DiSMEC++
Namespaces | Classes | Typedefs | Enumerations | Functions | Variables
dismec Namespace Reference

Main namespace in which all types, classes, and functions are defined. More...

Namespaces

 confusion_matrix_detail
 
 eigen_visitors
 
 init
 
 io
 
 l2_reg_sq_hinge_detail
 
 model
 
 objective
 
 parallel
 
 postproc
 
 prediction
 
 solvers
 
 stats
 
 types
 

Classes

class  DataProcessing
 
class  DatasetBase
 
class  BinaryData
 Collects the data related to a single optimization problem. More...
 
class  MultiLabelData
 
class  label_id_t
 Strong typedef for an int to signify a label id. More...
 
class  CascadeTraining
 
class  DiSMECTraining
 An implementation of TrainingSpec that models the DiSMEC algorithm. More...
 
class  TrainingSpec
 This class gathers the setting-specific parts of the training process. More...
 
struct  DismecTrainingConfig
 
struct  CascadeTrainingConfig
 
class  ResultStatsGatherer
 
class  TrainingStatsGatherer
 
class  TrainingTaskGenerator
 Generates tasks for training weights for the i'th label. More...
 
struct  TrainingResult
 
class  PropensityModel
 
class  WeightingScheme
 Base class for label-based weighting schemes. More...
 
class  ConstantWeighting
 Simple weighting scheme that assigns the same weighting to all label_ids. More...
 
class  PropensityWeighting
 
class  PropensityDownWeighting
 
class  CustomWeighting
 
class  FastSparseRowIter
 This is an almost verbatim copy of the SparseFeatures::InnerIterator provided by Eigen. More...
 
class  HashVector
 An Eigen vector with versioning information, to implement simple caching of results. More...
 
class  VectorHash
 A unique identifier for a HashVector. More...
 
class  CacheHelper
 
class  HyperParameterBase
 Base class for all objects that have adjustable hyper-parameters. More...
 
class  HyperParameters
 This class represents a set of hyper-parameters. More...
 
class  opaque_int_type
 An integer-like type that represents categorical values. More...
 
class  KahanAccumulator
 Implements a numerically stable sum algorithm. More...
 

Typedefs

using real_t = float
 The default type for floating point values. More...
 
using GenericOutMatrix = types::GenericMatrixRef< real_t >
 
using GenericInMatrix = types::GenericMatrixRef< const real_t >
 
using GenericOutVector = types::GenericVectorRef< real_t >
 
using GenericInVector = types::GenericVectorRef< const real_t >
 
using DenseRealVector = types::DenseVector< real_t >
 Any dense, real values vector. More...
 
using SparseRealVector = types::SparseVector< real_t >
 
using GenericRealVector = types::GenericVector< real_t >
 
using SparseFeatures = types::SparseRowMajor< real_t >
 Sparse Feature Matrix in Row Major format. More...
 
using DenseFeatures = types::DenseRowMajor< real_t >
 Dense Feature Matrix in Row Major format. More...
 
using GenericFeatureMatrix = types::GenericMatrix< DenseFeatures, SparseFeatures >
 
using BinaryLabelVector = types::DenseVector< std::int8_t >
 Dense vector for storing binary labels. More...
 
using PredictionMatrix = types::DenseRowMajor< real_t >
 Dense matrix in Row Major format used for predictions. More...
 
using IndexMatrix = types::DenseRowMajor< long >
 Matrix used for indices in sparse predictions. More...
 
using RegularizerSpec = std::variant< objective::SquaredNormConfig, objective::HuberConfig, objective::ElasticConfig >
 

Enumerations

enum class  DatasetTransform { IDENTITY , ONE_PLUS_LOG , LOG_ONE_PLUS , SQRT }
 
enum class  RegularizerType {
  REG_L2 , REG_L1 , REG_L1_RELAXED , REG_HUBER ,
  REG_ELASTIC_50_50 , REG_ELASTIC_90_10
}
 
enum class  LossType { SQUARED_HINGE , LOGISTIC , HUBER_HINGE , HINGE }
 

Functions

void augment_features_with_bias (DatasetBase &data, real_t bias=1)
 
