tsne

t-Distributed Stochastic Neighbor Embedding

Syntax

Y = tsne(X)

Y = tsne(X,Name,Value)

[Y,loss] = tsne(___)

Description

Y = tsne(X) returns a matrix of two-dimensional embeddings of the high-dimensional rows of X.

Y = tsne(X,Name,Value) modifies the embeddings using options specified by one or more name-value pair arguments.

[Y,loss] = tsne(___), for any input arguments, also returns the Kullback-Leibler divergence between the joint distributions that model the data X and the embedding Y.

Examples

Visualize Fisher Iris Data

The Fisher iris data set has four-dimensional measurements of irises, and corresponding classification into species. Visualize this data by reducing the dimension using tsne.

load fisheriris
rng default % for reproducibility
Y = tsne(meas);
gscatter(Y(:,1),Y(:,2),species)

Compare Distance Metrics

Use various distance metrics to try to obtain a better separation between species in the Fisher iris data.

load fisheriris

rng('default') % for reproducibility
Y = tsne(meas,'Algorithm','exact','Distance','mahalanobis');
subplot(2,2,1)
gscatter(Y(:,1),Y(:,2),species)
title('Mahalanobis')

rng('default') % for fair comparison
Y = tsne(meas,'Algorithm','exact','Distance','cosine');
subplot(2,2,2)
gscatter(Y(:,1),Y(:,2),species)
title('Cosine')

rng('default') % for fair comparison
Y = tsne(meas,'Algorithm','exact','Distance','chebychev');
subplot(2,2,3)
gscatter(Y(:,1),Y(:,2),species)
title('Chebychev')

rng('default') % for fair comparison
Y = tsne(meas,'Algorithm','exact','Distance','euclidean');
subplot(2,2,4)
gscatter(Y(:,1),Y(:,2),species)
title('Euclidean')

In this case, the cosine, Chebychev, and Euclidean distance metrics give reasonably good separation of clusters. But the Mahalanobis distance metric does not give a good separation.

Plot Results with NaN Input Data

tsne removes input data rows that contain any NaN entries. Therefore, you must remove any such rows from your classification data before plotting.

For example, change a few random entries in the Fisher iris data to NaN.

load fisheriris
rng default % for reproducibility
meas(rand(size(meas)) < 0.05) = NaN;

Embed the four-dimensional data into two dimensions using tsne.

Y = tsne(meas,'Algorithm','exact');

Warning: Rows with NaN missing values in X or 'InitialY' values are removed.

Determine how many rows were eliminated from the embedding.

length(species)-length(Y)

ans = 
22

Prepare to plot the result by locating the rows of meas that have no NaN values.

goodrows = not(any(isnan(meas),2));

Plot the results using only the rows of species that correspond to rows of meas with no NaN values.

gscatter(Y(:,1),Y(:,2),species(goodrows))

Compare t-SNE Loss

Find both 2-D and 3-D embeddings of the Fisher iris data, and compare the loss for each embedding. It is likely that the loss is lower for a 3-D embedding, because this embedding has more freedom to match the original data.

load fisheriris
rng default % for reproducibility
[Y,loss] = tsne(meas,'Algorithm','exact');
rng default % for fair comparison
[Y2,loss2] = tsne(meas,'Algorithm','exact','NumDimensions',3);
fprintf('2-D embedding has loss %g, and 3-D embedding has loss %g.\n',loss,loss2)

2-D embedding has loss 0.1255, and 3-D embedding has loss 0.0980872.

As expected, the 3-D embedding has lower loss.

View the embeddings. Use RGB colors [1 0 0], [0 1 0], and [0 0 1].

For the 3-D plot, convert the species to numeric values using the categorical command, then convert the numeric values to RGB colors using the sparse function as follows. If v is a vector of positive integers 1, 2, or 3, corresponding to the species data, then the command

sparse(1:numel(v),v,ones(size(v)))

is a sparse matrix whose rows are the RGB colors of the species.

gscatter(Y(:,1),Y(:,2),species,eye(3))
title('2-D Embedding')

figure
v = double(categorical(species));
c = full(sparse(1:numel(v),v,ones(size(v)),numel(v),3));
scatter3(Y2(:,1),Y2(:,2),Y2(:,3),15,c,'filled')
title('3-D Embedding')
view(-50,8)

Input Arguments

X — Data points

n-by-m matrix

Data points, specified as an n-by-m matrix, where each row is one m-dimensional point.

tsne removes rows of X that contain any NaN values before creating an embedding. See Plot Results with NaN Input Data.

