maml.apps.symbolic package

Symbolic learning.

class maml.apps.symbolic.AdaptiveLasso(lambd, gamma, **kwargs)

Bases: PenalizedLeastSquares

Adaptive lasso regression using OLS coefficients as the root-n estimator coefficients.

_penalty_jac(x, y, beta)

get_w(x, y)

Get adaptive weights from data.

Parameters
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets

Returns: coefficients array.

penalty(beta: np.ndarray, x: np.ndarray | None = None, y: np.ndarray | None = None)

Calculate the penalty from input x, output y and coefficient beta.

Parameters
- beta (np.ndarray) – N coefficients
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets

Returns: penalty value.

select(x, y, options=None)

Select feature indices from x.

Parameters
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets
- options (dict) – options in the optimizations provided to scipy.optimize.minimize

Returns: list of int indices.

class maml.apps.symbolic.DantzigSelector(lambd, sigma=1.0, **kwargs)

Bases: BaseSelector

Equation 11 in https://orfe.princeton.edu/~jqfan/papers/06/SIS.pdf and reference in https://projecteuclid.org/download/pdfview_1/euclid.aos/1201012958.

construct_constraints(x: np.ndarray, y: np.ndarray, beta: np.ndarray | None = None)

Get constraints dictionary from data, e.g., {“func”: lambda beta: fun(x, y, beta), “type”: “ineq”}.

Parameters
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets
- beta (np.ndarray) – placeholder

Returns: dict of constraints.

construct_jac(x: ndarray, y: ndarray)

Jacobian of cost functions.

Parameters
- x – ndarray
- y – ndarray

Returns: callable

construct_loss(x, y, beta)

Get loss function from data and tentative coefficients beta.

Parameters
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets
- beta (np.ndarray) – N coefficients

Returns: loss value.

class maml.apps.symbolic.FeatureGenerator(feature_df: pd.DataFrame, operators: list)

Bases: object

FeatureGenerator class for feature augmentation before selection.

augment(n: int = 1)

Augment features :param n: number of rounds of iteration. :type n: int

Returns: augmented dataframe

class maml.apps.symbolic.ISIS(sis: SIS | None = None, l0_regulate: bool = True)

Bases: object

Iterative SIS.

evaluate(x: ndarray, y: ndarray, metric: str = ‘neg_mean_absolute_error’)

Evaluate the linear models using x, and y test data.

Parameters
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets
- metric (str) – scorer function, used with sklearn.metrics.get_scorer

Returns: float.

run(x: np.ndarray, y: np.ndarray, max_p: int = 10, metric: str = ‘neg_mean_absolute_error’, options: dict | None = None, step: float = 0.5)

Run the ISIS :param x: input array :type x: np.ndarray :param y: target array :type y: np.ndarray :param max_p: Number of feature desired :type max_p: int :param metric: scorer function, used with

sklearn.metrics.get_scorer

Parameters
- options –
- step (float) – step to update gamma with.
Returns np.array of index of selected features coeff(np.array): np.array of coeff of selected features
Return type find_sel(np.array)

class maml.apps.symbolic.L0BrutalForce(lambd: float, **kwargs)

Bases: BaseSelector

Brutal force combinatorial screening of features. This method takes all possible combinations of features and optimize the following loss function

1/2 * mean((y-x @ beta)**2) + lambd *

beta _0.

select(x: np.ndarray, y: np.ndarray, options: dict | None = None, n_job: int = 1)

L0 combinatorial optimization.

Parameters
- x (np.ndarray) – design matrix
- y (np.ndarray) – target vector
- options – Dict of options.
- n_job (int) – number of cpu

Returns:

class maml.apps.symbolic.Lasso(lambd, **kwargs)

Bases: PenalizedLeastSquares

Simple Lasso regression.

_penalty_jac(x, y, beta)

penalty(beta: np.ndarray, x: np.ndarray | None = None, y: np.ndarray | None = None)

Calculate the penalty from input x, output y and coefficient beta.