SparseFeatures augment_features_with_bias (const SparseFeatures &features, real_t bias=1)
 
DenseFeatures augment_features_with_bias (const DenseFeatures &features, real_t bias=1)
 
DenseRealVector get_mean_feature (const GenericFeatureMatrix &features)
 
DenseRealVector get_mean_feature (const SparseFeatures &features)
 
DenseRealVector get_mean_feature (const DenseFeatures &features)
 
std::vector< long > count_features (const SparseFeatures &features)
 
void normalize_instances (DatasetBase &data)
 
void normalize_instances (SparseFeatures &features)
 
void normalize_instances (DenseFeatures &features)
 
Eigen::PermutationMatrix< Eigen::Dynamic, Eigen::Dynamic, int > sort_features_by_frequency (DatasetBase &data)
 
Eigen::PermutationMatrix< Eigen::Dynamic, Eigen::Dynamic, int > sort_features_by_frequency (SparseFeatures &features)
 
Eigen::PermutationMatrix< Eigen::Dynamic, Eigen::Dynamic, int > sort_features_by_frequency (DenseFeatures &features)
 
void hash_sparse_features (SparseFeatures &features, unsigned seed, int buckets, int repeats)
 
SparseFeatures shortlist_features (const SparseFeatures &source, const std::vector< long > &shortlist)
 
DenseFeatures shortlist_features (const DenseFeatures &source, const std::vector< long > &shortlist)
 
void transform_features (DatasetBase &data, DatasetTransform transform)
 
void transform_features (SparseFeatures &features, DatasetTransform transform)
 
void transform_features (DenseFeatures &features, DatasetTransform transform)
 
std::ptrdiff_t operator- (label_id_t a, label_id_t b)
 
label_id_t operator+ (label_id_t a, std::ptrdiff_t b)
 
std::shared_ptr< objective::Objective > make_loss (LossType type, std::shared_ptr< const GenericFeatureMatrix > X, std::unique_ptr< objective::Objective > regularizer)
 
std::shared_ptr< TrainingSpec > create_dismec_training (std::shared_ptr< const DatasetBase > data, HyperParameters params, DismecTrainingConfig config)
 
std::shared_ptr< TrainingSpec > create_cascade_training (std::shared_ptr< const DatasetBase > data, std::shared_ptr< const GenericFeatureMatrix > dense, std::shared_ptr< const std::vector< std::vector< long >>> shortlist, HyperParameters params, CascadeTrainingConfig config)
 
TrainingResult run_training (parallel::ParallelRunner &runner, std::shared_ptr< TrainingSpec > spec, label_id_t begin_label=label_id_t{0}, label_id_t end_label=label_id_t{-1})
 
template<class T >
constexpr long to_long (T value)
 Convert the given value to long, throwing an error if the conversion is not possible. More...
 
template<class T >
constexpr std::ptrdiff_t calc_ssizeof ()
 Gets the sizeof of a type as a signed integer. More...
 
template<class C >
constexpr auto ssize (const C &c) -> std::common_type_t< std::ptrdiff_t, std::make_signed_t< decltype(c.size())>>
 signed size free function. Taken from https://en.cppreference.com/w/cpp/iterator/size More...
 
