#include <data_reader_merge_samples.hpp>

Inheritance diagram for lbann::data_reader_merge_samples:

Collaboration diagram for lbann::data_reader_merge_samples:

Public Member Functions
	data_reader_merge_samples (std::vector< generic_data_reader *> data_readers, bool shuffle=true)

	data_reader_merge_samples (const data_reader_merge_samples &)

data_reader_merge_samples &	operator= (const data_reader_merge_samples &)

	~data_reader_merge_samples () override

data_reader_merge_samples *	copy () const override

std::string	get_type () const override

void	load () override
	Load subsidiary data readers. More...

int	get_num_labels () const override
	Return the number of labels (classes) in this dataset. More...

int	get_num_responses () const override
	Return the number of responses in this dataset. More...

int	get_linearized_data_size () const override
	Get the linearized size (i.e. number of elements) in a sample. More...

int	get_linearized_label_size () const override
	Get the linearized size (i.e. number of elements) in a label. More...

int	get_linearized_response_size () const override
	Get the linearized size (i.e. number of elements) in a response. More...

const std::vector< El::Int >	get_data_dims () const override
	Get the dimensions of the data. More...

const std::vector< int > &	get_num_samples_psum ()
	support for data store functionality More...

Public Member Functions inherited from lbann::generic_compound_data_reader
	generic_compound_data_reader (std::vector< generic_data_reader *> data_readers, bool shuffle=true)

	generic_compound_data_reader (const generic_compound_data_reader &other)

generic_compound_data_reader &	operator= (const generic_compound_data_reader &other)

	~generic_compound_data_reader () override

void	set_execution_mode_split_fraction (execution_mode m, double s) override
	Apply operations to subsidiary data readers. More...

void	set_role (std::string role) override

std::vector< generic_data_reader * > &	get_data_readers ()
	needed to support data_store_merge_samples More...

bool	has_labels () const override

bool	has_responses () const override

void	set_has_labels (const bool b) override
	Whether or not a data reader has labels. More...

void	set_has_responses (const bool b) override
	Whether or not a data reader has a response field. More...

Public Member Functions inherited from lbann::generic_data_reader
	generic_data_reader (bool shuffle=true)

	generic_data_reader (const generic_data_reader &)=default

generic_data_reader &	operator= (const generic_data_reader &)=default

virtual	~generic_data_reader ()

template<class Archive >
void	serialize (Archive &ar)

void	set_comm (lbann_comm *comm)
	set the comm object More...

lbann_comm *	get_comm () const
	returns a (possibly nullptr) to comm More...

virtual bool	has_conduit_output ()

void	set_file_dir (std::string s)

void	set_local_file_dir (std::string s)

void	set_max_files_to_load (size_t n)

std::string	get_file_dir () const

std::string	get_local_file_dir () const

void	set_data_sample_list (std::string s)

std::string	get_data_sample_list () const

void	keep_sample_order (bool same_order=false)

void	set_data_filename (std::string s)

std::string	get_data_filename () const

void	set_label_filename (std::string s)

std::string	get_label_filename () const

void	set_shuffle (bool b)

bool	is_shuffled () const

void	set_shuffled_indices (const std::vector< int > &indices)

const std::vector< int > &	get_shuffled_indices () const

void	set_first_n (int n)

void	set_absolute_sample_count (size_t s)

void	set_use_fraction (double s)

std::string	get_role () const

virtual void	setup (int num_io_threads, observer_ptr< thread_pool > io_thread_pool)

int	fetch (std::map< data_field_type, CPUMat *> &input_buffers, El::Matrix< El::Int > &indices_fetched, size_t mb_size)
	Fetch a mini-batch worth of data, including samples, labels, responses (as appropriate) More...

int	fetch (std::vector< conduit::Node > &samples, El::Matrix< El::Int > &indices_fetched, size_t mb_size)

virtual bool	has_data_field (data_field_type data_field) const
	Check to see if the data reader supports this specific data field. More...

void	set_has_data_field (data_field_type const data_field, const bool b)
	Whether or not a data reader has a data field. More...

