Noise filters and prototype selection¶

TomekLinkRemoval¶

API¶

class smote_variants.TomekLinkRemoval(strategy='remove_majority', n_jobs=1)[source]¶

__init__(strategy='remove_majority', n_jobs=1)[source]¶

Constructor of the noise filter.

Parameters:	strategy (str) – noise removal strategy: ‘remove_majority’/’remove_both’ n_jobs (int) – number of jobs

remove_noise(X, y)[source]¶

Removes noise from dataset

Parameters:	X (np.matrix) – features y (np.array) – target labels
Returns:	dataset after noise removal
Return type:	np.matrix, np.array

Example¶

>>> noise_filter= smote_variants.TomekLinkRemoval()
>>> X_samp, y_samp= noise_filter.remove_noise(X, y)

Tomek link removal

References:

BibTex:

@article{smoteNoise0,
         author = {Batista, Gustavo E. A. P. A. and Prati,
                    Ronaldo C. and Monard, Maria Carolina},
         title = {A Study of the Behavior of Several Methods for
                    Balancing Machine Learning Training Data},
         journal = {SIGKDD Explor. Newsl.},
         issue_date = {June 2004},
         volume = {6},
         number = {1},
         month = jun,
         year = {2004},
         issn = {1931-0145},
         pages = {20--29},
         numpages = {10},
         url = {http://doi.acm.org/10.1145/1007730.1007735},
         doi = {10.1145/1007730.1007735},
         acmid = {1007735},
         publisher = {ACM},
         address = {New York, NY, USA}
        }

CondensedNearestNeighbors¶

API¶

class smote_variants.CondensedNearestNeighbors(n_jobs=1)[source]¶

__init__(n_jobs=1)[source]¶

Constructor of the noise removing object

Parameters:	n_jobs (int) – number of jobs

remove_noise(X, y)[source]¶

Removes noise from dataset

Parameters:	X (np.matrix) – features y (np.array) – target labels
Returns:	dataset after noise removal
Return type:	np.matrix, np.array

Example¶

>>> noise_filter= smote_variants.CondensedNearestNeighbors()
>>> X_samp, y_samp= noise_filter.remove_noise(X, y)

Condensed nearest neighbors

References:

BibTex:

@ARTICLE{condensed_nn,
            author={Hart, P.},
            journal={IEEE Transactions on Information Theory},
            title={The condensed nearest neighbor rule (Corresp.)},
            year={1968},
            volume={14},
            number={3},
            pages={515-516},
            keywords={Pattern classification},
            doi={10.1109/TIT.1968.1054155},
            ISSN={0018-9448},
            month={May}}

OneSidedSelection¶

API¶

class smote_variants.OneSidedSelection(n_jobs=1)[source]¶

__init__(n_jobs=1)[source]¶

Constructor of the noise removal object

Parameters:	n_jobs (int) – number of jobs

remove_noise(X, y)[source]¶

Removes noise

Parameters:	X (np.matrix) – features y (np.array) – target labels
Returns:	cleaned features and target labels
Return type:	np.matrix, np.array

Example¶

>>> noise_filter= smote_variants.OneSidedSelection()
>>> X_samp, y_samp= noise_filter.remove_noise(X, y)

References:

BibTex:

@article{smoteNoise0,
         author = {Batista, Gustavo E. A. P. A. and Prati,
                    Ronaldo C. and Monard, Maria Carolina},
         title = {A Study of the Behavior of Several Methods
                    for Balancing Machine Learning Training Data},
         journal = {SIGKDD Explor. Newsl.},
         issue_date = {June 2004},
         volume = {6},
         number = {1},
         month = jun,
         year = {2004},
         issn = {1931-0145},
         pages = {20--29},
         numpages = {10},
         url = {http://doi.acm.org/10.1145/1007730.1007735},
         doi = {10.1145/1007730.1007735},
         acmid = {1007735},
         publisher = {ACM},
         address = {New York, NY, USA}
        }

