turi::topk_indexer Class Reference

#include <toolkits/feature_engineering/topk_indexer.hpp>

Public Member Functions
	topk_indexer (const size_t &_topk=(size_t) -1, const size_t &_threshold=1, const size_t &_max_threshold=(size_t) -1, const std::string _column_name="")
	topk_indexer (const topk_indexer &)=delete
void	initialize ()
void	insert_or_update (const flexible_type &value, size_t thread_idx=0, size_t count=1) GL_HOT
size_t	lookup (const flexible_type &value) const
size_t	lookup_counts (const flexible_type &value) const
void	finalize ()
flexible_type	inverse_lookup (size_t idx) const
std::vector< flexible_type >	get_values () const
size_t	size () const
size_t	get_version () const
void	save_impl (turi::oarchive &oarc) const
void	load_version (turi::iarchive &iarc, size_t version)

Detailed Description

Parallel top-k indexer for categorical variables (uses one-hot-encoding)

Note: This implementation is intended to be general and will be moved to some place more general later.

Construction

// Construct the indexer with the arguments. auto indexer = topk_indexer(10, 1, "column_name_for_error_messages"); indexer.initialize();

// Insert flexible types into the indexer for (const flexible_type& v: sa.range_iterator() { indexer.insert_or_update(v); }

// Finalize mapping (drops elements by frequency/threshold) indexer.finalize();

Lookups

size_t index = indexer.lookup(v); // Returns (size_t) -1 if not present.

size_t counts = indexer.lookup_counts(v); // Returns 0 if not present.

flexible_type v = indexer.inverse_lookup(1) // Fails if index doesn't exist.

Parallel construction

// Initialize indexer.initialize();

// Perform the indexing. in_parallel([&](size_t thread_idx, size_t num_threads) {

size_t start_idx = src_size * thread_idx / num_threads; size_t end_idx = src_size * (thread_idx + 1) / num_threads;

for (const flexible_type& v: sa.range_iterator(start_idx, end_idx) { indexer.insert_or_update(v, thread_id); }

// Finalize indexer.finalize();

Definition at line 71 of file topk_indexer.hpp.

Constructor & Destructor Documentation

topk_indexer() [1/2]

turi::topk_indexer::topk_indexer ( const size_t &  _topk = (size_t)-1, const size_t &  _threshold = 1, const size_t &  _max_threshold = (size_t) -1, const std::string  _column_name = ""  )

inline

Default constructor

Parameters

[in]	topk	Topk to retain (by counts)
[in]	threshold	Min count threshold to retain.
[in]	column_name	Column name for display.

Definition at line 83 of file topk_indexer.hpp.

topk_indexer() [2/2]

turi::topk_indexer::topk_indexer ( const topk_indexer &  )

delete

Copy constructor: Don't want to risk making copies of this.

Member Function Documentation

finalize()

void turi::topk_indexer::finalize

(

)

Finalize by dropping indices that dont meet

• Count requirement i.e count >= threshold.
• Topk requirement.

get_values()

std::vector<flexible_type> turi::topk_indexer::get_values ( ) const

inline

Returns the values (ordered by indices)

Definition at line 149 of file topk_indexer.hpp.

get_version()

size_t turi::topk_indexer::get_version

(

)

const

Returns the current version used for the serialization.

initialize()

void turi::topk_indexer::initialize

(

)

Initialize the index mapping and setup. Should be called before starting the map.

insert_or_update()

void turi::topk_indexer::insert_or_update	(	const flexible_type &	value,
		size_t	thread_idx = 0,
		size_t	count = 1
	)

Insert

Parameters

[in]	value	Flexible type.
[in]	thread_idx	Thread id (For parallel insertion).
[in]	count	Amount to increment for this value.

inverse_lookup()

flexible_type turi::topk_indexer::inverse_lookup

(

size_t

idx

)

const

Returns the "value" associated with the index.

Parameters

load_version()

void turi::topk_indexer::load_version	(	turi::iarchive &	iarc,
		size_t	version
	)

Load the object.

lookup()

size_t turi::topk_indexer::lookup

(

const flexible_type &

value

)

const

Returns the index associated with the value.

Parameters

[in]

value

Search for the value.

Returns

The index. (Returns size_t(-1) if not present).

lookup_counts()

size_t turi::topk_indexer::lookup_counts

(

const flexible_type &

value

)

const

Returns the counts associated with the value.

Parameters

[in]

value

Search for the value.

Returns

Counts (Returns 0 if not present).

save_impl()

void turi::topk_indexer::save_impl

(

turi::oarchive &

oarc

)

const

Serialize the object (save).

size()

size_t turi::topk_indexer::size ( ) const

inline

Returns the number of categorical variables.

Returns

Column size.

Definition at line 158 of file topk_indexer.hpp.

---

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

• toolkits/feature_engineering/topk_indexer.hpp