activeSVC 4.0.1

Active feature selection method with support vector classifier.

activeSVC

ActiveSVC selects features for large matrix data with reduced computational complexity or limited data acquisition. It approaches Sequential Feature Selection through an active learning strategy with a support vector machine classifier. At each round of iteration, the procedure analyzes only the samples that classify poorly with the current feature set, and the procedure extends the feature set by identifying features within incorrectly classified samples that will maximally shift the classification margin. There are two strategy, min_complexity and min_acquisition. Min_complexity strategy tends to use less samples each iteration while min_acquisition strategy tends to re-use samples used in previous iterations to minimize the total samples we acquired during the procedure.

Why is activeSVC better than other feature selection methods?

• Easy to use
• Good for large datasets
• Reduce Memory
• Reduce computational complexity
• Minimize the data size we need

Usage

ActiveSVC processes a datasets with training set and test set and returns the features selected, training accuracy, test accuracy, training mean squared error, test mean squared error, the number of samples acquired after every features are selected. We highly recommend to do l2-normalization for each sample before activeSVC to improve accuracy and speed up model training.

Requires

numpy, random, math, os, time, multiprocessing, sklearn, spcipy

Import

from activeSVC import min_complexity
from activeSVC import min_acquisition
from activeSVC import min_complexity_cv
from activeSVC import min_acquisition_cv
from activeSVC import min_complexity_h5py
from activeSVC import min_acquisition_h5py

Function

• min_complexity: fix SVM parameters for each loop
• min_acquisition: fix SVM parameters for each loop
• min_complexity_cv: Every SVMs trained during the procedure are the best estimator by cross validation on parameters "C" and "tol".
• min_acquisition_cv: Every SVMs trained during the procedure are the best estimator by cross validation on parameters "C" and "tol".
• min_complexity_h5py: only load part of data matrix of selected features and samples into memory instead of loading the entire dataset to save memory usage. The data should be h5py file.
• min_acquisition_h5py: only load part of data matrix of selected features and samples into memory instead of loading the entire dataset to save memory usage. The data should be h5py file.

min_complexity

Parameters

X_train: {ndarray, sparse matrix} of shape {n_samples_X, n_features}
        Input data of training set.
y_train: ndarray of shape {n_samples_X,}
        Input classification labels of training set.
X_test: {ndarray, sparse matrix} of shape {n_samples_X, n_features}
        Input data of test set.
y_test: ndarray of shape {n_samples_X,}
        Input classification labels of test set.
num_features: integer
        The total number of features to select.
num_samples: integer
        The number of samples to use in each iteration (for each feature).
init_features: integer, default=1
        The number of features to select in the first iteration.
init_samples: integer, default=None
        The number of samples to use in the first iteration.
balance: bool, default=False
        Balance samples of each classes when sampling misclassified samples at each 
        iteration or randomly sample misclassified samples.
penalty: {‘l1’, ‘l2’}, default=’l2’
        Specifies the norm used in the penalization. The ‘l2’ penalty is the 
        standard used in SVC. The ‘l1’ leads to sparse weight.
loss: {‘hinge’, ‘squared_hinge’}, default=’squared_hinge’
        Specifies the loss function for each SVC to train. ‘hinge’ is the standard 
        SVM loss while ‘squared_hinge’ is the square of the hinge loss. 
        The combination of penalty='l1' and loss='hinge' is not supported.
dual: bool, default=True
        Select the algorithm to either solve the dual or primal optimization 
        problem for each SVC. Prefer dual=False when n_samples > n_features.
tol: float, default=1e-4
        Tolerance for stopping criteria for each SVC.
C: float, default=1.0
        Regularization parameter for each SVC. The strength of the regularization 
        is inversely proportional to C. Must be strictly positive.
fit_intercept: bool, default=True
        Whether to calculate the intercept for each SVC. If set to false, no 
        intercept will be used in calculations (i.e. data is already centered).
intercept_scaling: float, default=1
        When self.fit_intercept is True, instance vector x becomes 
        [x, self.intercept_scaling], i.e. a “synthetic” feature with constant value 
        equals to intercept_scaling is appended to the instance vector. The intercept 
        becomes intercept_scaling * synthetic feature weight Note! the synthetic 
        feature weight is subject to l1/l2 regularization as all other features. 
        To lessen the effect of regularization on synthetic feature weight 
        (and therefore on the intercept) intercept_scaling has to be increased.
class_weight: dict or ‘balanced’, default=None
        Set the parameter C of class i to class_weight[i]*C for SVC. If not given, 
        all classes are supposed to have weight one. The “balanced” mode uses the 
        values of y to automatically adjust weights inversely proportional to class 
        frequencies in the input data as n_samples / (n_classes * np.bincount(y)).
random_state: int, RandomState instance or None, default=None
        Controls the pseudo random number generation for shuffling data for the dual 
        coordinate descent (if dual=True). When dual=False underlying implementation 
        of LinearSVC is not random and random_state has no effect on the results. 
        Pass an int for reproducible output across multiple function calls.
max_iter: int, default=1000
        The maximum number of iterations to be run for each SVC.