void	start_data_store_mini_batch_exchange ()

void	finish_data_store_mini_batch_exchange ()

virtual bool	update (bool is_active_reader)

virtual int	get_linearized_size (data_field_type const &data_field) const
	get the linearized size of what is identified by desc. More...

virtual std::vector< El::Int >	get_slice_points (const slice_points_mode var_category, bool &is_supported)

virtual bool	position_valid () const
	True if the data reader's current position is valid. More...

virtual bool	position_is_overrun () const

bool	at_new_epoch () const
	True if the data reader is at the start of an epoch. More...

void	set_mini_batch_size (const int s)
	Set the mini batch size. More...

int	get_mini_batch_size () const
	Get the mini batch size. More...

int	get_loaded_mini_batch_size () const
	Get the loaded mini-batch size. More...

int	get_current_mini_batch_size () const
	Get the current mini-batch size. More...

int	get_mini_batch_max () const
	Return the full mini_batch_size. More...

void	set_stride_to_next_mini_batch (const int s)
	Set the mini batch stride. More...

int	get_stride_to_next_mini_batch () const
	Return the mini batch stride. More...

void	set_sample_stride (const int s)
	Set the sample stride. More...

int	get_sample_stride () const
	Return the sample stride. More...

void	set_iteration_stride (const int s)
	Set the iteration stride. More...

int	get_iteration_stride () const
	Return the iteration stride. More...

virtual void	set_base_offset (const int s)
	Return the base offset. More...

int	get_base_offset () const
	Return the base offset. More...

void	set_last_mini_batch_size (const int s)
	Set the last mini batch size. More...

int	get_last_mini_batch_size () const
	Return the last mini batch size. More...

void	set_stride_to_last_mini_batch (const int s)
	Set the last mini batch stride. More...

int	get_stride_to_last_mini_batch () const
	Return the last mini batch stride. More...

void	set_num_parallel_readers (const int s)
	Set the number of parallel readers per model. More...

int	get_num_parallel_readers () const
	Return the number of parallel readers per model. More...

virtual void	set_reset_mini_batch_index (const int s)
	Set the starting mini-batch index for the epoch. More...

int	get_reset_mini_batch_index () const
	Return the starting mini-batch index for the epoch. More...

int	get_loaded_mini_batch_index () const
	Return the current mini-batch index for the epoch. More...

int	get_current_mini_batch_index () const
	Return the current mini-batch index for the epoch. More...

void	set_initial_position ()
	Set the current position based on the base and model offsets. More...

int	get_position () const
	Get the current position in the data reader. More...

int	get_next_position () const
	Get the next position in the data reader. More...

int *	get_indices ()
	Get a pointer to the start of the shuffled indices. More...

virtual int	get_num_data () const
	Get the number of samples in this dataset. More...

int	get_num_unused_data (execution_mode m) const
	Get the number of unused samples in this dataset. More...

int *	get_unused_data (execution_mode m)
	Get a pointer to the start of the unused sample indices. More...

const std::vector< int > &	get_unused_indices (execution_mode m)

void	set_num_iterations_per_epoch (int num_iterations_per_epoch)
	Set the number of iterations in each epoch. More...

int	get_num_iterations_per_epoch () const
	Get the number of iterations in each epoch. More...

int	get_current_step_in_epoch () const

void	resize_shuffled_indices ()

void	select_subset_of_data ()

virtual void	use_unused_index_set (execution_mode m)

virtual bool	has_list_per_model () const
	Does the data reader have a unique sample list per model. More...

virtual bool	has_list_per_trainer () const
	Does the data reader have a unique sample list per trainer. More...

bool	save_to_checkpoint_shared (persist &p, execution_mode mode)
	Given directory to store checkpoint files, write state to file and add to number of bytes written. More...

bool	load_from_checkpoint_shared (persist &p, execution_mode mode)
	Given directory to store checkpoint files, read state from file and add to number of bytes read. More...

