gssl.classifiers package

Submodules

gssl.classifiers.CLGC module

Created on 23 de jun de 2020

@author: klaus

class gssl.classifiers.CLGC.CLGCClassifier(alpha=0.1, num_iter=None, useEstimatedFreq=True)

Bases: gssl.classifiers.classifier.GSSLClassifier

Constrained Local and Global Consistency Classifier. See [sousaconstrained].

__init__(alpha=0.1, num_iter=None, useEstimatedFreq=True)

Constructor for the LGC classifier.

Parameters

  • alpha (float) – A value between 0 and 1 (not inclusive) for alpha.

  • num_iter (float) – Optional. If not Ǹone, the number of iteration updates to perform

property alpha

\(\alpha\) such that \(\mu = \frac{1-\alpha}{\alpha}\) is the regularization factor multiplying the label fitting criterion. It is required that \(0 < \alpha < 1\).

fit(X, W, Y, labeledIndexes, hook=None)

Classifies the input data.

Parameters

  • X (NDArray[float].shape[N,D]) – Input matrix of N instances of dimension D.

  • W (NDArray[float].shape[N,N]) – The affinity matrix encoding the weighted edges.

  • Y (NDArray[float].shape[N,C]) – The initial belief matrix

  • hook (GSSLHook) – Optional. A hook to execute extra operations (e.g. plots) during the algorithm

Returns

An updated belief matrix.

Return type

NDArray[float].shape[N,C]

gssl.classifiers.GFHF module

Created on 27 de mar de 2019

@author: klaus

class gssl.classifiers.GFHF.GFHF(num_iter=None)

Bases: gssl.classifiers.classifier.GSSLClassifier

Label Propagation based on Gaussian Fields and Harmonic Functions. See [ZGL03].

__init__(num_iter=None)

Constructor for GFHF classifier.

Parameters

None –

fit(X, W, Y, labeledIndexes, hook=None)

Classifies the input data.

Parameters

  • X (NDArray[float].shape[N,D]) – Input matrix of N instances of dimension D.

  • W (NDArray[float].shape[N,N]) – The affinity matrix encoding the weighted edges.

  • Y (NDArray[float].shape[N,C]) – The initial belief matrix

  • hook (GSSLHook) – Optional. A hook to execute extra operations (e.g. plots) during the algorithm

Returns

An updated belief matrix.

Return type

NDArray[float].shape[N,C]

gssl.classifiers.GTAM module

Created on 27 de mar de 2019

@author: klaus

class gssl.classifiers.GTAM.GTAMClassifier(mu=99.0, num_iter=None, useEstimatedFreq=True, constantProp=False, know_true_freq=True, weigh_by_degree=True)

Bases: gssl.classifiers.classifier.GSSLClassifier

Classifier using Graph Transduction Through Alternating Minimization (GTAM - see [WJC08]).

__init__(mu=99.0, num_iter=None, useEstimatedFreq=True, constantProp=False, know_true_freq=True, weigh_by_degree=True)

” Constructor for GTAM classifier.

Parameters

  • mu (float) – a parameter determining the importance of the fitting term. Default is 99.0.

  • num_iter (int) – Optional. The number of iterations to run. The default behaviour makes it N iterations given a NDArray[float].shape[N,D] input matrix.

  • useEstimatedFreq (Union[bool,NDArray[C],None]) – If True, then use estimated class freq. to balance the propagation. If it is a float array, it uses that as the frequency. If None, assumes classes are equiprobable. Default is True.

  • useConstantProp (bool) – If True, then use try to maintain a constant proportion of labels in all iterations.

fit(X, W, Y, labeledIndexes, hook=None)

Classifies the input data.

Parameters

  • X (NDArray[float].shape[N,D]) – Input matrix of N instances of dimension D.

  • W (NDArray[float].shape[N,N]) – The affinity matrix encoding the weighted edges.

  • Y (NDArray[float].shape[N,C]) – The initial belief matrix

  • hook (GSSLHook) – Optional. A hook to execute extra operations (e.g. plots) during the algorithm

Returns

An updated belief matrix.

Return type

NDArray[float].shape[N,C]

gssl.classifiers.LGC module

class gssl.classifiers.LGC.LGCClassifier(alpha=0.1, num_iter=None)

Bases: gssl.classifiers.classifier.GSSLClassifier

Local and Global Consistency Classifier. See [ZBL+04].

__init__(alpha=0.1, num_iter=None)

Constructor for the LGC classifier.

Parameters

  • alpha (float) – A value between 0 and 1 (not inclusive) for alpha.

  • num_iter (float) – Optional. If not Ǹone, the number of iteration updates to perform

property alpha

\(\alpha\) such that \(\mu = \frac{1-\alpha}{\alpha}\) is the regularization factor multiplying the label fitting criterion. It is required that \(0 < \alpha < 1\).

fit(X, W, Y, labeledIndexes, hook=None)

Classifies the input data.

Parameters

  • X (NDArray[float].shape[N,D]) – Input matrix of N instances of dimension D.

  • W (NDArray[float].shape[N,N]) – The affinity matrix encoding the weighted edges.

  • Y (NDArray[float].shape[N,C]) – The initial belief matrix

  • hook (GSSLHook) – Optional. A hook to execute extra operations (e.g. plots) during the algorithm

Returns

An updated belief matrix.