template<typename Scalar , int Options, typename StorageIndex , typename OtherDerived >
auto fast_dot (const Eigen::SparseMatrix< Scalar, Options, StorageIndex > &first, int row, const Eigen::MatrixBase< OtherDerived > &other) -> Scalar
 
template<class T >
auto operator+ (const HashVector &vec, T &&other)
 
template<class T >
auto operator+ (T &&other, const HashVector &vec)
 
template<class T >
auto operator+= (T &&other, const HashVector &vec)
 
template<class T >
auto operator* (const HashVector &vec, T &&other)
 
template<class T >
auto operator* (T &&other, const HashVector &vec)
 
template<class T >
auto operator*= (T &&other, const HashVector &vec)
 
template<class Tag , class T >
constexpr bool operator== (opaque_int_type< Tag, T > a, opaque_int_type< Tag, T > b)
 
template<class Tag , class T >
constexpr bool operator!= (opaque_int_type< Tag, T > a, opaque_int_type< Tag, T > b)
 
template<class Tag , class T >
constexpr bool operator<= (opaque_int_type< Tag, T > a, opaque_int_type< Tag, T > b)
 
template<class Tag , class T >
constexpr bool operator< (opaque_int_type< Tag, T > a, opaque_int_type< Tag, T > b)
 
template<class Tag , class T >
constexpr bool operator> (opaque_int_type< Tag, T > a, opaque_int_type< Tag, T > b)
 
template<class Tag , class T >
constexpr bool operator>= (opaque_int_type< Tag, T > a, opaque_int_type< Tag, T > b)
 
template<class Tag , class T >
std::ostream & operator<< (std::ostream &stream, opaque_int_type< Tag, T > a)
 
SparseFeatures make_uniform_sparse_matrix (int rows, int cols, int non_zeros_per_row)
 Creates a sparse matrix with the given number of rows and columns. More...
 

Variables

constexpr const long CG_MIN_ITER_BOUND = 5
 
constexpr const real_t CG_DEFAULT_EPSILON = 0.5
 The default value for the tolerance parameter of the conjugate gradient optimization. More...
 
constexpr const int MIN_TIME_PER_CHUNK_MS = 5
 If the time needed per chunk of work is less than this, we display a warning. More...
 
constexpr const int PREDICTION_RUN_CHUNK_SIZE = 1024
 Default chunk size for predicting scores. More...
 
constexpr const int PREDICTION_METRICS_CHUNK_SIZE = 4096
 Default chunk size for calculating metrics. More...
 
template<class T >
constexpr std::ptrdiff_t ssizeof = calc_ssizeof<T>()
 Signed size of type T More...
 

Detailed Description

Main namespace in which all types, classes, and functions are defined.

Typedef Documentation

◆ BinaryLabelVector

using dismec::BinaryLabelVector = typedef types::DenseVector<std::int8_t>

Dense vector for storing binary labels.

We use this type to store a dense representation of a binary label where +1 represents presence of the label and -1 represents its absence.

Todo:
decide whether this should be a matrix or an array type

Definition at line 68 of file matrix_types.h.

◆ DenseFeatures

using dismec::DenseFeatures = typedef types::DenseRowMajor<real_t>

Dense Feature Matrix in Row Major format.

This is the format in which we store the features of a dense dataset. We use a RowMajor format, because we usually want to iterate over all the features of one example, but not over all the examples of a given feature. Each row corresponds to one instance, and each column to a feature.

Definition at line 58 of file matrix_types.h.

◆ DenseRealVector

using dismec::DenseRealVector = typedef types::DenseVector<real_t>

Any dense, real values vector.

Definition at line 40 of file matrix_types.h.

◆ GenericFeatureMatrix

using dismec::GenericFeatureMatrix = typedef types::GenericMatrix<DenseFeatures, SparseFeatures>

Definition at line 60 of file matrix_types.h.

◆ GenericInMatrix

using dismec::GenericInMatrix = typedef types::GenericMatrixRef<const real_t>

Definition at line 33 of file matrix_types.h.

◆ GenericInVector

using dismec::GenericInVector = typedef types::GenericVectorRef<const real_t>

Definition at line 35 of file matrix_types.h.

◆ GenericOutMatrix

using dismec::GenericOutMatrix = typedef types::GenericMatrixRef<real_t>

Definition at line 32 of file matrix_types.h.