Data Types: single | double

Name-Value Arguments

Specify optional pairs of arguments as Name1=Value1,...,NameN=ValueN, where Name is the argument name and Value is the corresponding value. Name-value arguments must appear after other arguments, but the order of the pairs does not matter.

Before R2021a, use commas to separate each name and value, and enclose Name in quotes.

Example: Y = tsne(X,'Algorithm','Exact','NumPCAComponents',50)

Algorithm Control

Algorithm — tsne algorithm

'barneshut' (default) | 'exact'

tsne algorithm, specified as 'barneshut' or 'exact'. The 'exact' algorithm optimizes the Kullback-Leibler divergence of distributions between the original space and the embedded space. The 'barneshut' algorithm performs an approximate optimization that is faster and uses less memory when the number of data rows is large.

Note

For the 'barneshut' algorithm, tsne uses knnsearch to find the nearest neighbors.

Example: 'exact'

CacheSize — Size of Gram matrix in megabytes

1e3 (default) | positive scalar | "maximal"

Size of the Gram matrix in megabytes, specified as a positive scalar or "maximal". The tsne function can use CacheSize only when the Distance name-value argument begins with fast.

If you set CacheSize to "maximal", tsne tries to allocate enough memory for an entire intermediate matrix whose size is M-by-M, where M is the number of rows of the input data X. The cache size does not have to be large enough for an entire intermediate matrix, but must be at least large enough to hold an M-by-1 vector. Otherwise, tsne uses the standard algorithm for computing Euclidean distances.

If the value of the Distance argument begins with fast, and the value of CacheSize is too large or "maximal", tsne might try to allocate a Gram matrix that exceeds the available memory. In this case, MATLAB^® issues an error.

Example: CacheSize="maximal"

Data Types: double | char | string

Distance — Distance metric

'euclidean' (default) | 'seuclidean' | 'fasteuclidean' | 'fastseuclidean' | 'cityblock' | 'chebychev' | 'minkowski' | 'mahalanobis' | 'cosine' | 'correlation' | 'spearman' | 'hamming' | 'jaccard' | function handle

Distance metric, specified as one of the following:

• 'euclidean' — Euclidean distance.

• 'seuclidean' — Standardized Euclidean distance. Each coordinate difference between the rows in X and the query matrix is scaled by dividing by the corresponding element of the standard deviation computed from S = std(X,'omitnan').

• 'fasteuclidean' — Euclidean distance computed by using an alternative algorithm that saves time when the number of predictors is at least 10. In some cases, this faster algorithm can reduce accuracy. Algorithms starting with 'fast' do not support sparse data. For details, see Algorithms.

• 'fastseuclidean' — Standardized Euclidean distance computed by using an alternative algorithm that saves time when the number of predictors is at least 10. In some cases, this faster algorithm can reduce accuracy. Algorithms starting with 'fast' do not support sparse data. For details, see Algorithms.

• 'cityblock' — City block distance.

• 'chebychev' — Chebychev distance, which is the maximum coordinate difference.

• 'minkowski' — Minkowski distance with exponent 2. This distance is the same as the Euclidean distance.

• 'mahalanobis' — Mahalanobis distance, computed using the positive definite covariance matrix cov(X,'omitrows').

• 'cosine' — One minus the cosine of the included angle between observations (treated as vectors).

• 'correlation' — One minus the sample linear correlation between observations (treated as sequences of values).

• 'spearman' — One minus the sample Spearman's rank correlation between observations (treated as sequences of values).

• 'hamming' — Hamming distance, which is the percentage of coordinates that differ.

• 'jaccard' — One minus the Jaccard coefficient, which is the percentage of nonzero coordinates that differ.

