tsne_dist: Dimension reduction with t-SNE

In opisthokonta/tsnemicrobiota: t-distributed stochastic neighbor embedding for microbiota data

Description

t-Distributed Stochastic Neighbor Embedding (t-SNE).

Usage

tsne_dist(dist_, dimensions = 2, verbose = 0, rng_seed = NULL,
  control = list())

tsne(xdata, perplexity = 25, dimensions = 2, verbose = 0,
  rng_seed = NULL, control = list())

Arguments

dist_	A matrix or dist object woth pairwise dissimilarities. All elements must be positive and sum to 1.
dimensions	The number of dimensions of the t-SNE embedding. Default is 2, which is usually adequate.
verbose	The amount of output during the t-SNE estimation. Can be 0 (no output), 1 (some output, including a progress bar) and 2 (detailed output, mostly usefull for debugging purposes)
rng_seed	Provide a seed for generating intiial values for the coordiantes. The final t-SNE result is highly sensitive for starting values, so to get reproducible results you should provide a seed.
control	A list of control parameters. See 'Details'.
xdata	A matrix or data frame.
perplexity	The perplexity. Usually a number between 5 and 50. Default is 25.

Details

tsne_from_dist assumes that dist_ is a distance matrix interpretable as a nearest neighbor distribution. If you only have a distance matrix, you can use the dist_to_sne function to create a SNE dissimilarity matrix.

Function tsne provides a convenient interface to use t-SNE using raw data. Under the hood it calls the dist_sne and tsne_dist functions.

Value

A list of class 'tsne' with the following elements:

par:	A matrix with the t-SNE embedding.
trace:	A vector of the Kullback-Leibler divergences. If you used early exaggeration (default) you will see a large jump in the values at iteration 50.

Control

The control argument is a list that can supply any of the following components:

max_iter:	The number of iterations. Default is 1000.
step_size:	The gradient descent step size. Default is 100, but it can sometimes be usefull to set this to a lower value.
momentum_values:	A vector of values that determines the momentum. The default is c(0.5, 0.8), following the original t-SNE paper.
momentum_iter:	A vector of values that determines the iterations the momentum_values should be used. Default is c(250, 1000), following the original t-SNE paper.
early_exageration_iter:	Numerical. The number of iterations the early exageration trick should be used. Default 50.
early_exageration_factor:	Numerical. The exageration factor used in the early exacegration phase. Default is 4.

References

• L.J.P. van der Maaten and G.E. Hinton. Visualizing High-Dimensional Data Using t-SNE. Journal of Machine Learning Research 9(Nov):2579-2605, 2008.

• L.J.P. van der Maaten. The t-SNE FAQ https://lvdmaaten.github.io/tsne/

See Also

The tsne_phyloseq function provides a nice interface for using t-SNE with microbiota data.

Examples

xx <- make_swiss_roll(150)
tsne_res <- tsne(xx[,1:3])
plot(tsne_res$par, col=xx[,4], pch=16)

opisthokonta/tsnemicrobiota documentation built on June 1, 2019, 9:44 a.m.