◆ GenericOutVector

using dismec::GenericOutVector = typedef types::GenericVectorRef<real_t>

Definition at line 34 of file matrix_types.h.

◆ GenericRealVector

using dismec::GenericRealVector = typedef types::GenericVector<real_t>

Definition at line 42 of file matrix_types.h.

◆ IndexMatrix

using dismec::IndexMatrix = typedef types::DenseRowMajor<long>

Matrix used for indices in sparse predictions.

This matrix is used for predictions, thus it is in row-major order.

Definition at line 81 of file matrix_types.h.

◆ PredictionMatrix

using dismec::PredictionMatrix = typedef types::DenseRowMajor<real_t>

Dense matrix in Row Major format used for predictions.

This is the matrix type used for dense predictions. To facility predictions on an instance-by-instance basis, this is a RowMajor matrix type. Each row corresponds to one instance, and each column to a label.

Definition at line 75 of file matrix_types.h.

◆ real_t

typedef float dismec::real_t

The default type for floating point values.

Definition at line 17 of file config.h.

◆ RegularizerSpec

using dismec::RegularizerSpec = typedef std::variant<objective::SquaredNormConfig, objective::HuberConfig, objective::ElasticConfig>

Definition at line 143 of file spec.h.

◆ SparseFeatures

using dismec::SparseFeatures = typedef types::SparseRowMajor<real_t>

Sparse Feature Matrix in Row Major format.

This is the format in which we store the features of a sparse dataset. We use a RowMajor format, because we usually want to iterate over all the features of one example, but not over all the examples of a given feature. Each row corresponds to one instance, and each column to a feature.

Definition at line 50 of file matrix_types.h.

◆ SparseRealVector

using dismec::SparseRealVector = typedef types::SparseVector<real_t>

Definition at line 41 of file matrix_types.h.

Enumeration Type Documentation

◆ DatasetTransform

enum dismec::DatasetTransform
strong
Enumerator
IDENTITY 
ONE_PLUS_LOG 
LOG_ONE_PLUS 
SQRT 

Definition at line 37 of file transform.h.

◆ LossType

enum dismec::LossType
strong
Enumerator
SQUARED_HINGE 
LOGISTIC 
HUBER_HINGE 
HINGE 

Definition at line 129 of file spec.h.

◆ RegularizerType

enum dismec::RegularizerType
strong
Enumerator
REG_L2 
REG_L1 
REG_L1_RELAXED 
REG_HUBER 
REG_ELASTIC_50_50 
REG_ELASTIC_90_10 

Definition at line 120 of file spec.h.

Function Documentation

◆ augment_features_with_bias() [1/3]

DenseFeatures dismec::augment_features_with_bias ( const DenseFeatures &  features,
real_t  bias = 1 
)

Definition at line 44 of file transform.cpp.

◆ augment_features_with_bias() [2/3]

SparseFeatures dismec::augment_features_with_bias ( const SparseFeatures &  features,
real_t  bias = 1 
)

Definition at line 29 of file transform.cpp.

◆ augment_features_with_bias() [3/3]

void dismec::augment_features_with_bias ( DatasetBase &  data,
real_t  bias = 1 
)

Definition at line 25 of file transform.cpp.

References dismec::DatasetBase::edit_features(), and dismec::types::visit().

Referenced by dismec::DataProcessing::load(), anonymous_namespace{transform.cpp}::VisitorBias::operator()(), TrainingProgram::run(), and TEST_CASE().

◆ calc_ssizeof()

template<class T >
constexpr std::ptrdiff_t dismec::calc_ssizeof ( )
constexpr

Gets the sizeof of a type as a signed integer.

Definition at line 32 of file conversion.h.

◆ count_features()

std::vector< long > dismec::count_features ( const SparseFeatures &  features)

Definition at line 114 of file transform.cpp.

Referenced by main(), and sort_features_by_frequency().

