qdtsne::Status< num_dim_, Index_, Float_ > Class Template Reference

Status of the t-SNE iterations. More...

#include <Status.hpp>

Public Member Functions
int	iteration () const
int	max_iterations () const
Index_	num_observations () const
void	run (Float_ *const Y, const int limit)
void	run (Float_ *const Y)
Float_	cost (const Float_ *const Y)

Detailed Description

template<std::size_t num_dim_, typename Index_, typename Float_>
class qdtsne::Status< num_dim_, Index_, Float_ >

Status of the t-SNE iterations.

Template Parameters

num_dim_	Number of dimensions in the t-SNE embedding.
Index_	Integer type of the observation indices.
Float_	Floating-point type of the neighbor distances and output embedding.

This class holds the precomputed structures required to perform the t-SNE iterations. Instances should not be constructed directly but instead created by initialize().

Member Function Documentation

cost()

template<std::size_t num_dim_, typename Index_ , typename Float_ >

Float_ qdtsne::Status< num_dim_, Index_, Float_ >::cost ( const Float_ *const Y )

inline

Parameters

[in]

Y

Pointer to a array containing a column-major matrix with number of rows and columns equal to num_dim_ and num_observations(), respectively. This should contain the location of each observation.

Returns

The Kullback-Leibler divergence for the current embedding.

This method is not const as it re-uses some of the pre-allocated buffers in the Status object for efficiency. Calling cost() at any time will not affect the results of subsequent calls to run().

This method does not consider any exaggeration of the conditional probabilities for iterations at or before Options::early_exaggeration_iterations. That is, all probabilities used here will not be exaggerated, regardless of the iteration.

iteration()

template<std::size_t num_dim_, typename Index_ , typename Float_ >

int qdtsne::Status< num_dim_, Index_, Float_ >::iteration ( ) const

inline

Returns

The number of iterations performed on this object so far.

max_iterations()

template<std::size_t num_dim_, typename Index_ , typename Float_ >

int qdtsne::Status< num_dim_, Index_, Float_ >::max_iterations ( ) const

inline

Returns

The maximum number of iterations, as specified in Options::max_iterations.

num_observations()

template<std::size_t num_dim_, typename Index_ , typename Float_ >

Index_ qdtsne::Status< num_dim_, Index_, Float_ >::num_observations ( ) const

inline

Returns

The number of observations in the dataset.

run() [1/2]

template<std::size_t num_dim_, typename Index_ , typename Float_ >

void qdtsne::Status< num_dim_, Index_, Float_ >::run ( Float_ *const Y )

inline

Run the algorithm to the maximum number of iterations. If run() has already been invoked with an iteration limit, this method will only perform the remaining iterations required for iteration() to reach max_iterations(). If iteration() is already greater than max_iterations(), this method is a no-op.

Parameters

[in,out]

Y

Pointer to a array containing a column-major matrix with number of rows and columns equal to num_dim_ and num_observations(), respectively. Each row corresponds to a dimension of the embedding while each column corresponds to an observation. On input, this should contain the initial location of each observation; on output, it is updated to the t-SNE location at the maximum number of iterations.

run() [2/2]

template<std::size_t num_dim_, typename Index_ , typename Float_ >

void qdtsne::Status< num_dim_, Index_, Float_ >::run ( Float_ *const Y, const int limit )

inline

Run the algorithm to the specified number of iterations. This can be invoked repeatedly with increasing limit to run the algorithm incrementally.

Parameters

[in,out]	Y	Pointer to a array containing a column-major matrix with number of rows and columns equal to num_dim_ and num_observations(), respectively. Each row corresponds to a dimension of the embedding while each column corresponds to an observation. On input, this should contain the initial location of each observation; on output, it is updated to the t-SNE location at the specified number of iterations.
	limit	Number of iterations to run up to. The actual number of iterations performed will be the difference between limit and iteration(), i.e., iteration() will be equal to limit on completion. limit may be greater than max_iterations(), to run the algorithm for more iterations than specified during construction of this Status object.

---

The documentation for this class was generated from the following file:

• qdtsne/Status.hpp