• Custom distance function — A distance function specified using @ (for example, @distfun). For details, see More About.

In all cases, tsne uses squared pairwise distances to calculate the Gaussian kernel in the joint distribution of X.

Example: 'mahalanobis'

Data Types: char | string | function_handle

Exaggeration — Size of natural clusters in data

4 (default) | scalar value 1 or greater

Size of natural clusters in data, specified as a scalar value 1 or greater.

A large exaggeration makes tsne learn larger joint probabilities of Y and creates relatively more space between clusters in Y. tsne uses exaggeration in the first 99 optimization iterations.

If the value of Kullback-Leibler divergence increases in the early stage of the optimization, try reducing the exaggeration. See tsne Settings.

Example: 10

Data Types: single | double

NumDimensions — Dimension of the output Y

2 (default) | positive integer

Dimension of the output Y, specified as a positive integer. Generally, set NumDimensions to 2 or 3.

Example: 3

Data Types: single | double

NumPCAComponents — PCA dimension reduction

0 (default) | nonnegative integer

PCA dimension reduction, specified as a nonnegative integer. Before tsne embeds the high-dimensional data, it first reduces the dimensionality of the data to NumPCAComponents using the pca function. When NumPCAComponents is 0, tsne does not use PCA.

Example: 50

Data Types: single | double

Perplexity — Effective number of local neighbors of each point

30 (default) | positive scalar

Effective number of local neighbors of each point, specified as a positive scalar. See t-SNE Algorithm.

Larger perplexity causes tsne to use more points as nearest neighbors. Use a larger value of Perplexity for a large dataset. Typical Perplexity values are from 5 to 50. In the Barnes-Hut algorithm, tsne uses min(3*Perplexity,N-1) as the number of nearest neighbors. See tsne Settings.

Example: 10

Data Types: single | double

Standardize — Flag to normalize input data

false (default) | true

Flag to normalize input data, specified as false or true. When the value is true, tsne centers and scales each column of X by first subtracting its mean, and then dividing by its standard deviation.

When features in X are on different scales, set 'Standardize' to true. The learning process is based on nearest neighbors, so features with large scales can override the contribution of features with small scales.

Example: true

Data Types: logical

Optimization Control

InitialY — Initial embedded points

1e-4*randn(N,NumDimensions) (default) | n-by-NumDimensions real matrix

Initial embedded points, specified as an n-by-NumDimensions real matrix, where n is the number of rows of X. The tsne optimization algorithm uses these points as initial values.

Data Types: single | double

LearnRate — Learning rate for optimization process

500 (default) | positive scalar

Learning rate for optimization process, specified as a positive scalar. Typically, set values from 100 through 1000.

When LearnRate is too small, tsne can converge to a poor local minimum. When LearnRate is too large, the optimization can initially have the Kullback-Leibler divergence increase rather than decrease. See tsne Settings.

Example: 1000

Data Types: single | double

NumPrint — Iterative display frequency

20 (default) | positive integer

Iterative display frequency, specified as a positive integer. When the Verbose name-value pair is not 0, tsne returns iterative display after every NumPrint iterations. If the Options name-value pair contains a nonempty 'OutputFcn' entry, then output functions run after every NumPrint iterations.

Example: 20

Data Types: single | double

Options — Optimization options

structure containing the fields 'MaxIter', 'OutputFcn', and 'TolFun'

Optimization options, specified as a structure containing the fields 'MaxIter', 'OutputFcn', and 'TolFun'. Create 'Options' using statset or struct.

• 'MaxIter' — Positive integer specifying the maximum number of optimization iterations. Default: 1000.

• 'OutputFcn' — Function handle or cell array of function handles specifying one or more functions to call after every NumPrint optimization iterations. For syntax details, see t-SNE Output Function. Default: [].

• 'TolFun' — Stopping criterion for the optimization. The optimization exits when the norm of the gradient of the Kullback-Leibler divergence is less than 'TolFun'. Default: 1e-10.