◆ create_cascade_training()

std::shared_ptr< TrainingSpec > dismec::create_cascade_training ( std::shared_ptr< const DatasetBase >  data,
std::shared_ptr< const GenericFeatureMatrix >  dense,
std::shared_ptr< const std::vector< std::vector< long >>>  shortlist,
HyperParameters  params,
CascadeTrainingConfig  config 
)

Definition at line 161 of file cascade.cpp.

References dismec::postproc::create_identity(), dismec::init::create_zero_initializer(), dismec::CascadeTrainingConfig::DenseInit, dismec::CascadeTrainingConfig::DenseReg, dismec::CascadeTrainingConfig::PostProcessing, dismec::CascadeTrainingConfig::SparseInit, dismec::CascadeTrainingConfig::SparseReg, and dismec::CascadeTrainingConfig::StatsGatherer.

Referenced by TrainingProgram::run().

◆ create_dismec_training()

std::shared_ptr< TrainingSpec > dismec::create_dismec_training ( std::shared_ptr< const DatasetBase >  data,
HyperParameters  params,
DismecTrainingConfig  config 
)

Definition at line 157 of file dismec.cpp.

References dismec::postproc::create_identity(), dismec::init::create_zero_initializer(), dismec::DismecTrainingConfig::Init, dismec::DismecTrainingConfig::Loss, dismec::DismecTrainingConfig::PostProcessing, dismec::DismecTrainingConfig::Regularizer, dismec::DismecTrainingConfig::Sparse, dismec::DismecTrainingConfig::StatsGatherer, and dismec::DismecTrainingConfig::Weighting.

Referenced by register_training().

◆ fast_dot()

template<typename Scalar , int Options, typename StorageIndex , typename OtherDerived >
auto dismec::fast_dot ( const Eigen::SparseMatrix< Scalar, Options, StorageIndex > &  first,
int  row,
const Eigen::MatrixBase< OtherDerived > &  other 
) -> Scalar

Definition at line 55 of file fast_sparse_row_iter.h.

Referenced by dismec::l2_reg_sq_hinge_detail::__attribute__().

◆ get_mean_feature() [1/3]

DenseRealVector dismec::get_mean_feature ( const DenseFeatures &  features)

Definition at line 75 of file transform.cpp.

◆ get_mean_feature() [2/3]

DenseRealVector dismec::get_mean_feature ( const GenericFeatureMatrix &  features)

Definition at line 52 of file transform.cpp.

References dismec::types::visit().

Referenced by TrainingProgram::make_config(), and dismec::init::SubsetFeatureMeanStrategy::SubsetFeatureMeanStrategy().

◆ get_mean_feature() [3/3]

DenseRealVector dismec::get_mean_feature ( const SparseFeatures &  features)

Definition at line 57 of file transform.cpp.

◆ hash_sparse_features()

void dismec::hash_sparse_features ( SparseFeatures &  features,
unsigned  seed,
int  buckets,
int  repeats 
)

Definition at line 183 of file transform.cpp.

Referenced by dismec::DataProcessing::load().

◆ make_loss()

std::shared_ptr< objective::Objective > dismec::make_loss ( LossType  type,
std::shared_ptr< const GenericFeatureMatrix >  X,
std::unique_ptr< objective::Objective >  regularizer 
)

Definition at line 41 of file dismec.cpp.

References HINGE, HUBER_HINGE, LOGISTIC, dismec::objective::make_huber_hinge(), dismec::objective::make_logistic_loss(), dismec::objective::make_squared_hinge(), SQUARED_HINGE, and THROW_EXCEPTION.

Referenced by dismec::init::create_ova_primal_initializer(), and dismec::DiSMECTraining::make_objective().

◆ make_uniform_sparse_matrix()

dismec::SparseFeatures dismec::make_uniform_sparse_matrix ( int  rows,
int  cols,
int  non_zeros_per_row 
)

Creates a sparse matrix with the given number of rows and columns.