Return

feature_selected: list of integer
        The sequence of features selected.
num_samples_list: list of integer
        The number of unique samples acquired totally after every features are selected.
train_errors: list of float
        Mean squared error of training set after every features are selected.
test_errors: list of float
        Mean squared error of test set after every features are selected.
train_accuracy: list of float
        Classification accuracy of training set after every features are selected.
test_accuracy: list of float
        Classification accuracy of test set after every features are selected.
step_times: list of float
        The run time of each iteration. Unit is second. 

min_acquisition

Parameters

X_train: {ndarray, sparse matrix} of shape {n_samples_X, n_features}
        Input data of training set.
y_train: ndarray of shape {n_samples_X,}
        Input classification labels of training set.
X_test: {ndarray, sparse matrix} of shape {n_samples_X, n_features}
        Input data of test set.
y_test: ndarray of shape {n_samples_X,}
        Input classification labels of test set.
num_features: integer
        The total number of features to select.
num_samples: integer
        The number of misclassified samples randomly sampled, which are taken union with 
        samples already acquired before. The union of samples are used for next ietration.
init_features: integer, default=1
        The number of features to select in the first iteration.
init_samples: integer, default=None
        The number of samples to use in the first iteration.
balance: bool, default=False
        Balance samples of each classes when sampling misclassified samples at each 
        iteration or randomly sample misclassified samples.
penalty: {‘l1’, ‘l2’}, default=’l2’
        Specifies the norm used in the penalization. The ‘l2’ penalty is the 
        standard used in SVC. The ‘l1’ leads to sparse weight.
loss: {‘hinge’, ‘squared_hinge’}, default=’squared_hinge’
        Specifies the loss function for each SVC to train. ‘hinge’ is the standard 
        SVM loss while ‘squared_hinge’ is the square of the hinge loss. 
        The combination of penalty='l1' and loss='hinge' is not supported.
dual: bool, default=True
        Select the algorithm to either solve the dual or primal optimization 
        problem for each SVC. Prefer dual=False when n_samples > n_features.
tol: float, default=1e-4
        Tolerance for stopping criteria for each SVC.
C: float, default=1.0
        Regularization parameter for each SVC. The strength of the regularization 
        is inversely proportional to C. Must be strictly positive.
fit_intercept: bool, default=True
        Whether to calculate the intercept for each SVC. If set to false, no 
        intercept will be used in calculations (i.e. data is already centered).
intercept_scaling: float, default=1
        When self.fit_intercept is True, instance vector x becomes 
        [x, self.intercept_scaling], i.e. a “synthetic” feature with constant value 
        equals to intercept_scaling is appended to the instance vector. The intercept 
        becomes intercept_scaling * synthetic feature weight Note! the synthetic 
        feature weight is subject to l1/l2 regularization as all other features. 
        To lessen the effect of regularization on synthetic feature weight 
        (and therefore on the intercept) intercept_scaling has to be increased.
class_weight: dict or ‘balanced’, default=None
        Set the parameter C of class i to class_weight[i]*C for SVC. If not given, 
        all classes are supposed to have weight one. The “balanced” mode uses the 
        values of y to automatically adjust weights inversely proportional to class 
        frequencies in the input data as n_samples / (n_classes * np.bincount(y)).
random_state: int, RandomState instance or None, default=None
        Controls the pseudo random number generation for shuffling data for the dual 
        coordinate descent (if dual=True). When dual=False underlying implementation 
        of LinearSVC is not random and random_state has no effect on the results. 
        Pass an int for reproducible output across multiple function calls.
max_iter: int, default=1000
        The maximum number of iterations to be run for each SVC.

