean? Thank you

[the cyclist]

There is no single "best" choice of distance metric (as far as I can tell), and it is not the job of statistical software to decide which distance metric is better for your data. MATLAB provides options, and sets a default option. Euclidean definitely seems to be the most commonly used metric, so it is sensible as a default.

[Neo]

@the cyclist okay thanks