Parameters
rowsNumber of rows in the matrix.
colsNumber of columns in the martix.
non_zeros_per_rowNumber of nonzero entries in each row. The non-zeros will be distributed uniformly among the columns.
Returns
The resulting sparse matrix.

Definition at line 8 of file test_utils.cpp.

Referenced by TEST_CASE().

◆ normalize_instances() [1/3]

void dismec::normalize_instances ( DatasetBase &  data)

Definition at line 88 of file transform.cpp.

References dismec::DatasetBase::edit_features(), and dismec::types::visit().

Referenced by apply_tfidf(), dismec::DataProcessing::load(), and TrainingProgram::run().

◆ normalize_instances() [2/3]

void dismec::normalize_instances ( DenseFeatures &  features)

Definition at line 101 of file transform.cpp.

◆ normalize_instances() [3/3]

void dismec::normalize_instances ( SparseFeatures &  features)

Definition at line 92 of file transform.cpp.

◆ operator!=()

template<class Tag , class T >
constexpr bool dismec::operator!= ( opaque_int_type< Tag, T >  a,
opaque_int_type< Tag, T >  b 
)
inlineconstexpr

Definition at line 45 of file opaque_int.h.

References dismec::opaque_int_type< Tag, T >::to_index().

◆ operator*() [1/2]

template<class T >
auto dismec::operator* ( const HashVector &  vec,
T &&  other 
)

Definition at line 167 of file hash_vector.h.

References dismec::HashVector::get().

◆ operator*() [2/2]

template<class T >
auto dismec::operator* ( T &&  other,
const HashVector &  vec 
)

Definition at line 172 of file hash_vector.h.

References dismec::HashVector::get().

◆ operator*=()

template<class T >
auto dismec::operator*= ( T &&  other,
const HashVector &  vec 
)

Definition at line 177 of file hash_vector.h.

References dismec::HashVector::get().

◆ operator+() [1/3]

template<class T >
auto dismec::operator+ ( const HashVector &  vec,
T &&  other 
)

Definition at line 152 of file hash_vector.h.

References dismec::HashVector::get().

◆ operator+() [2/3]

label_id_t dismec::operator+ ( label_id_t  a,
std::ptrdiff_t  b 
)
inline

Definition at line 34 of file types.h.

References dismec::opaque_int_type< Tag, T >::to_index().

◆ operator+() [3/3]

template<class T >
auto dismec::operator+ ( T &&  other,
const HashVector &  vec 
)

Definition at line 157 of file hash_vector.h.

References dismec::HashVector::get().

◆ operator+=()

template<class T >
auto dismec::operator+= ( T &&  other,
const HashVector &  vec 
)

Definition at line 162 of file hash_vector.h.

References dismec::HashVector::get().

◆ operator-()

std::ptrdiff_t dismec::operator- ( label_id_t  a,
label_id_t  b 
)
inline

Definition at line 30 of file types.h.

References dismec::opaque_int_type< Tag, T >::to_index().

◆ operator<()

template<class Tag , class T >
constexpr bool dismec::operator< ( opaque_int_type< Tag, T >  a,
opaque_int_type< Tag, T >  b 
)
inlineconstexpr

Definition at line 55 of file opaque_int.h.

References dismec::opaque_int_type< Tag, T >::to_index().

◆ operator<<()

template<class Tag , class T >
std::ostream& dismec::operator<< ( std::ostream &  stream,
opaque_int_type< Tag, T >  a 
)

Definition at line 70 of file opaque_int.h.

References dismec::opaque_int_type< Tag, T >::to_index().

◆ operator<=()

template<class Tag , class T >
constexpr bool dismec::operator<= ( opaque_int_type< Tag, T >  a,
opaque_int_type< Tag, T >  b 
)
inlineconstexpr

Definition at line 50 of file opaque_int.h.

References dismec::opaque_int_type< Tag, T >::to_index().

