Feature Selection Toolbox FST4
UTIA · Czech Academy of Sciences
Loading...
Searching...
No Matches
fst4_cli::Pipeline Struct Reference

everything the run phase needs, assembled by the factories below More...

Public Member Functions

void invalidate_engine_clones () const
 Drops the cached per-worker criterion clones of every engine (see Candidate_Evaluator::invalidate_clones()). The threaded engine re-uses its clones across batches while the same criterion object is supplied, so once the ORIGINAL criterion's evaluation context has changed - the next trial's depth-0 split, a re-estimated filter model, a re-tuned classifier or regressor parameter - the clones must be re-made from it, or the workers keep evaluating in the old context: without this call a threaded trial protocol repeats its first trial in every trial. A no-op for the sequential engine.

Public Attributes

std::shared_ptr< ACCESSOR_MEM > da_mem
 concrete-side handle: initialize(), setSplittingDepth()
std::shared_ptr< DATAACCESSOR > da
 the type-erased handle every component sees
std::shared_ptr< CRITERION > crit
 counting decorator around the selected criterion
std::shared_ptr< CLASSIFIER > clf
 the wrapper criterion's classifier (empty for filters); --validate same reuses it
std::shared_ptr< REGRESSOR > regr
 the type-erased regressor handle (--validate same on the regression pipelines)
std::string regr_label
 its validation-entry label ("regressor" | "ridge")
std::shared_ptr< RWN > rwn
 concrete regressor handle (only where the concrete API is needed: the kernel-multiplier sweep)
std::shared_ptr< SVM > svm
 concrete SVM handle (parameter optimization needs the non-erased type)
std::shared_ptr< RBFSVM > rbfsvm
 concrete native RBF SVM handle (--rbfsvm-optimize needs the non-erased type)
std::shared_ptr< std::atomic< long long > > evals
 shared evaluation counter, incremented by the counting criterion decorator; feeds the "evaluations" stats field
std::function< bool(RETURNTYPE &, const std::shared_ptr< SUBSET > &)> split_eval
 re-evaluates a subset on the accessor's current split
std::function< void()> crit_reinit
 re-estimates a filter criterion's model on the current split (per-trial re-init, demo55-B); empty for wrappers
std::shared_ptr< SEARCH > search
 the constructed search algorithm
std::shared_ptr< TRACKER > tracker
 dupless-family handle (top_k source); null for stabileval
std::shared_ptr< TRACKER_REG > tracker_reg
 regularizer handle (optimize_within_margin)
std::shared_ptr< TRACKER_STABIL > tracker_stabil
 stability-evaluator handle (manual add per trial)
std::shared_ptr< TRACKER_STATS > tracker_stats
 feature-stats handle (compute_stats, DAF extraction)
std::shared_ptr< WEIGHTER_SAMPLED > sampled_stats
 sampled-evaluator statistics engine (the warm-up probe loop feeds it)
PENGINE sampled_engine
 the sampled evaluator's candidate-evaluation engine; the threaded warm-up loop reuses it so criterion clones persist into the search
std::vector< PENGINE > engines
 every candidate-evaluation engine built for this run (the search's and the step evaluator's), registered by make_engine() so that invalidate_engine_clones() reaches them all

Detailed Description

everything the run phase needs, assembled by the factories below


The documentation for this struct was generated from the following file:
Copyright © 2000–present ÚTIA, Institute of Information Theory and Automation of the Czech Academy of Sciences · Contacts · This site sets no cookies and uses no analytics.