Parameters
- beta (np.ndarray) – N coefficients
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets

Returns: penalty value.

class maml.apps.symbolic.Operator(operation: Callable[[…], Any], rep: str, unary: bool, commutative: bool)

Bases: object

Operator class. Wrap math operators with more attributes including check is_unary, is_binary, and is_commutative, and generate name string for the output.

compute(i1: np.ndarray, i2: np.ndarray | None = None)

Compute the results :param i1: first input array :type i1: np.ndarray :param i2: second input array (for binary operators). :type i2: np.ndarray

Returns: array of computed results

classmethod from_str(op_name: str)

Operator from name of the operator :param op_name: string representation of the operator, :type op_name: str :param check Operator.support_op_rep for reference.:

Returns: Operator

gen_name(f1: str, f2: str | None = None)

Generate string representation for output :param f1: name of the first input array :type f1: str :param f2: name of the second input array. :type f2: str

Returns: name of the output

property is_binary(: boo )

True if the operator takes two arguments else False.

Type Returns

property is_commutative(: boo )

True if the operator is commutative else False.

Type Returns

property is_unary(: boo )

True if the operator takes one argument else False.

Type Returns

support_op_rep(_ = [‘^2’, ‘^3’, ‘sqrt’, ‘abssqrt’, ‘cbrt’, ‘exp’, ‘abs’, ‘log10’, ‘abslog10’, ‘+’, ‘-’, ‘*’, ‘/’, ‘|+|’, ‘|-|’, ‘sum_power_2’, ‘sum_exp’_ )

class maml.apps.symbolic.SCAD(lambd: float | np.ndarray, a: float = 3.7, **kwargs)

Bases: PenalizedLeastSquares

Smoothly clipped absolute deviation (SCAD), equation 12 and 13 in https://orfe.princeton.edu/~jqfan/papers/06/SIS.pdf.

_penalty_jac(x, y, beta)

penalty(beta: np.ndarray, x: np.ndarray | None = None, y: np.ndarray | None = None)

Calculate the SCAD penalty from input x, output y

and coefficient beta

Parameters
- beta (np.ndarray) – N coefficients
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets

Returns: penalty value.

class maml.apps.symbolic.SIS(gamma=0.1, selector: BaseSelector | None = None, verbose: bool = True)

Bases: object

Sure independence screening method. The method consists of two steps:

Screen

Select.

compute_residual(x, y)

Compute residual :param x: input array :type x: np.ndarray :param y: target array. :type y: np.ndarray

Returns: residual vector

run(x, y, select_options=None)

Run the SIS with selector :param x: MxN input data array :type x: np.ndarray :param y: M output targets :type y: np.ndarray :param select_options: options in the optimizations provided

to scipy.optimize.minimize. If the selector is using cvxpy optimization package, this option is fed into cp.Problem.solve.

Returns: selected feature indices

screen(x, y)

Simple screening method by comparing the correlation between features and the target.

Parameters
- x (np.ndarray) – input array
- y (np.ndarray) – target array

Returns: top indices

select(x, y, options=None)

Select features using selectors :param x: input array :type x: np.ndarray :param y: target array :type y: np.ndarray :param options: options for the optimization. :type options: dict

set_gamma(gamma)

Set gamma.

Parameters gamma (float) – new gamma value

set_selector(selector: BaseSelector)

Set new selector :param selector: a feature selector. :type selector: BaseSelector

update_gamma(ratio: float = 0.5)

Update the sis object so that sis.select return at least one feature.

Parameters ratio (float) – ratio to update the parameters

maml.apps.symbolic._feature_generator module

Feature Generator.

class maml.apps.symbolic._feature_generator.FeatureGenerator(feature_df: pd.DataFrame, operators: list)

Bases: object

FeatureGenerator class for feature augmentation before selection.

augment(n: int = 1)

Augment features :param n: number of rounds of iteration. :type n: int

Returns: augmented dataframe

class maml.apps.symbolic._feature_generator.Operator(operation: Callable[[…], Any], rep: str, unary: bool, commutative: bool)

Bases: object

Operator class. Wrap math operators with more attributes including check is_unary, is_binary, and is_commutative, and generate name string for the output.

compute(i1: np.ndarray, i2: np.ndarray | None = None)

Compute the results :param i1: first input array :type i1: np.ndarray :param i2: second input array (for binary operators). :type i2: np.ndarray

Returns: array of computed results

classmethod from_str(op_name: str)

Operator from name of the operator :param op_name: string representation of the operator, :type op_name: str :param check Operator.support_op_rep for reference.:

Returns: Operator

gen_name(f1: str, f2: str | None = None)

Generate string representation for output :param f1: name of the first input array :type f1: str :param f2: name of the second input array. :type f2: str

Returns: name of the output

property is_binary(: boo )

True if the operator takes two arguments else False.

