feature_normalization

Module: missions.nodes.postprocessing.feature_normalization

Normalize FeatureVector

Inheritance diagram for pySPACE.missions.nodes.postprocessing.feature_normalization:

Class Summary

InconsistentFeatureVectorsException
FeatureNormalizationNode([load_path])	General node for Feature Normalization
OutlierFeatureNormalizationNode([...])	Map the feature vectors of the training set to the range [0,1]^n
GaussianFeatureNormalizationNode(\*\*kwargs)	Transform the features, such that they have zero mean and variance one
HistogramFeatureNormalizationNode([load_path])	Transform the features, such that they have zero mean in the main bit in the histogram and variance one on that bit.
EuclideanFeatureNormalizationNode([...])	Normalize feature vectors to Euclidean norm with respect to dimensions
InfinityNormFeatureNormalizationNode(\*\*kwargs)	Normalize feature vectors with infinity norm to [-1,1]

Classes

InconsistentFeatureVectorsException

class pySPACE.missions.nodes.postprocessing.feature_normalization.InconsistentFeatureVectorsException

Bases: exceptions.Exception

__weakref__

list of weak references to the object (if defined)

FeatureNormalizationNode

class pySPACE.missions.nodes.postprocessing.feature_normalization.FeatureNormalizationNode(load_path=None, **kwargs)

Bases: pySPACE.missions.nodes.base_node.BaseNode

General node for Feature Normalization

The node should simply shift the data with the translation variable and afterwards scale it with the mult variable.

This transformation can be loaded and stored and given to visualization tools.

When used as a standalone node, loading a transformation is mandatory because the translation and mult variables are otherwise not available.

Parameter

load_path:

An absolute path from which the normalization vectors are loaded. If not specified, these vectors are learned from the training data.

(optional, default: None)

Exemplary Call

-
    node : FeatureNormalizationNode
    parameters :
        load_path: "/Users/mustermann/proj/examples/FN.pickle"

Warning

This base node only works alone, when load_path is specified.

Input:

FeatureVector

Output:

FeatureVector

Author:

Mario Krell (mario.krell@dfki.de)

Created:

2012/03/28

POSSIBLE NODE NAMES:  

• FeatureNormalizationNode
• FeatureNormalization

POSSIBLE INPUT TYPES:  

• FeatureVector

Class Components Summary

_execute(data)	Normalizes the feature vector data.
_train(data)	Collects the values each feature takes on in the training set.
collect_data(data)
extract_feature_names(data)
get_own_transformation([sample])
input_types
is_trainable()
store_state(result_dir[, index])	Stores transformation and feature names in the given directory result_dir

__init__(load_path=None, **kwargs)

is_trainable()

get_own_transformation(sample=None)

store_state(result_dir, index=None)

Stores transformation and feature names in the given directory result_dir

_train(data)

Collects the values each feature takes on in the training set.

extract_feature_names(data)

_execute(data)

Normalizes the feature vector data.

Normalizes the feature vector data by subtracting the translation variable and scaling it with mult.

collect_data(data)

input_types = ['FeatureVector']

OutlierFeatureNormalizationNode

class pySPACE.missions.nodes.postprocessing.feature_normalization.OutlierFeatureNormalizationNode(outlier_percentage=0, **kwargs)

Bases: pySPACE.missions.nodes.postprocessing.feature_normalization.FeatureNormalizationNode

Map the feature vectors of the training set to the range [0,1]^n

A class that normalizes each dimension of the feature vector so that an upper boundary value (learned from in the training set) is mapped to 1, and a lower boundary value to 0. All other values are linearly interpolated. Optionally, one can specify an outlier_percentage that determines which ratio of the training data is considered to be a potential outlier. outlier_percentage/2 samples are allowed to be larger than the determined upper boundary, and outlier_percentage/2 samples are allowed to be smaller than the determined lower boundary.

Parameters

outlier_percentage:  

The percentage of training instances that are potential outliers.

(optional, default: 0)

Exemplary Call

-
    node : OutlierFeatureNormalization
    parameters :
        outlier_percentage : 10

Author:

Jan Hendrik Metzen (jhm@informatik.uni-bremen.de)

Created:

??

Revised (1):

2009/07/16

Revised (2):

2009/09/03

POSSIBLE NODE NAMES:  

• FN
• Outlier_Feature_Normalization
• Feature_Normalization
• OutlierFeatureNormalizationNode
• O_FN
• OutlierFeatureNormalization

POSSIBLE INPUT TYPES:  

• FeatureVector

Class Components Summary

__hyperparameters
_stop_training()	Computes the upper and lower boundary for normalization.
collect_data(data)

__init__(outlier_percentage=0, **kwargs)

collect_data(data)

_stop_training()

Computes the upper and lower boundary for normalization.

For this computation, the largest and smallest outlier_percentage/2 examples for each feature dimension are ignored. The smallest and largest remaining example are used as lower and upper boundary.

