# Classifiers implemented in TRIOSlib¶

All classifiers in TRIOSlib inherit from trios.Classifier, documented below. Classifiers should be interchangeable and typically do not depend on a specific FeatureExtractor to work. bla

These methods are not usually directly called in scripts. Instances of WOperator automatically call these methods during its training and application phase.

class trios.WOperator.Classifier

Classifies patterns extracted from the images. This is an abstract class.

__init__

Classifiers have two basic attributes: minimize and ordered.

• if minimize is True then WOperator modifies all labels in the training set such that train receives a training set with only consistent patterns (no $$x_i = x_j$$ and $$y_i \neq y_j$$).
• if ordered is True train receives as input a tuple $$(X, y)$$, where $$X$$ contains input patterns in its rows and $$y$$ contains the labels. If ordered is False train receives a dictionary with patterns and keys and a dictionary with the frequency of each output as values.
apply()

Override this method with the application procedure for a single pattern.

apply_batch()

Classifies a batch of patterns. Each one is stored on the rows of fmatrix.

Override this method if the classifier can do batch classification faster than classifiyng each pattern individually in a loop.

partial_train()

Executes one training iteration using inputs X and labels y.

train()

Each classification method must override this method with its training procedure.

The following classifiers are implemented in TRIOSlib:

 trios.classifiers.isi.ISI([win]) description here