Type Returns

property is_commutative(: boo )

True if the operator is commutative else False.

Type Returns

property is_unary(: boo )

True if the operator takes one argument else False.

Type Returns

support_op_rep(_ = [‘^2’, ‘^3’, ‘sqrt’, ‘abssqrt’, ‘cbrt’, ‘exp’, ‘abs’, ‘log10’, ‘abslog10’, ‘+’, ‘-’, ‘*’, ‘/’, ‘|+|’, ‘|-|’, ‘sum_power_2’, ‘sum_exp’_ )

maml.apps.symbolic._feature_generator._my_abs_diff(x, y)

maml.apps.symbolic._feature_generator._my_abs_log10(x)

maml.apps.symbolic._feature_generator._my_abs_sqrt(x)

maml.apps.symbolic._feature_generator._my_abs_sum(x, y)

maml.apps.symbolic._feature_generator._my_diff(x, y)

maml.apps.symbolic._feature_generator._my_div(x, y)

maml.apps.symbolic._feature_generator._my_exp(x)

maml.apps.symbolic._feature_generator._my_exp_power_2(x)

maml.apps.symbolic._feature_generator._my_exp_power_3(x)

maml.apps.symbolic._feature_generator._my_mul(x, y)

maml.apps.symbolic._feature_generator._my_power(x: float, n: int)

maml.apps.symbolic._feature_generator._my_sum(x, y)

maml.apps.symbolic._feature_generator._my_sum_exp(x, y)

maml.apps.symbolic._feature_generator._my_sum_power_2(x, y)

maml.apps.symbolic._feature_generator._my_sum_power_3(x, y)

maml.apps.symbolic._feature_generator._update_df(df, op, fn1, fn2=None)

Helper function to update the dataframe with new generated feature array.

maml.apps.symbolic._feature_generator.generate_feature(feature_df: pd.DataFrame, operators: list)

Generate new features by applying operators to columns in feature_df.

Parameters
- feature_df (pd.DataFrame) – dataframe of original features
- operators (list) – list of str of operators (check Operator.support_op_rep for reference)

Returns: dataframe of augmented features

maml.apps.symbolic._selectors module

Selectors.

class maml.apps.symbolic._selectors.AdaptiveLasso(lambd, gamma, **kwargs)

Bases: PenalizedLeastSquares

Adaptive lasso regression using OLS coefficients as the root-n estimator coefficients.

_penalty_jac(x, y, beta)

get_w(x, y)

Get adaptive weights from data.

Parameters
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets

Returns: coefficients array.

penalty(beta: np.ndarray, x: np.ndarray | None = None, y: np.ndarray | None = None)

Calculate the penalty from input x, output y and coefficient beta.

Parameters
- beta (np.ndarray) – N coefficients
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets

Returns: penalty value.

select(x, y, options=None)

Select feature indices from x.

Parameters
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets
- options (dict) – options in the optimizations provided to scipy.optimize.minimize

Returns: list of int indices.

class maml.apps.symbolic._selectors.BaseSelector(coef_thres: float = 1e-06, method: str = ‘SLSQP’)

Bases: object

Feature selector. This is meant to work on relatively smaller number of features.

classmethod _get_param_names()

compute_residual(x: ndarray, y: ndarray)

Compute.

Parameters
- x (np.ndarray) – design matrix
- y (np.ndarray) – target vector

Returns: residual vector.

construct_constraints(x: np.ndarray, y: np.ndarray, beta: np.ndarray | None = None)

Get constraints dictionary from data, e.g., {“func”: lambda beta: fun(x, y, beta), “type”: “ineq”}.

