Open rvosa opened 7 years ago
Response from Yuanhao:
I used to work on this topic for a while and learned that this technique mostly works on low-dimensional (usually in 2D or 3D) visualization for high-dimensional features. It has not been clear how to apply it on dimensionality reduction. Probably there have been related works published. This algorithm is very powerful but probably you would like to look up improved version of this method to make it suitable in our work. Another possible replacements are probably non-linear dimensionality reduction http://scikit-learn.org/stable/modules/manifold.html.
In addition to, or in replacement of, PCAs, we might experiment with t-SNE: https://github.com/cemoody/topicsne