◆ operator==()

template<class Tag , class T >
constexpr bool dismec::operator== ( opaque_int_type< Tag, T >  a,
opaque_int_type< Tag, T >  b 
)
inlineconstexpr

Definition at line 40 of file opaque_int.h.

References dismec::opaque_int_type< Tag, T >::to_index().

◆ operator>()

template<class Tag , class T >
constexpr bool dismec::operator> ( opaque_int_type< Tag, T >  a,
opaque_int_type< Tag, T >  b 
)
inlineconstexpr

Definition at line 60 of file opaque_int.h.

References dismec::opaque_int_type< Tag, T >::to_index().

◆ operator>=()

template<class Tag , class T >
constexpr bool dismec::operator>= ( opaque_int_type< Tag, T >  a,
opaque_int_type< Tag, T >  b 
)
inlineconstexpr

Definition at line 65 of file opaque_int.h.

References dismec::opaque_int_type< Tag, T >::to_index().

◆ run_training()

TrainingResult dismec::run_training ( parallel::ParallelRunner &  runner,
std::shared_ptr< TrainingSpec >  spec,
label_id_t  begin_label = label_id_t{0},
label_id_t  end_label = label_id_t{-1} 
)

Definition at line 122 of file training.cpp.

References dismec::parallel::ParallelRunner::run().

Referenced by register_training(), and TrainingProgram::run().

◆ shortlist_features() [1/2]

DenseFeatures dismec::shortlist_features ( const DenseFeatures &  source,
const std::vector< long > &  shortlist 
)

Definition at line 235 of file transform.cpp.

◆ shortlist_features() [2/2]

SparseFeatures dismec::shortlist_features ( const SparseFeatures &  source,
const std::vector< long > &  shortlist 
)

Definition at line 219 of file transform.cpp.

Referenced by dismec::CascadeTraining::update_objective().

◆ sort_features_by_frequency() [1/3]

Eigen::PermutationMatrix< Eigen::Dynamic, Eigen::Dynamic, int > dismec::sort_features_by_frequency ( DatasetBase &  data)

Definition at line 110 of file transform.cpp.

References dismec::DatasetBase::edit_features(), and dismec::types::visit().

Referenced by TEST_CASE().

◆ sort_features_by_frequency() [2/3]

Eigen::PermutationMatrix< Eigen::Dynamic, Eigen::Dynamic, int > dismec::sort_features_by_frequency ( DenseFeatures &  features)

Definition at line 146 of file transform.cpp.

◆ sort_features_by_frequency() [3/3]

Eigen::PermutationMatrix< Eigen::Dynamic, Eigen::Dynamic, int > dismec::sort_features_by_frequency ( SparseFeatures &  features)

Definition at line 127 of file transform.cpp.

References count_features().

◆ ssize()

template<class C >
constexpr auto dismec::ssize ( const C &  c) -> std::common_type_t<std::ptrdiff_t, std::make_signed_t<decltype(c.size())>>
constexpr

signed size free function. Taken from https://en.cppreference.com/w/cpp/iterator/size

Definition at line 42 of file conversion.h.

Referenced by dismec::l2_reg_sq_hinge_detail::__attribute__(), dismec::TrainingStatsGatherer::add_accu(), dismec::stats::StatisticsCollection::declare_stat(), dismec::stats::StatisticsCollection::declare_tag(), dismec::prediction::EvaluateMetrics::EvaluateMetrics(), dismec::prediction::EvaluateMetrics::finalize(), dismec::prediction::ConfusionMatrixRecorder::get_confusion_matrix(), dismec::prediction::MacroMetricReporter::get_values(), dismec::l2_reg_sq_hinge_detail::htd_sum(), dismec::prediction::InstanceRankedPositives::InstanceRankedPositives(), main(), anonymous_namespace{regularizers_imp.cpp}::make_vec(), dismec::stats::TaggedStat::merge_imp(), dismec::MultiLabelData::num_labels(), dismec::MultiLabelData::num_negatives(), dismec::MultiLabelData::num_positives(), obesity(), anonymous_namespace{sparse.cpp}::PredictVisitor::operator()(), anonymous_namespace{xmc.cpp}::read_into_buffers(), anonymous_namespace{numpy.cpp}::read_key_value(), dismec::io::read_xmc_dataset(), dismec::stats::TaggedStat::record_real(), dismec::prediction::ConfusionMatrixRecorder::reduce(), anonymous_namespace{numpy.cpp}::skip_whitespace(), anonymous_namespace{collection.cpp}::str_to_id(), TEST_CASE(), dismec::prediction::InstanceRankedPositives::update(), dismec::prediction::AbandonmentAtK::update(), dismec::CascadeTraining::update_objective(), and anonymous_namespace{xmc.cpp}::write_label_list().