Parameters
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets
- beta (np.ndarray) – parameter to optimize

Returns: dict of constraints.

construct_jac(x: np.ndarray, y: np.ndarray)

Jacobian of cost function :param x: MxN input data array :type x: np.ndarray :param y: M output targets :type y: np.ndarray

Returns: Jacobian function.

construct_loss(x: ndarray, y: ndarray, beta: ndarray)

Get loss function from data and tentative coefficients beta :param x: MxN input data array :type x: np.ndarray :param y: M output targets :type y: np.ndarray :param beta: N coefficients :type beta: np.ndarray

Returns: loss value.

evaluate(x: ndarray, y: ndarray, metric: str = ‘neg_mean_absolute_error’)

Evaluate the linear models using x, and y test data.

Parameters
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets
- metric (str) – scorer function, used with sklearn.metrics.get_scorer

Returns:

get_coef()

Get coefficients Returns: the coefficients array.

get_feature_indices()

Get selected feature indices.

Returns: ndarray

get_params()

Get params for this selector.

Returns: mapping of string to any

parameter names mapped to their values

predict(x: ndarray)

Predict the results using sparsified coefficients.

Parameters x (np.ndarray) – design matrix

Returns: ndarray

select(x: np.ndarray, y: np.ndarray, options: dict | None = None)

Select feature indices from x :param x: MxN input data array :type x: np.ndarray :param y: M output targets :type y: np.ndarray :param options: options in the optimizations provided

to scipy.optimize.minimize

Returns: list of int indices.

set_params(**params)

Set the parameters of this selector :param **params: dict :param Selector parameters.:

Returns selector instance
Return type self

class maml.apps.symbolic._selectors.DantzigSelector(lambd, sigma=1.0, **kwargs)

Bases: BaseSelector

Equation 11 in https://orfe.princeton.edu/~jqfan/papers/06/SIS.pdf and reference in https://projecteuclid.org/download/pdfview_1/euclid.aos/1201012958.

construct_constraints(x: np.ndarray, y: np.ndarray, beta: np.ndarray | None = None)

Get constraints dictionary from data, e.g., {“func”: lambda beta: fun(x, y, beta), “type”: “ineq”}.

Parameters
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets
- beta (np.ndarray) – placeholder

Returns: dict of constraints.

construct_jac(x: ndarray, y: ndarray)

Jacobian of cost functions.

Parameters
- x – ndarray
- y – ndarray

Returns: callable

construct_loss(x, y, beta)

Get loss function from data and tentative coefficients beta.

Parameters
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets
- beta (np.ndarray) – N coefficients

Returns: loss value.

class maml.apps.symbolic._selectors.L0BrutalForce(lambd: float, **kwargs)

Bases: BaseSelector

Brutal force combinatorial screening of features. This method takes all possible combinations of features and optimize the following loss function

1/2 * mean((y-x @ beta)**2) + lambd *

beta _0.

select(x: np.ndarray, y: np.ndarray, options: dict | None = None, n_job: int = 1)

L0 combinatorial optimization.

Parameters
- x (np.ndarray) – design matrix
- y (np.ndarray) – target vector
- options – Dict of options.
- n_job (int) – number of cpu

Returns:

class maml.apps.symbolic._selectors.Lasso(lambd, **kwargs)

Bases: PenalizedLeastSquares

Simple Lasso regression.

_penalty_jac(x, y, beta)

penalty(beta: np.ndarray, x: np.ndarray | None = None, y: np.ndarray | None = None)

Calculate the penalty from input x, output y and coefficient beta.

Parameters
- beta (np.ndarray) – N coefficients
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets

Returns: penalty value.

class maml.apps.symbolic._selectors.PenalizedLeastSquares(coef_thres: float = 1e-06, method: str = ‘SLSQP’)

Bases: BaseSelector

Penalized least squares. In addition to minimizing the sum of squares loss, it adds an additional penalty to the coefficients.

