classifier_wrapper

Module: missions.nodes.meta.classifier_wrapper

Operate on classifiers

Inheritance diagram for pySPACE.missions.nodes.meta.classifier_wrapper:

Class Summary

SplitClassifierLayerNode(classifier[, store])	Split the overrepresented class in the training set for multiple training.
SVMComplexityLayerNode(classifier[, store, eps])	Calculate the minimal complexity, where the soft margin is inactive

Classes

SplitClassifierLayerNode

class pySPACE.missions.nodes.meta.classifier_wrapper.SplitClassifierLayerNode(classifier, store=False, *args, **kwargs)

Bases: pySPACE.missions.nodes.base_node.BaseNode

Split the overrepresented class in the training set for multiple training.

The node trains several classifiers with this splits such that both classes are nearly equally distributed.

Output is a set of predictions in a prediction vector. An ensemble node should follow to combine these classifications.

Parameter

Classifier:

Classifier to be split. Notation is as usual as in the YAML file.

Maybe this will be changed later on.

Exemplary Call

- 
    node : Split_Classifier
    parameters :
        classifier: 
                -
                    node : LibSVM_Classifier
                    parameters :
                        complexity : 1
                        weight : [1,3]
                        debug : False
                        store : False
                        class_labels : ['Standard', 'Target']

Author:

Mari Krell (mario.krell@dfki.de)

POSSIBLE NODE NAMES:  

• Split_Classifier
• SplitClassifierLayerNode
• SplitClassifierLayer

POSSIBLE INPUT TYPES:  

• FeatureVector

Class Components Summary

_execute(data)	Process the data through the internal nodes.
_stop_training([debug])
_train(data, class_label)	collects the data for training
get_output_type(input_type[, as_string])	overwritten method from BaseNode
input_types
is_supervised()	Returns whether this node requires supervised training
is_trainable()	Returns whether this node is trainable.
reset()	Reset the state to the clean state it had after its initialization

input_types = ['FeatureVector']

__init__(classifier, store=False, *args, **kwargs)

is_trainable()

Returns whether this node is trainable.

is_supervised()

Returns whether this node requires supervised training

reset()

Reset the state to the clean state it had after its initialization

_execute(data)

Process the data through the internal nodes.

_train(data, class_label)

collects the data for training

It is assumed that the class_label parameter contains information about the true class the data belongs to

_stop_training(debug=False)

get_output_type(input_type, as_string=True)

overwritten method from BaseNode

returns PredictionVector(as string or class) since this is the only possible output of the current node

SVMComplexityLayerNode

class pySPACE.missions.nodes.meta.classifier_wrapper.SVMComplexityLayerNode(classifier, store=False, eps=0.001, *args, **kwargs)

Bases: pySPACE.missions.nodes.meta.classifier_wrapper.SplitClassifierLayerNode

Calculate the minimal complexity, where the soft margin is inactive

This node uses nested intervals and a tolerance variable is used to define when the accuracy is high enough and the slack variables are small enough. This was necessary because the libsvm classifier gives no exact solution and the slack variables may be never zero.

Output is the prediction of the given classifier with the given complexity multiplied by the found complexity. Wrapper around a classifier. The result should be analyzed with the classification performance sink node.

Parameter

Classifier:

SVM Classifier to be analysed. Notation is as usual as in the YAML file.

Maybe this will be changed later on.

Exemplary Call

- 
    node : Get_Complexity
    parameters :
        classifier: 
                -
                    node : LibSVM_Classifier
                    parameters :
                        complexity : 1
                        weight : [1,3]
                        debug : False
                        store : False
                        class_labels : ['Standard', 'Target']

Author:

Mario Krell (mario.krell@dfki.de)

POSSIBLE NODE NAMES:  

• SVMComplexityLayer
• Get_Complexity
• SVMComplexityLayerNode

POSSIBLE INPUT TYPES:  

• FeatureVector

Class Components Summary

_execute(data)	Process the data through the internal nodes.
_stop_training([debug])
_train(data, class_label)	It is assumed that the class_label parameter
reset()	Reset the state to the clean state it had after its initialization

__init__(classifier, store=False, eps=0.001, *args, **kwargs)

reset()

Reset the state to the clean state it had after its initialization

_execute(data)

Process the data through the internal nodes.

_train(data, class_label)

It is assumed that the class_label parameter contains information about the true class the data belongs to

_stop_training(debug=False)