Return type

NDArray[float].shape[N,C]

gssl.classifiers.LGC_tf module

Created on 23 de nov de 2019

@author: klaus

gssl.classifiers.LGC_tf.LGC_iter_TF(X, W, Y, labeledIndexes, alpha=0.1, num_iter=1000, hook=None)
gssl.classifiers.LGC_tf.convert_sparse_matrix_to_sparse_tensor(X, var_values=False)
gssl.classifiers.LGC_tf.gather(x, F)
gssl.classifiers.LGC_tf.get_P(TOTAL_ITER, ALPHA, S, F_0)
gssl.classifiers.LGC_tf.get_S_fromtensor(W)
gssl.classifiers.LGC_tf.repeat(x, n)
gssl.classifiers.LGC_tf.row_normalize(x)
gssl.classifiers.LGC_tf.update_F(TOTAL_ITER, ALPHA, S, F_0)

gssl.classifiers.LapEigLS module

Created on 27 de mar de 2019

@author: klaus

class gssl.classifiers.LapEigLS.LapEigLS(p=0.2)

Bases: gssl.classifiers.classifier.GSSLClassifier

Manifold Regularization with Laplacian Eigenmaps. Minimizes the Least Squares in a semi-supervised way by using a linear combination of the first \(p\) eigenfunctions. See [BN03].

__init__(p=0.2)

Constructor for LapEigLS classifier.

Parameters

p (Union[float,int]) – the labeled data. if float. Default is 0.2

fit(X, W, Y, labeledIndexes, hook=None)

Classifies the input data.

Parameters

  • X (NDArray[float].shape[N,D]) – Input matrix of N instances of dimension D.

  • W (NDArray[float].shape[N,N]) – The affinity matrix encoding the weighted edges.

  • Y (NDArray[float].shape[N,C]) – The initial belief matrix

  • hook (GSSLHook) – Optional. A hook to execute extra operations (e.g. plots) during the algorithm

Returns

An updated belief matrix.

Return type

NDArray[float].shape[N,C]

gssl.classifiers.RF module

class gssl.classifiers.RF.RFClassifier(n_estimators=10)

Bases: gssl.classifiers.classifier.GSSLClassifier

Supervised Random Forest classifier

__init__(n_estimators=10)

Constructor for the LGC classifier.

Parameters

alpha (float) – A value between 0 and 1 (not inclusive) for alpha.

fit(X, W, Y, labeledIndexes, hook=None)

Classifies the input data.

Parameters

  • X (NDArray[float].shape[N,D]) – Input matrix of N instances of dimension D.

  • W (NDArray[float].shape[N,N]) – The affinity matrix encoding the weighted edges.

  • Y (NDArray[float].shape[N,C]) – The initial belief matrix

  • hook (GSSLHook) – Optional. A hook to execute extra operations (e.g. plots) during the algorithm

Returns

An updated belief matrix.

Return type

NDArray[float].shape[N,C]

gssl.classifiers.SIIS module

Created on 27 de mar de 2019

@author: klaus

class gssl.classifiers.SIIS.SIISClassifier(m=100, beta=10, alpha=100, rho=1.2, max_iter=1000)

Bases: gssl.classifiers.classifier.GSSLClassifier

Label Propagation based on Semi-Supervised Learning under Inadequate and Incorrect Supervision (SIIS).

__init__(m=100, beta=10, alpha=100, rho=1.2, max_iter=1000)

Constructor for SIIS classifier.

Parameters

  • m (Union[float,int]) – The number of eigenvectors. It is given as either the absolute value (int), or a percentage of the labeled data (float).

  • alpha (float) – Value of parameter \(\alpha\) controlling the importance of the \(l_{2,1}\) norm smoothness term.

  • beta (float) – Value of parameter \(\beta\) controlling the importance of using smooth eigenfunctions within the already restricted eigenbasis.

  • max_iter (float) – Number of iterations.

  • rho (float) – Parameter for the convergence process. Default is 1.2. Should not be changed unless there is a good reason.

static edge_mat(W)
fit(X, W, Y, labeledIndexes, hook=None)

Classifies the input data.

Parameters

  • X (NDArray[float].shape[N,D]) – Input matrix of N instances of dimension D.

  • W (NDArray[float].shape[N,N]) – The affinity matrix encoding the weighted edges.

  • Y (NDArray[float].shape[N,C]) – The initial belief matrix

  • hook (GSSLHook) – Optional. A hook to execute extra operations (e.g. plots) during the algorithm

Returns

An updated belief matrix.

Return type

NDArray[float].shape[N,C]

gssl.classifiers.classifier module

class gssl.classifiers.classifier.GSSLClassifier

Bases: object

Skeleton class for GSSL Classifiers.

classmethod autohooks(fun)

Automatically calls the begin and end method of the hook. The classifier result is passed as the ‘Y’ argument at the end.

fit(X, W, Y, labeledIndexes, hook=None)

Classifies the input data.

Parameters

  • X (NDArray[float].shape[N,D]) – Input matrix of N instances of dimension D.

  • W (NDArray[float].shape[N,N]) – The affinity matrix encoding the weighted edges.

  • Y (NDArray[float].shape[N,C]) – The initial belief matrix

  • hook (GSSLHook) – Optional. A hook to execute extra operations (e.g. plots) during the algorithm

Returns

An updated belief matrix.

Return type

NDArray[float].shape[N,C]

Module contents