Return

feature_selected: list of integer
        The sequence of features selected.
num_samples_list: list of integer
        The number of unique samples acquired totally after every features are selected.
samples_global: list of integer
        The indices of samples that are acquired.
train_errors: list of float
        Mean squared error of training set after every features are selected.
test_errors: list of float
        Mean squared error of test set after every features are selected.
train_accuracy: list of float
        Classification accuracy of training set after every features are selected.
test_accuracy: list of float
        Classification accuracy of test set after every features are selected.
step_times: list of float
        The run time of each iteration. Unit is second. 

min_complexity_cv

Parameters

X_train: {ndarray, sparse matrix} of shape {n_samples_X, n_features}
        Input data of training set.
y_train: ndarray of shape {n_samples_X,}
        Input classification labels of training set.
X_test: {ndarray, sparse matrix} of shape {n_samples_X, n_features}
        Input data of test set.
y_test: ndarray of shape {n_samples_X,}
        Input classification labels of test set.
num_features: integer
        The total number of features to select.
num_samples: integer
        The number of samples to use in each iteration (for each feature).
init_features: integer, default=1
        The number of features to select in the first iteration.
init_samples: integer, default=None
        The number of samples to use in the first iteration.
balance: bool, default=False
        Balance samples of each classes when sampling misclassified samples at each 
        iteration or randomly sample misclassified samples.
tol: list of float, default=[1e-4]
        Tolerance for stopping criteria for each SVC. 
        The list of tolerance for gridSearch cross validation. 
C: list of float, default=[1.0]
        Regularization parameter for each SVC. The strength of the regularization 
        is inversely proportional to C. Must be strictly positive.
        The list of C for gridSearch cross validation. 
n_splits: integer, default=5
        N-fold for gridSearch cross validation. 
penalty: {‘l1’, ‘l2’}, default=’l2’
        Specifies the norm used in the penalization. The ‘l2’ penalty is the 
        standard used in SVC. The ‘l1’ leads to sparse weight.
loss: {‘hinge’, ‘squared_hinge’}, default=’squared_hinge’
        Specifies the loss function for each SVC to train. ‘hinge’ is the standard 
        SVM loss while ‘squared_hinge’ is the square of the hinge loss. 
        The combination of penalty='l1' and loss='hinge' is not supported.
dual: bool, default=True
        Select the algorithm to either solve the dual or primal optimization 
        problem for each SVC. Prefer dual=False when n_samples > n_features.
fit_intercept: bool, default=True
        Whether to calculate the intercept for each SVC. If set to false, no 
        intercept will be used in calculations (i.e. data is already centered).
intercept_scaling: float, default=1
        When self.fit_intercept is True, instance vector x becomes 
        [x, self.intercept_scaling], i.e. a “synthetic” feature with constant value 
        equals to intercept_scaling is appended to the instance vector. The intercept 
        becomes intercept_scaling * synthetic feature weight Note! the synthetic 
        feature weight is subject to l1/l2 regularization as all other features. 
        To lessen the effect of regularization on synthetic feature weight 
        (and therefore on the intercept) intercept_scaling has to be increased.
class_weight: dict or ‘balanced’, default=None
        Set the parameter C of class i to class_weight[i]*C for SVC. If not given, 
        all classes are supposed to have weight one. The “balanced” mode uses the 
        values of y to automatically adjust weights inversely proportional to class 
        frequencies in the input data as n_samples / (n_classes * np.bincount(y)).
random_state: int, RandomState instance or None, default=None
        Controls the pseudo random number generation for shuffling data for the dual 
        coordinate descent (if dual=True). When dual=False underlying implementation 
        of LinearSVC is not random and random_state has no effect on the results. 
        Pass an int for reproducible output across multiple function calls.
max_iter: int, default=1000
        The maximum number of iterations to be run for each SVC.

