PercolatorTypes.h File Reference

#include <OpenMS/CONCEPT/Types.h>
#include <OpenMS/DATASTRUCTURES/TypeAliases.h>
#include <OpenMS/DATASTRUCTURES/StringUtils.h>
#include <string>
#include <vector>

Go to the source code of this file.

Classes
struct	RescoreInput
	Input to domain-agnostic Percolator::rescore. More...
struct	RescoreOutput
	Output from Percolator::rescore. Aligned 1:1 with the rows of the RescoreInput (RescoreInput::rowCount()). More...
struct	PercolatorModel
	Trained Percolator model: averaged SVM weights in raw feature space. More...

Namespaces
namespace	OpenMS
	Main OpenMS namespace.

---

Class Documentation

OpenMS::RescoreOutput

struct OpenMS::RescoreOutput

Output from Percolator::rescore. Aligned 1:1 with the rows of the RescoreInput (RescoreInput::rowCount()).

Class Members
vector< double >	peps	posterior error probability per row
vector< double >	peptide_peps
vector< double >	peptide_q_values	With Percolator parameter 'peptide_level' (and RescoreInput::peptide_keys): per row, the q-value and PEP of its peptide (key and label) at the peptide level; 1 for a row whose peptide has no PSM left after the target-decoy competition. Empty otherwise.
vector< double >	q_values	q-value per row
vector< double >	scores	SVM discriminant score per row.

OpenMS::PercolatorModel

struct OpenMS::PercolatorModel

Trained Percolator model: averaged SVM weights in raw feature space.

Produced by Percolator::train and consumed by Percolator::score. The weights are un-normalized: they are intended to multiply raw input features directly. The normalization transform learned by the SVM has already been folded into the weights and bias by Normalizer::unnormalizeweight(). Callers must therefore not normalize features before score(); doing so would apply the transform twice.

The raw SVM dot product for a row with feature vector f is raw = sum_j(f[j] * weights[j]) + weights[n_features] // bias last Percolator::score() applies a further FDR-based rescaling on top of this raw value to produce the final SVM discriminant reported in RescoreOutput.scores; see Percolator::score for the exact formula.

Class Members
StringList	feature_names	Feature column names. Must be non-empty and must match RescoreInput::feature_names positionally at score time. Any string value is permitted: the bias is stored in the header by saveModel(), so feature names carry no reserved meaning.
int	format_version = 1	Integer schema version for the on-disk format.
string	normalizer_type	"stdv" | "uni" | "none" — the normalizer used during training. Informational only; all three produce raw-space weights that score() can apply directly, since the normalization transform is already folded into the weights and bias. Recorded so that reproducibility tooling can identify the learner configuration that produced the model.
int	seed = 0	Random seed used during training. Informational.
vector< double >	weights	Size = n_features + 1. The last entry is the bias.