normalize_input: Normalize input data matrix

In Rtsne: T-Distributed Stochastic Neighbor Embedding using a Barnes-Hut Implementation

Normalize input data matrix

Description

Mean centers each column of an input data matrix so that it has a mean of zero. Scales the entire matrix so that the largest absolute of the centered matrix is equal to unity.

Usage

normalize_input(X)

Arguments

X	matrix; Input data matrix with rows as observations and columns as variables/dimensions.

Details

Normalization avoids numerical problems when the coordinates (and thus the distances between observations) are very large. Directly computing distances on this scale may lead to underflow when computing the probabilities in the t-SNE algorithm. Rescaling the input values mitigates these problems to some extent.

Value

A numeric matrix of the same dimensions as X but centred by column and scaled to have a maximum deviation of 1.

Author(s)

Aaron Lun

Examples

iris_unique <- unique(iris) # Remove duplicates
iris_matrix <- as.matrix(iris_unique[,1:4])
X <- normalize_input(iris_matrix)
colMeans(X)
range(X)

Rtsne documentation built on May 29, 2024, 1:30 a.m.