_penalty_jac(x, y, beta)

_sse_jac(x, y, beta)

construct_constraints(x: np.ndarray, y: np.ndarray, beta: np.ndarray | None = None)

No constraints :param x: MxN input data array :type x: np.ndarray :param y: M output targets :type y: np.ndarray :param beta: placeholder only :type beta: np.ndarray

Returns: a list of dictionary constraints.

construct_jac(x: ndarray, y: ndarray)

Construct the jacobian of loss function :param x: MxN input data array :type x: np.ndarray :param y: M output targets :type y: np.ndarray

Returns: jacobian vector.

construct_loss(x: ndarray, y: ndarray, beta: ndarray)

Construct the loss function. An extra penalty term is added :param x: MxN input data array :type x: np.ndarray :param y: M output targets :type y: np.ndarray :param beta: N coefficients :type beta: np.ndarray

Returns: sum of errors.

penalty(beta: np.ndarray, x: np.ndarray | None = None, y: np.ndarray | None = None)

Calculate the penalty from input x, output y and coefficient beta :param x: MxN input data array :type x: np.ndarray :param y: M output targets :type y: np.ndarray :param beta: N coefficients :type beta: np.ndarray

Returns: penalty value.

class maml.apps.symbolic._selectors.SCAD(lambd: float | np.ndarray, a: float = 3.7, **kwargs)

Bases: PenalizedLeastSquares

Smoothly clipped absolute deviation (SCAD), equation 12 and 13 in https://orfe.princeton.edu/~jqfan/papers/06/SIS.pdf.

_penalty_jac(x, y, beta)

penalty(beta: np.ndarray, x: np.ndarray | None = None, y: np.ndarray | None = None)

Calculate the SCAD penalty from input x, output y

and coefficient beta

Parameters
- beta (np.ndarray) – N coefficients
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets

Returns: penalty value.

maml.apps.symbolic._selectors_cvxpy module

This module implements more robust optimization using the cvxpy package.

class maml.apps.symbolic._selectors_cvxpy.AdaptiveLassoCP(lambd, gamma, **kwargs)

Bases: PenalizedLeastSquaresCP

Adaptive lasso regression using OLS coefficients as the root-n estimator coefficients.

get_w(x: ndarray, y: ndarray)

Get adaptive weights from data :param x: MxN input data array :type x: np.ndarray :param y: M output targets :type y: np.ndarray

Returns: coefficients array.

penalty(beta: cp.Variable, x: np.ndarray | None = None, y: np.ndarray | None = None)

Returns: penalty value.

select(x: np.ndarray, y: np.ndarray, options: dict | None = None)

Select feature indices from x :param x: MxN input data array :type x: np.ndarray :param y: M output targets :type y: np.ndarray :param options: options in the cp.Problem.solve :type options: dict

Returns: array int indices.

class maml.apps.symbolic._selectors_cvxpy.BaseSelectorCP(coef_thres: float = 1e-06, method: str = ‘ECOS’)

Bases: BaseSelector

Base selector using cvxpy (CP).

construct_constraints(x: np.ndarray, y: np.ndarray, beta: cp.Variable | None = None)

Get constraints dictionary from data, e.g., {“func”: lambda beta: fun(x, y, beta), “type”: “ineq”}.

Parameters
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets
- beta – (np.ndarray): target variable for optimization

Returns: dict of constraints.

construct_loss(x: np.ndarray, y: np.ndarray, beta: cp.Variable)

Returns: loss value.

select(x: np.ndarray, y: np.ndarray, options: dict | None = None)

Select feature indices from x :param x: MxN input data array :type x: np.ndarray :param y: M output targets :type y: np.ndarray :param options: kwargs for cp.Problem.solve :type options: dict

Returns: list of int indices.

class maml.apps.symbolic._selectors_cvxpy.DantzigSelectorCP(lambd, sigma=1.0, **kwargs)

Bases: BaseSelectorCP

Equation 11 in https://orfe.princeton.edu/~jqfan/papers/06/SIS.pdf and reference in https://projecteuclid.org/download/pdfview_1/euclid.aos/1201012958.

construct_constraints(x: np.ndarray, y: np.ndarray, beta: cp.Variable | None = None)