Return

feature_selected: list of integer
        The sequence of features selected.
num_samples_list: list of integer
        The number of unique samples acquired totally after every features are selected.
train_errors: list of float
        Mean squared error of training set after every features are selected.
test_errors: list of float
        Mean squared error of test set after every features are selected.
train_accuracy: list of float
        Classification accuracy of training set after every features are selected.
test_accuracy: list of float
        Classification accuracy of test set after every features are selected.
Paras: list of dictionary
        The best estimator parameters of every SVMs trained. 
step_times: list of float
        The run time of each iteration. Unit is second. 

min_acquisition_cv

Parameters

X_train: {ndarray, sparse matrix} of shape {n_samples_X, n_features}
        Input data of training set.
y_train: ndarray of shape {n_samples_X,}
        Input classification labels of training set.
X_test: {ndarray, sparse matrix} of shape {n_samples_X, n_features}
        Input data of test set.
y_test: ndarray of shape {n_samples_X,}
        Input classification labels of test set.
num_features: integer
        The total number of features to select.
num_samples: integer
        The number of misclassified samples randomly sampled, which are taken union with 
        samples already acquired before. The union of samples are used for next ietration.
init_features: integer, default=1
        The number of features to select in the first iteration.
init_samples: integer, default=None
        The number of samples to use in the first iteration.
balance: bool, default=False
        Balance samples of each classes when sampling misclassified samples at each 
        iteration or randomly sample misclassified samples.
tol: list of float, default=[1e-4]
        Tolerance for stopping criteria for each SVC. 
        The list of tolerance for gridSearch cross validation. 
C: list of float, default=[1.0]
        Regularization parameter for each SVC. The strength of the regularization 
        is inversely proportional to C. Must be strictly positive.
        The list of C for gridSearch cross validation. 
n_splits: integer, default=5
        N-fold for gridSearch cross validation. 
penalty: {‘l1’, ‘l2’}, default=’l2’
        Specifies the norm used in the penalization. The ‘l2’ penalty is the 
        standard used in SVC. The ‘l1’ leads to sparse weight.
loss: {‘hinge’, ‘squared_hinge’}, default=’squared_hinge’
        Specifies the loss function for each SVC to train. ‘hinge’ is the standard 
        SVM loss while ‘squared_hinge’ is the square of the hinge loss. 
        The combination of penalty='l1' and loss='hinge' is not supported.
dual: bool, default=True
        Select the algorithm to either solve the dual or primal optimization 
        problem for each SVC. Prefer dual=False when n_samples > n_features.
fit_intercept: bool, default=True
        Whether to calculate the intercept for each SVC. If set to false, no 
        intercept will be used in calculations (i.e. data is already centered).
intercept_scaling: float, default=1
        When self.fit_intercept is True, instance vector x becomes 
        [x, self.intercept_scaling], i.e. a “synthetic” feature with constant value 
        equals to intercept_scaling is appended to the instance vector. The intercept 
        becomes intercept_scaling * synthetic feature weight Note! the synthetic 
        feature weight is subject to l1/l2 regularization as all other features. 
        To lessen the effect of regularization on synthetic feature weight 
        (and therefore on the intercept) intercept_scaling has to be increased.
class_weight: dict or ‘balanced’, default=None
        Set the parameter C of class i to class_weight[i]*C for SVC. If not given, 
        all classes are supposed to have weight one. The “balanced” mode uses the 
        values of y to automatically adjust weights inversely proportional to class 
        frequencies in the input data as n_samples / (n_classes * np.bincount(y)).
random_state: int, RandomState instance or None, default=None
        Controls the pseudo random number generation for shuffling data for the dual 
        coordinate descent (if dual=True). When dual=False underlying implementation 
        of LinearSVC is not random and random_state has no effect on the results. 
        Pass an int for reproducible output across multiple function calls.
max_iter: int, default=1000
        The maximum number of iterations to be run for each SVC.

Return

feature_selected: list of integer
        The sequence of features selected.
num_samples_list: list of integer
        The number of unique samples acquired totally after every features are selected.
samples_global: list of integer
        The indices of samples that are acquired.
train_errors: list of float
        Mean squared error of training set after every features are selected.
test_errors: list of float
        Mean squared error of test set after every features are selected.
train_accuracy: list of float
        Classification accuracy of training set after every features are selected.
test_accuracy: list of float
        Classification accuracy of test set after every features are selected.
Paras: list of dictionary
        The best estimator parameters of every SVMs trained. 
step_times: list of float
        The run time of each iteration. Unit is second. 