Example: options = statset('MaxIter',500)

Data Types: struct

Theta — Barnes-Hut tradeoff parameter

0.5 (default) | scalar from 0 through 1

Barnes-Hut tradeoff parameter, specified as a scalar from 0 through 1. Higher values give a faster but less accurate optimization. Applies only when Algorithm is 'barneshut'.

Example: 0.1

Data Types: single | double

Verbose — Iterative display

0 (default) | 1 | 2

Iterative display, specified as 0, 1, or 2. When Verbose is not 0, tsne prints a summary table of the Kullback-Leibler divergence and the norm of its gradient every NumPrint iterations.

When Verbose is 2, tsne also prints the variances of Gaussian kernels. tsne uses these kernels in its computation of the joint probability of X. If you see a large difference in the scales of the minimum and maximum variances, you can sometimes get more suitable results by rescaling X.

Example: 2

Data Types: single | double

Output Arguments

Y — Embedded points

n-by-NumDimensions matrix

Embedded points, returned as an n-by-NumDimensions matrix. Each row represents one embedded point. n is the number of rows of data X that do not contain any NaN entries. See Plot Results with NaN Input Data.

loss — Kullback-Leibler divergence

nonnegative scalar

Kullback-Leibler divergence between modeled input and output distributions, returned as a nonnegative scalar. For details, see t-SNE Algorithm.

More About

Custom Distance Function

The syntax of a custom distance function is as follows.

function D2 = distfun(ZI,ZJ)

tsne passes ZI and ZJ to your function, and your function computes the distance.

• ZI is a 1-by-n vector containing a single row from X.

• ZJ is an m-by-n matrix containing multiple rows of X.

Your function returns D2, which is an m-by-1 vector of distances. The jth element of D2 is the distance between the observations ZI and ZJ(j,:).

Tip

If your data are not sparse, then usually the built-in distance functions are faster than a function handle.

Algorithms

tsne constructs a set of embedded points in a low-dimensional space whose relative similarities mimic those of the original high-dimensional points. The embedded points show the clustering in the original data.

Roughly, the algorithm models the original points as coming from a Gaussian distribution, and the embedded points as coming from a Student’s t distribution. The algorithm tries to minimize the Kullback-Leibler divergence between these two distributions by moving the embedded points.

For details, see t-SNE.

Fast Euclidean Distance Algorithm

The values of the Distance argument that begin fast (such as "fasteuclidean" and "fastseuclidean") calculate Euclidean distances using an algorithm that uses extra memory to save computational time. This algorithm is named "Euclidean Distance Matrix Trick" in Albanie [1] and elsewhere. Internal testing shows that this algorithm saves time when the number of predictors is at least 10. Algorithms starting with fast do not support sparse data.

To find the matrix D of distances between all the points x_i and x_j, where each x_i has n variables, the algorithm computes distance using the final line in the following equations:

$$\begin{array}{c}{D}_{i,j}^2=\Vert x_i-x_j{\Vert }^2\\ ={(}^{x_i}(x_i-x_j)\\ =\Vert x_i{\Vert }^2-2x_i^T{x}_j+\Vert x_j{\Vert }^2.\end{array}$$

The matrix $$x_i^T{x}_j$$ in the last line of the equations is called the Gram matrix. Computing the set of squared distances is faster, but slightly less numerically stable, when you compute and use the Gram matrix instead of computing the squared distances by squaring and summing. For more details, see Albanie [1].

To store the Gram matrix, the software uses a cache with the default size of 1e3 megabytes. You can set the cache size using the CacheSize name-value argument. If the value of CacheSize is too large or "maximal", then the software might try to allocate a Gram matrix that exceeds the available memory. In this case, the software issues an error.

References

[1] Albanie, Samuel. Euclidean Distance Matrix Trick. June, 2019. Available at https://samuelalbanie.com/files/Euclidean_distance_trick.pdf.

Version History

Introduced in R2017a

R2023a: Fast Euclidean distance using a cache

The 'fasteuclidean' and 'fastseuclidean' distance metrics accelerate the computation of Euclidean distances by using a cache and a different algorithm (see Algorithms). Set the size of the cache using the CacheSize name-value argument.