bool	save_to_checkpoint_distributed (persist &p, execution_mode mode)

bool	load_from_checkpoint_distributed (persist &p, execution_mode mode)
	Given directory to store checkpoint files, read state from file and add to number of bytes read. More...

const data_store_conduit &	get_data_store () const
	returns a const ref to the data store More...

data_store_conduit &	get_data_store ()
	returns a non-const ref to the data store More...

data_store_conduit *	get_data_store_ptr () const

void	setup_data_store (int mini_batch_size)

void	instantiate_data_store ()

virtual void	preload_data_store ()

void	set_gan_labelling (bool has_gan_labelling)

void	set_gan_label_value (int gan_label_value)

void	set_data_store (data_store_conduit *g)
	support of data store functionality More...

virtual bool	data_store_active () const

virtual bool	priming_data_store () const

virtual void	post_update ()

void	set_transform_pipeline (transform::transform_pipeline &&tp)

void	print_get_methods (const std::string filename)
	Print the return values from various get_X methods to file. More...

size_t	get_num_indices_to_use () const

void	set_use_data_store (bool s)

Protected Member Functions
bool	fetch_datum (CPUMat &X, int data_id, int mb_idx) override

bool	fetch_label (CPUMat &Y, int data_id, int mb_idx) override

bool	fetch_response (CPUMat &Y, int data_id, int mb_idx) override

void	setup_indices (int num_samples)
	code common to both load() and load_using_data_store() More...

size_t	compute_num_samples_psum ()
	code common to both load() and load_using_data_store() More...

void	sanity_check_for_consistency (int num_labels, int data_size, int label_size, const std::vector< El::Int > &data_dims)
	code common to both load() and load_using_data_store() More...

Protected Member Functions inherited from lbann::generic_data_reader
size_t	get_absolute_sample_count () const

double	get_use_fraction () const

double	get_execution_mode_split_fraction (execution_mode m) const

virtual bool	fetch_data_block (std::map< data_field_type, CPUMat *> &input_buffers, El::Int block_offset, El::Int block_stride, El::Int mb_size, El::Matrix< El::Int > &indices_fetched)

bool	fetch_data_block_conduit (std::vector< conduit::Node > &samples, El::Int block_offset, El::Int block_stride, El::Int mb_size, El::Matrix< El::Int > &indices_fetched)

virtual bool	fetch_data_field (data_field_type data_field, CPUMat &Y, int data_id, int mb_idx)
	Called by fetch_data, fetch_label, fetch_response. More...

virtual bool	fetch_conduit_node (conduit::Node &sample, int data_id)

CPUMat	create_datum_view (CPUMat &X, const int mb_idx)

virtual void	preprocess_data_source (int tid)

virtual void	postprocess_data_source (int tid)

virtual void	shuffle_indices ()
	Shuffle indices (uses the data_seq_generator) More...

virtual void	shuffle_indices (rng_gen &gen)
	Shuffle indices and profide a random number generator. More...

void	error_check_counts () const

Protected Attributes
std::vector< int >	m_num_samples_psum
	Partial sums of the number of samples in each reader. More...

Protected Attributes inherited from lbann::generic_compound_data_reader
std::vector< generic_data_reader * >	m_data_readers
	List of readers providing data. More...

Protected Attributes inherited from lbann::generic_data_reader
bool	m_verbose = false

std::unordered_set< int >	m_using_random_node

data_store_conduit *	m_data_store

lbann_comm *	m_comm

bool	m_use_data_store = false

std::map< data_field_type, bool >	m_supported_input_types
	Holds a true value for each input data type that is supported. Use an ordered map so that checkpoints are stable. More...

bool	m_gan_labelling

int	m_gan_label_value

observer_ptr< thread_pool >	m_io_thread_pool

bool	m_keep_sample_order

transform::transform_pipeline	m_transform_pipeline

bool	m_issue_warning

Additional Inherited Members
Public Types inherited from lbann::generic_data_reader
using	unused_index_map_t = std::map< execution_mode, std::vector< int > >