CNNTomekLinks¶

API¶

class smote_variants.CNNTomekLinks(n_jobs=1)[source]¶

__init__(n_jobs=1)[source]¶

Constructor of the noise removal object

Parameters:	n_jobs (int) – number of parallel jobs

remove_noise(X, y)[source]¶

Removes noise

Parameters:	X (np.matrix) – features y (np.array) – target labels
Returns:	cleaned features and target labels
Return type:	np.matrix, np.array

Example¶

>>> noise_filter= smote_variants.CNNTomekLinks()
>>> X_samp, y_samp= noise_filter.remove_noise(X, y)

References:

BibTex:

@article{smoteNoise0,
         author = {Batista, Gustavo E. A. P. A. and Prati,
                    Ronaldo C. and Monard, Maria Carolina},
         title = {A Study of the Behavior of Several Methods
                    for Balancing Machine Learning Training Data},
         journal = {SIGKDD Explor. Newsl.},
         issue_date = {June 2004},
         volume = {6},
         number = {1},
         month = jun,
         year = {2004},
         issn = {1931-0145},
         pages = {20--29},
         numpages = {10},
         url = {http://doi.acm.org/10.1145/1007730.1007735},
         doi = {10.1145/1007730.1007735},
         acmid = {1007735},
         publisher = {ACM},
         address = {New York, NY, USA}
        }

NeighborhoodCleaningRule¶

API¶

class smote_variants.NeighborhoodCleaningRule(n_jobs=1)[source]¶

__init__(n_jobs=1)[source]¶

Constructor of the noise removal object

Parameters:	n_jobs (int) – number of parallel jobs

remove_noise(X, y)[source]¶

Removes noise

Parameters:	X (np.matrix) – features y (np.array) – target labels
Returns:	cleaned features and target labels
Return type:	np.matrix, np.array

Example¶

>>> noise_filter= smote_variants.NeighborhoodCleaningRule()
>>> X_samp, y_samp= noise_filter.remove_noise(X, y)

References:

BibTex:

@article{smoteNoise0,
         author = {Batista, Gustavo E. A. P. A. and Prati,
                    Ronaldo C. and Monard, Maria Carolina},
         title = {A Study of the Behavior of Several Methods for
                    Balancing Machine Learning Training Data},
         journal = {SIGKDD Explor. Newsl.},
         issue_date = {June 2004},
         volume = {6},
         number = {1},
         month = jun,
         year = {2004},
         issn = {1931-0145},
         pages = {20--29},
         numpages = {10},
         url = {http://doi.acm.org/10.1145/1007730.1007735},
         doi = {10.1145/1007730.1007735},
         acmid = {1007735},
         publisher = {ACM},
         address = {New York, NY, USA}
        }

EditedNearestNeighbors¶

API¶

class smote_variants.EditedNearestNeighbors(remove='both', n_jobs=1)[source]¶

__init__(remove='both', n_jobs=1)[source]¶

Constructor of the noise removal object

Parameters:	remove (str) – class to remove from ‘both’/’min’/’maj’ n_jobs (int) – number of parallel jobs

get_params()[source]¶

Get noise removal parameters

Returns:	dictionary of parameters
Return type:	dict

remove_noise(X, y)[source]¶

Removes noise

Parameters:	X (np.matrix) – features y (np.array) – target labels
Returns:	cleaned features and target labels
Return type:	np.matrix, np.array

Example¶

>>> noise_filter= smote_variants.EditedNearestNeighbors()
>>> X_samp, y_samp= noise_filter.remove_noise(X, y)

References:

BibTex:

@article{smoteNoise0,
         author = {Batista, Gustavo E. A. P. A. and Prati,
                    Ronaldo C. and Monard, Maria Carolina},
         title = {A Study of the Behavior of Several Methods for
                    Balancing Machine Learning Training Data},
         journal = {SIGKDD Explor. Newsl.},
         issue_date = {June 2004},
         volume = {6},
         number = {1},
         month = jun,
         year = {2004},
         issn = {1931-0145},
         pages = {20--29},
         numpages = {10},
         url = {http://doi.acm.org/10.1145/1007730.1007735},
         doi = {10.1145/1007730.1007735},
         acmid = {1007735},
         publisher = {ACM},
         address = {New York, NY, USA}
        }