votes.tsne: Votes t-sne

In pmartinezarbizu/RFtools: Miscellaneous Tools For Random Forest Models

Description

Calculate t-distributed stochastic neighbor embedding on randomForest votes.

Usage

votes.tsne(x, method = "none", perplex.factor = 0.15, max_iter = 1000)

Arguments

x	A randomForest object
method	One of the transformations offered by decostand from package vegan. Default is 'none'
perplex.factor	Multiply the number of observations by the perplex.factor to get the perpelexity value passed to t-sne
max_iter	Maximum number of iterations: parameter passed to t-sne

Details

VotesPlot will use the votes matrix of a randomForest object to calculate the probabiliy that two observations belong to the same cluster using t-distributed stochastic neighbor embedding from package tsne. An t-sne ordination graph is plotted to trace the results after every 100 iterations until max_iter is reached. Main control parameter of t-sne analysis is the perplexity value. Here we use a default perplexity of 15 Contrasting plot symbols and colors are assigned automatically, but can be adjusted manually. Although not recommended here, it might be interesting to transform the votes table prior to t-sne. The argument method can be used to pass the method to decostand. Hellinger transformation might be meaningful. Consider to increase max_iter if there is no stabilization of the error after the default 1000 iterations.

returns an object of type votes_tsne A plot method is available

Author(s)

Pedro Martinez Arbizu

See Also

plot.votes_tsne

Examples

data(iris)
rf <- randomForest(Species ~.,data = iris)
vt <- votes.tsne(rf)

plot(vt,margin=1,y.low=5)

pmartinezarbizu/RFtools documentation built on March 10, 2021, 12:11 p.m.