Public Attributes inherited from lbann::generic_data_reader
int	m_mini_batch_size

int	m_current_pos

int	m_stride_to_next_mini_batch

int	m_base_offset

int	m_sample_stride

int	m_iteration_stride
	Stride used by parallel data readers within the model. More...

std::vector< int >	m_shuffled_indices

unused_index_map_t	m_unused_indices
	Record of the indicies that are not being used for training. More...

int	m_last_mini_batch_size

int	m_stride_to_last_mini_batch

int	m_reset_mini_batch_index
	The index at which this data reader starts its epoch. More...

int	m_loaded_mini_batch_idx
	The index of the current mini-batch that has been loaded. More...

int	m_current_mini_batch_idx

int	m_num_iterations_per_epoch

int	m_num_parallel_readers
	How many iterations all readers will execute. More...

size_t	m_max_files_to_load
	How many parallel readers are being used. More...

std::string	m_file_dir

std::string	m_local_file_dir

std::string	m_data_sample_list

std::string	m_data_fn

std::string	m_label_fn

bool	m_shuffle

size_t	m_absolute_sample_count

std::map< execution_mode, double >	m_execution_mode_split_fraction

double	m_use_fraction

int	m_first_n

std::string	m_role

Detailed Description

Data reader for merging the samples from multiple data readers into a single dataset.

Definition at line 40 of file data_reader_merge_samples.hpp.

Constructor & Destructor Documentation

◆ data_reader_merge_samples() [1/2]

lbann::data_reader_merge_samples::data_reader_merge_samples	(	std::vector< generic_data_reader *>	data_readers,
		bool	shuffle = `true`
	)

Here is the caller graph for this function:

◆ data_reader_merge_samples() [2/2]

lbann::data_reader_merge_samples::data_reader_merge_samples ( const data_reader_merge_samples & )

◆ ~data_reader_merge_samples()

lbann::data_reader_merge_samples::~data_reader_merge_samples ( )

override

Member Function Documentation

◆ compute_num_samples_psum()

size_t lbann::data_reader_merge_samples::compute_num_samples_psum ( )

protected

code common to both load() and load_using_data_store()

◆ copy()

data_reader_merge_samples* lbann::data_reader_merge_samples::copy ( ) const

inlineoverridevirtual

Implements lbann::generic_compound_data_reader.

Definition at line 48 of file data_reader_merge_samples.hpp.

Here is the call graph for this function:

◆ fetch_datum()

bool lbann::data_reader_merge_samples::fetch_datum	(	CPUMat &	X,
		int	data_id,
		int	mb_idx
	)

overrideprotectedvirtual

Fetch a single sample into a matrix.

Parameters

X	The matrix to load data into.
data_id	The index of the datum to fetch.
mb_idx	The index within the mini-batch.

Reimplemented from lbann::generic_data_reader.

Here is the caller graph for this function:

◆ fetch_label()

bool lbann::data_reader_merge_samples::fetch_label	(	CPUMat &	Y,
		int	data_id,
		int	mb_idx
	)

overrideprotectedvirtual

Fetch a single label into a matrix.

Parameters

Y	The matrix to load data into.
data_id	The index of the datum to fetch.
mb_idx	The index within the mini-batch.

Reimplemented from lbann::generic_data_reader.

Here is the caller graph for this function:

◆ fetch_response()

bool lbann::data_reader_merge_samples::fetch_response	(	CPUMat &	Y,
		int	data_id,
		int	mb_idx
	)

overrideprotectedvirtual

Fetch a single response into a matrix.

Parameters

Y	The matrix to load data into.
data_id	The index of the datum to fetch.
mb_idx	The index within the mini-batch.

Reimplemented from lbann::generic_data_reader.

Here is the caller graph for this function:

◆ get_data_dims()

const std::vector<El::Int> lbann::data_reader_merge_samples::get_data_dims ( ) const

inlineoverridevirtual

Get the dimensions of the data.

Reimplemented from lbann::generic_data_reader.

Definition at line 78 of file data_reader_merge_samples.hpp.

◆ get_linearized_data_size()

int lbann::data_reader_merge_samples::get_linearized_data_size ( ) const

inlineoverridevirtual

Get the linearized size (i.e. number of elements) in a sample.