◆ to_long()

template<class T >
constexpr long dismec::to_long ( T  value)
constexpr

Convert the given value to long, throwing an error if the conversion is not possible.

Definition at line 14 of file conversion.h.

References THROW_EXCEPTION.

Referenced by dismec::objective::Regularized_SquaredHingeSVC::gradient_and_pre_conditioner_tpl(), anonymous_namespace{numpy.cpp}::read_key_value(), dismec::parallel::ParallelRunner::run(), and dismec::l2_reg_sq_hinge_detail::value_from_xTw().

◆ transform_features() [1/3]

void dismec::transform_features ( DatasetBase &  data,
DatasetTransform  transform 
)

Definition at line 152 of file transform.cpp.

References dismec::DatasetBase::edit_features(), and dismec::types::visit().

Referenced by dismec::DataProcessing::load(), and TrainingProgram::run().

◆ transform_features() [2/3]

void dismec::transform_features ( DenseFeatures &  features,
DatasetTransform  transform 
)

Definition at line 179 of file transform.cpp.

References anonymous_namespace{transform.cpp}::transform_features_imp().

◆ transform_features() [3/3]

void dismec::transform_features ( SparseFeatures &  features,
DatasetTransform  transform 
)

Definition at line 175 of file transform.cpp.

References anonymous_namespace{transform.cpp}::transform_features_imp().

Variable Documentation

◆ CG_DEFAULT_EPSILON

constexpr const real_t dismec::CG_DEFAULT_EPSILON = 0.5
constexpr

The default value for the tolerance parameter of the conjugate gradient optimization.

Definition at line 25 of file config.h.

◆ CG_MIN_ITER_BOUND

constexpr const long dismec::CG_MIN_ITER_BOUND = 5
constexpr

The minimum upper bound for the number of CG iterations. If the problem has less than this many dimensions, the upper-bound for the number of CG iterations is still given by this number. TODO is it even sensible to define something like this?

Definition at line 22 of file config.h.

Referenced by dismec::solvers::CGMinimizer::do_minimize().

◆ MIN_TIME_PER_CHUNK_MS

constexpr const int dismec::MIN_TIME_PER_CHUNK_MS = 5
constexpr

If the time needed per chunk of work is less than this, we display a warning.

Definition at line 28 of file config.h.

Referenced by dismec::parallel::ParallelRunner::run().

◆ PREDICTION_METRICS_CHUNK_SIZE

constexpr const int dismec::PREDICTION_METRICS_CHUNK_SIZE = 4096
constexpr

Default chunk size for calculating metrics.

Definition at line 42 of file config.h.

Referenced by main().

◆ PREDICTION_RUN_CHUNK_SIZE

constexpr const int dismec::PREDICTION_RUN_CHUNK_SIZE = 1024
constexpr

Default chunk size for predicting scores.

Definition at line 39 of file config.h.

Referenced by main().

◆ ssizeof

template<class T >
constexpr std::ptrdiff_t dismec::ssizeof = calc_ssizeof<T>()
constexpr

Signed size of type T

Definition at line 38 of file conversion.h.