Dantzig selector constraints :param x: MxN input data array :type x: np.ndarray :param y: M output targets :type y: np.ndarray :param beta: dimension N vector for optimization :type beta: cp.Variable

Returns: List of constraints.

construct_loss(x: np.ndarray, y: np.ndarray, beta: cp.Variable)

L1 loss :param x: MxN input data array :type x: np.ndarray :param y: M output targets :type y: np.ndarray :param beta: dimension N vector for optimization :type beta: cp.Variable

Returns: loss expression.

class maml.apps.symbolic._selectors_cvxpy.LassoCP(lambd, **kwargs)

Bases: PenalizedLeastSquaresCP

Simple Lasso regression.

penalty(beta: cp.Variable, x: np.ndarray | None = None, y: np.ndarray | None = None)

Returns: penalty value.

class maml.apps.symbolic._selectors_cvxpy.PenalizedLeastSquaresCP(coef_thres: float = 1e-06, method: str = ‘ECOS’)

Bases: BaseSelectorCP

Penalized least squares. In addition to minimizing the sum of squares loss, it adds an additional penalty to the coefficients.

construct_loss(x: np.ndarray, y: np.ndarray, beta: cp.Variable)

L1 loss :param x: MxN input data array :type x: np.ndarray :param y: M output targets :type y: np.ndarray :param beta: dimension N vector for optimization :type beta: cp.Variable

Returns: loss expression.

penalty(beta: cp.Variable, x: np.ndarray | None = None, y: np.ndarray | None = None)

Returns: penalty value.

maml.apps.symbolic._sis module

Sure Independence Screening.

https://orfe.princeton.edu/~jqfan/papers/06/SIS.pdf

class maml.apps.symbolic._sis.ISIS(sis: SIS | None = None, l0_regulate: bool = True)

Bases: object

Iterative SIS.

evaluate(x: ndarray, y: ndarray, metric: str = ‘neg_mean_absolute_error’)

Evaluate the linear models using x, and y test data.

Parameters
- x (np.ndarray) – MxN input data array
- y (np.ndarray) – M output targets
- metric (str) – scorer function, used with sklearn.metrics.get_scorer

Returns: float.

run(x: np.ndarray, y: np.ndarray, max_p: int = 10, metric: str = ‘neg_mean_absolute_error’, options: dict | None = None, step: float = 0.5)

Run the ISIS :param x: input array :type x: np.ndarray :param y: target array :type y: np.ndarray :param max_p: Number of feature desired :type max_p: int :param metric: scorer function, used with

sklearn.metrics.get_scorer

Parameters
- options –
- step (float) – step to update gamma with.
Returns np.array of index of selected features coeff(np.array): np.array of coeff of selected features
Return type find_sel(np.array)

class maml.apps.symbolic._sis.SIS(gamma=0.1, selector: BaseSelector | None = None, verbose: bool = True)

Bases: object

Sure independence screening method. The method consists of two steps:

Screen

Select.

compute_residual(x, y)

Compute residual :param x: input array :type x: np.ndarray :param y: target array. :type y: np.ndarray

Returns: residual vector

run(x, y, select_options=None)

Run the SIS with selector :param x: MxN input data array :type x: np.ndarray :param y: M output targets :type y: np.ndarray :param select_options: options in the optimizations provided

to scipy.optimize.minimize. If the selector is using cvxpy optimization package, this option is fed into cp.Problem.solve.

Returns: selected feature indices

screen(x, y)

Simple screening method by comparing the correlation between features and the target.

Parameters
- x (np.ndarray) – input array
- y (np.ndarray) – target array

Returns: top indices

select(x, y, options=None)

Select features using selectors :param x: input array :type x: np.ndarray :param y: target array :type y: np.ndarray :param options: options for the optimization. :type options: dict

set_gamma(gamma)

Set gamma.

Parameters gamma (float) – new gamma value

set_selector(selector: BaseSelector)

Set new selector :param selector: a feature selector. :type selector: BaseSelector

update_gamma(ratio: float = 0.5)

Update the sis object so that sis.select return at least one feature.

Parameters ratio (float) – ratio to update the parameters