__hyperparameters = set([NoOptimizationParameter<kwargs_warning>, NoOptimizationParameter<dtype>, NoOptimizationParameter<output_dim>, UniformParameter<outlier_percentage>, NoOptimizationParameter<retrain>, NoOptimizationParameter<input_dim>, NoOptimizationParameter<store>])

GaussianFeatureNormalizationNode

class pySPACE.missions.nodes.postprocessing.feature_normalization.GaussianFeatureNormalizationNode(**kwargs)

Bases: pySPACE.missions.nodes.postprocessing.feature_normalization.FeatureNormalizationNode

Transform the features, such that they have zero mean and variance one

A class that normalizes each dimension of the feature vector so that it has zero mean and variance one. The relevant values are learned from the training set.

Exemplary Call

-
    node : Gaussian_Feature_Normalization

Author:

Mario Krell (Mario.Krell@dfki.de)

Created:

2011/04/15

POSSIBLE NODE NAMES:  

• GaussianFeatureNormalizationNode
• GaussianFeatureNormalization
• G_FN
• Gaussian_Feature_Normalization

POSSIBLE INPUT TYPES:  

• FeatureVector

Class Components Summary

_inc_train(data[, class_label])
_stop_training()	Computes mean and std deviation of each feature
_train(data)

__init__(**kwargs)

_stop_training()

Computes mean and std deviation of each feature

_train(data)

_inc_train(data, class_label=None)

HistogramFeatureNormalizationNode

class pySPACE.missions.nodes.postprocessing.feature_normalization.HistogramFeatureNormalizationNode(load_path=None, **kwargs)

Bases: pySPACE.missions.nodes.postprocessing.feature_normalization.FeatureNormalizationNode

Transform the features, such that they have zero mean in the main bit in the histogram and variance one on that bit.

The relevant values are learned from the training set.

Exemplary Call

-
    node : Histogram_Feature_Normalization

Author:

Mario Krell (Mario.Krell@dfki.de)

Created:

2011/04/15

POSSIBLE NODE NAMES:  

• HistogramFeatureNormalization
• HistogramFeatureNormalizationNode
• H_FN
• Histogram_Feature_Normalization

POSSIBLE INPUT TYPES:  

• FeatureVector

Class Components Summary

_stop_training()

Computes mean and std deviation of each feature

_stop_training()

Computes mean and std deviation of each feature

EuclideanFeatureNormalizationNode

class pySPACE.missions.nodes.postprocessing.feature_normalization.EuclideanFeatureNormalizationNode(dimension_scale=False, **kwargs)

Bases: pySPACE.missions.nodes.base_node.BaseNode

Normalize feature vectors to Euclidean norm with respect to dimensions

Parameters

dimension_scale:  

Scale the output to ||x|| * dim(x) (to get bigger values)

(optional, default: False)

Exemplary Call

-
    node : Euclidean_Feature_Normalization
    parameters :
        dimension_scale : True

Author:

Mario Krell (Mario.Krell@dfki.de)

Created:

2011/04/15

POSSIBLE NODE NAMES:  

• Euclidean_Feature_Normalization
• EuclideanFeatureNormalizationNode
• EuclideanFeatureNormalization
• E_FN

POSSIBLE INPUT TYPES:  

• FeatureVector

Class Components Summary

__hyperparameters
_execute(data)	Normalizes the samples vector to norm one
input_types
store_state(result_dir[, index])	Stores this node in the given directory result_dir

__init__(dimension_scale=False, **kwargs)

_execute(data)

Normalizes the samples vector to norm one

store_state(result_dir, index=None)

Stores this node in the given directory result_dir

__hyperparameters = set([BooleanParameter<dimension_scale>, NoOptimizationParameter<kwargs_warning>, NoOptimizationParameter<dtype>, NoOptimizationParameter<output_dim>, NoOptimizationParameter<retrain>, NoOptimizationParameter<input_dim>, NoOptimizationParameter<store>])

input_types = ['FeatureVector']

InfinityNormFeatureNormalizationNode

class pySPACE.missions.nodes.postprocessing.feature_normalization.InfinityNormFeatureNormalizationNode(**kwargs)

Bases: pySPACE.missions.nodes.base_node.BaseNode

Normalize feature vectors with infinity norm to [-1,1]

Parameters

Exemplary Call

-
    node : I_FN

Author:

Mario Krell and Marc Tabie (Mario.Krell and Marc.Tabie@dfki.de)

Created:

2012/07/16

POSSIBLE NODE NAMES:  

• InfinityNormFeatureNormalizationNode
• I_FN
• InfinityNormFeatureNormalization

POSSIBLE INPUT TYPES:  

• FeatureVector

Class Components Summary

_execute(data)	Normalizes the samples vector to inf norm one
input_types

__init__(**kwargs)

_execute(data)

Normalizes the samples vector to inf norm one

input_types = ['FeatureVector']