hello all,
I use TSNE dimentionality reduction algorithme to reduce my data to 2D data.
Now, I need to integrate a new data to my mapped 2D data without rexcute the tSNE algorithme for the hole data.
Is there any technique or ameliorate version for that ?
Is there any matlab code ?
Thank you.
Best regards
Hi Hadjer,
According Laurens van der Maaten, the TSNE author ("https://lvdmaaten.github.io/tsne/"):
"A potential approach to deal with this would be to train a multivariate regressor to predict the map location from the input data."
Kind regards,
Olivier PARISOT
Hi Hadjer,
According Laurens van der Maaten, the TSNE author ("https://lvdmaaten.github.io/tsne/"):
"A potential approach to deal with this would be to train a multivariate regressor to predict the map location from the input data."
Kind regards,
Olivier PARISOT
I am also looking for something like that. I konw that there was attempts to add transform function to t_sne.py in scikit-learn python library. You can investigete on the https://github.com/scikit-learn/scikit-learn/pull/4025
This feature was added but later removed from main branch
https://github.com/scikit-learn/scikit-learn/commit/fe49161fc7ee478dd035c2d0fc3a0a527e9d4ec0
You ask for matlab code, but it will be easy to call Python script from Matlab or write wrapper function.
I did not play around with this transform function yet, but maybe you will find it usefull or you will implement it in matlab yoursefl.
Have a look at the paper
Learning a Parametric Embedding by Preserving Local Structure
(also by Laurens van der Maaten). This paper addresses the problem how new data points may be included in the soring.
Kai
- Badges
- Science topic