singlepp::TrainSingleOptions< Index_, Float_ > Struct Template Reference

Options for train_single() and friends. More...

#include <train_single.hpp>

Public Attributes
int	top = -1
std::shared_ptr< knncolle::Builder< knncolle::SimpleMatrix< Index_, Index_, Float_ >, Float_ > >	trainer
int	num_threads = 1

Detailed Description

template<typename Index_ = DefaultIndex, typename Float_ = DefaultFloat>
struct singlepp::TrainSingleOptions< Index_, Float_ >

Options for train_single() and friends.

Template Parameters

Index_	Integer type for the row/column indices of the matrix.
Float_	Floating-point type for the correlations and scores.

Member Data Documentation

num_threads

template<typename Index_ = DefaultIndex, typename Float_ = DefaultFloat>

int singlepp::TrainSingleOptions< Index_, Float_ >::num_threads = 1

Number of threads to use. The parallelization scheme is determined by tatami::parallelize().

top

template<typename Index_ = DefaultIndex, typename Float_ = DefaultFloat>

int singlepp::TrainSingleOptions< Index_, Float_ >::top = -1

Number of top markers to use from each pairwise comparison between labels. Larger values improve the stability of the correlations at the cost of increasing noise and computational work.

Setting this to a negative value will instruct train_single() to use all supplied markers. This is useful in situations where the supplied markers have already been curated.

trainer

template<typename Index_ = DefaultIndex, typename Float_ = DefaultFloat>

std::shared_ptr<knncolle::Builder<knncolle::SimpleMatrix<Index_, Index_, Float_>, Float_> > singlepp::TrainSingleOptions< Index_, Float_ >::trainer

Algorithm for the nearest-neighbor search. This allows us to skip the explicit calculation of correlations between each test cell and every reference sample. If NULL, this defaults to an exact search based on knncolle::VptreeBuilder.

---

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

• singlepp/train_single.hpp