min_complexity_h5py

Parameters

data_cell: {h5py._hl.dataset.Dataset} of {scipy.sparse.csc matrix} 
        with shape { n_features, n_samples}.
        The h5py data as csc matrix.
indices_cell: {h5py._hl.dataset.Dataset}
        The indices of csc matrix.
indptr_cell: {h5py._hl.dataset.Dataset}
        The indptr of csc matrix.
data_gene: {h5py._hl.dataset.Dataset} of {scipy.sparse.csr matrix} 
        with shape {n_samples, n_features}.
        The h5py data as csr matrix.
indices_cell: {h5py._hl.dataset.Dataset}
        The indices of csr matrix.
indptr_cell: {h5py._hl.dataset.Dataset}
        The indptr of csr matrix.
y: ndarray of shape {n_samples,}
        Input classification labels of entire dataset.
shape: {h5py._hl.dataset.Dataset}
        The shape of { n_features, n_samples}.
idx_train: list of integer
        The indices of samples from entire dataset to produce training set.
idx_test: list of integer
        The indices of samples from entire dataset to produce test set.
num_features: integer
        The total number of features to select.
num_samples: integer
        The number of samples to use in each iteration (for each feature).
init_features: integer, default=1
        The number of features to select in the first iteration.
init_samples: integer, default=None
        The number of samples to use in the first iteration.
balance: bool, default=False
        Balance samples of each classes when sampling misclassified samples at each 
        iteration or randomly sample misclassified samples.
penalty: {‘l1’, ‘l2’}, default=’l2’
        Specifies the norm used in the penalization. The ‘l2’ penalty is the 
        standard used in SVC. The ‘l1’ leads to sparse weight.
loss: {‘hinge’, ‘squared_hinge’}, default=’squared_hinge’
        Specifies the loss function for each SVC to train. ‘hinge’ is the standard 
        SVM loss while ‘squared_hinge’ is the square of the hinge loss. 
        The combination of penalty='l1' and loss='hinge' is not supported.
dual: bool, default=True
        Select the algorithm to either solve the dual or primal optimization 
        problem for each SVC. Prefer dual=False when n_samples > n_features.
tol: float, default=1e-4
        Tolerance for stopping criteria for each SVC.
C: float, default=1.0
        Regularization parameter for each SVC. The strength of the regularization 
        is inversely proportional to C. Must be strictly positive.
fit_intercept: bool, default=True
        Whether to calculate the intercept for each SVC. If set to false, no 
        intercept will be used in calculations (i.e. data is already centered).
intercept_scaling: float, default=1
        When self.fit_intercept is True, instance vector x becomes 
        [x, self.intercept_scaling], i.e. a “synthetic” feature with constant value 
        equals to intercept_scaling is appended to the instance vector. The intercept 
        becomes intercept_scaling * synthetic feature weight Note! the synthetic 
        feature weight is subject to l1/l2 regularization as all other features. 
        To lessen the effect of regularization on synthetic feature weight 
        (and therefore on the intercept) intercept_scaling has to be increased.
class_weight: dict or ‘balanced’, default=None
        Set the parameter C of class i to class_weight[i]*C for SVC. If not given, 
        all classes are supposed to have weight one. The “balanced” mode uses the 
        values of y to automatically adjust weights inversely proportional to class 
        frequencies in the input data as n_samples / (n_classes * np.bincount(y)).
random_state: int, RandomState instance or None, default=None
        Controls the pseudo random number generation for shuffling data for the dual 
        coordinate descent (if dual=True). When dual=False underlying implementation 
        of LinearSVC is not random and random_state has no effect on the results. 
        Pass an int for reproducible output across multiple function calls.
max_iter: int, default=1000
        The maximum number of iterations to be run for each SVC.