Reimplemented from lbann::generic_data_reader.

Definition at line 66 of file data_reader_merge_samples.hpp.

◆ get_linearized_label_size()

int lbann::data_reader_merge_samples::get_linearized_label_size ( ) const

inlineoverridevirtual

Get the linearized size (i.e. number of elements) in a label.

Reimplemented from lbann::generic_data_reader.

Definition at line 70 of file data_reader_merge_samples.hpp.

◆ get_linearized_response_size()

int lbann::data_reader_merge_samples::get_linearized_response_size ( ) const

inlineoverridevirtual

Get the linearized size (i.e. number of elements) in a response.

Reimplemented from lbann::generic_data_reader.

Definition at line 74 of file data_reader_merge_samples.hpp.

◆ get_num_labels()

int lbann::data_reader_merge_samples::get_num_labels ( ) const

inlineoverridevirtual

Return the number of labels (classes) in this dataset.

This is called at the end of update; it permits data readers to perform actions that are specific to their data sets, for example, data_reader_jag_conduit_hdf5 has the 'primary' data reader bcast its shuffled indices to the other data readers. In general most data readers will probably not overide this method. It may also be called outside of update.

Reimplemented from lbann::generic_data_reader.

Definition at line 58 of file data_reader_merge_samples.hpp.

◆ get_num_responses()

int lbann::data_reader_merge_samples::get_num_responses ( ) const

inlineoverridevirtual

Return the number of responses in this dataset.

Reimplemented from lbann::generic_data_reader.

Definition at line 62 of file data_reader_merge_samples.hpp.

◆ get_num_samples_psum()

const std::vector<int>& lbann::data_reader_merge_samples::get_num_samples_psum ( )

inline

support for data store functionality

Definition at line 84 of file data_reader_merge_samples.hpp.

Here is the call graph for this function:

◆ get_type()

std::string lbann::data_reader_merge_samples::get_type ( ) const

inlineoverridevirtual

Return this data_reader's type

Implements lbann::generic_data_reader.

Definition at line 53 of file data_reader_merge_samples.hpp.

Here is the call graph for this function:

◆ load()

void lbann::data_reader_merge_samples::load ( )

overridevirtual

Load subsidiary data readers.

Implements lbann::generic_data_reader.

Here is the caller graph for this function:

◆ operator=()

data_reader_merge_samples& lbann::data_reader_merge_samples::operator= ( const data_reader_merge_samples & )

◆ sanity_check_for_consistency()

void lbann::data_reader_merge_samples::sanity_check_for_consistency	(	int	num_labels,
		int	data_size,
		int	label_size,
		const std::vector< El::Int > &	data_dims
	)

protected

code common to both load() and load_using_data_store()

◆ setup_indices()

void lbann::data_reader_merge_samples::setup_indices ( int num_samples )

protected

code common to both load() and load_using_data_store()

Member Data Documentation

◆ m_num_samples_psum

std::vector<int> lbann::data_reader_merge_samples::m_num_samples_psum

protected

Partial sums of the number of samples in each reader.

Definition at line 92 of file data_reader_merge_samples.hpp.

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

data_reader_merge_samples.hpp

Public Member Functions

Protected Member Functions

Protected Attributes

Additional Inherited Members

Detailed Description

Constructor & Destructor Documentation

◆ data_reader_merge_samples() [1/2]

◆ data_reader_merge_samples() [2/2]

◆ ~data_reader_merge_samples()

Member Function Documentation

◆ compute_num_samples_psum()

◆ copy()

◆ fetch_datum()

◆ fetch_label()

◆ fetch_response()

◆ get_data_dims()

◆ get_linearized_data_size()

◆ get_linearized_label_size()

◆ get_linearized_response_size()

◆ get_num_labels()

◆ get_num_responses()

◆ get_num_samples_psum()

◆ get_type()

◆ load()

◆ operator=()

◆ sanity_check_for_consistency()

◆ setup_indices()

Member Data Documentation

◆ m_num_samples_psum