Open lgatto opened 7 years ago
Did a first test run on 5 Datasets and will run over some night the next days the same procedure over all datasets (or directly on a server).
The TSNE: The perplexity fixed to 35 and the iterations indicated on top of each plot.
As comparrison the PCA of the same data.
It is important in such cases to be systematic. Best approach would be to do this analysis for each dataset separately, and for a range of perplexities and steps, as the number of step to converge will depend on the perplexity itself.
To set reasonable parameters for the t-SNE visualisation in the online app, we need to assess the hyper-paramters on all (or many) dataset. An example of how to do this is shown in the vignette. In a nutshell, one needs to iterate over a range of
steps
andperplexity
values and produce the grid of PCA plots.@Kohze - could you create an
Rmd
file with this analysis insci-reports/rsne.Rmd
and update this issue once it is ready. Thanks!