Return

feature_selected: list of integer
        The sequence of features selected.
num_samples_list: list of integer
        The number of unique samples acquired totally after every features are selected.
train_errors: list of float
        Mean squared error of training set after every features are selected.
test_errors: list of float
        Mean squared error of test set after every features are selected.
train_accuracy: list of float
        Classification accuracy of training set after every features are selected.
test_accuracy: list of float
        Classification accuracy of test set after every features are selected.
step_times: list of float
        The run time of each iteration. Unit is second. 

min_acquisition

Parameters

data_cell: {h5py._hl.dataset.Dataset} of {scipy.sparse.csc matrix} 
        with shape { n_features, n_samples}.
        The h5py data as csc matrix.
indices_cell: {h5py._hl.dataset.Dataset}
        The indices of csc matrix.
indptr_cell: {h5py._hl.dataset.Dataset}
        The indptr of csc matrix.
data_gene: {h5py._hl.dataset.Dataset} of {scipy.sparse.csr matrix} 
        with shape {n_samples, n_features}.
        The h5py data as csr matrix.
indices_cell: {h5py._hl.dataset.Dataset}
        The indices of csr matrix.
indptr_cell: {h5py._hl.dataset.Dataset}
        The indptr of csr matrix.
y: ndarray of shape {n_samples,}
        Input classification labels of entire dataset.
shape: {h5py._hl.dataset.Dataset}
        The shape of { n_features, n_samples}.
idx_train: list of integer
        The indices of samples from entire dataset to produce training set.
idx_test: list of integer
        The indices of samples from entire dataset to produce test set.
num_features: integer
        The total number of features to select.
num_samples: integer
        The number of misclassified samples randomly sampled, which are taken union with 
        samples already acquired before. The union of samples are used for next ietration.
init_features: integer, default=1
        The number of features to select in the first iteration.
init_samples: integer, default=None
        The number of samples to use in the first iteration.
balance: bool, default=False
        Balance samples of each classes when sampling misclassified samples at each 
        iteration or randomly sample misclassified samples.
penalty: {‘l1’, ‘l2’}, default=’l2’
        Specifies the norm used in the penalization. The ‘l2’ penalty is the 
        standard used in SVC. The ‘l1’ leads to sparse weight.
loss: {‘hinge’, ‘squared_hinge’}, default=’squared_hinge’
        Specifies the loss function for each SVC to train. ‘hinge’ is the standard 
        SVM loss while ‘squared_hinge’ is the square of the hinge loss. 
        The combination of penalty='l1' and loss='hinge' is not supported.
dual: bool, default=True
        Select the algorithm to either solve the dual or primal optimization 
        problem for each SVC. Prefer dual=False when n_samples > n_features.
tol: float, default=1e-4
        Tolerance for stopping criteria for each SVC.
C: float, default=1.0
        Regularization parameter for each SVC. The strength of the regularization 
        is inversely proportional to C. Must be strictly positive.
fit_intercept: bool, default=True
        Whether to calculate the intercept for each SVC. If set to false, no 
        intercept will be used in calculations (i.e. data is already centered).
intercept_scaling: float, default=1
        When self.fit_intercept is True, instance vector x becomes 
        [x, self.intercept_scaling], i.e. a “synthetic” feature with constant value 
        equals to intercept_scaling is appended to the instance vector. The intercept 
        becomes intercept_scaling * synthetic feature weight Note! the synthetic 
        feature weight is subject to l1/l2 regularization as all other features. 
        To lessen the effect of regularization on synthetic feature weight 
        (and therefore on the intercept) intercept_scaling has to be increased.
class_weight: dict or ‘balanced’, default=None
        Set the parameter C of class i to class_weight[i]*C for SVC. If not given, 
        all classes are supposed to have weight one. The “balanced” mode uses the 
        values of y to automatically adjust weights inversely proportional to class 
        frequencies in the input data as n_samples / (n_classes * np.bincount(y)).
random_state: int, RandomState instance or None, default=None
        Controls the pseudo random number generation for shuffling data for the dual 
        coordinate descent (if dual=True). When dual=False underlying implementation 
        of LinearSVC is not random and random_state has no effect on the results. 
        Pass an int for reproducible output across multiple function calls.
max_iter: int, default=1000
        The maximum number of iterations to be run for each SVC.

Return

feature_selected: list of integer
        The sequence of features selected.
num_samples_list: list of integer
        The number of unique samples acquired totally after every features are selected.
samples_global: list of integer
        The indices of samples that are acquired.
train_errors: list of float
        Mean squared error of training set after every features are selected.
test_errors: list of float
        Mean squared error of test set after every features are selected.
train_accuracy: list of float
        Classification accuracy of training set after every features are selected.
test_accuracy: list of float
        Classification accuracy of test set after every features are selected.
step_times: list of float
        The run time of each iteration. Unit is second.