secure_learning.models.common_gradient_forms module

Provides classes for computing the gradient of objective functions

class secure_learning.models.common_gradient_forms.GradientFunction(*args, **kwargs)

Bases: Protocol

Class for objective function.

class secure_learning.models.common_gradient_forms.WeightedDifferencesGradient(predictive_func)

Bases: GradientFunction

Class for computing the gradient of objective functions. The gradient is assumed to have the following form: $$g(X, y, w) = X^T (f(X, w) - y)$$

We refer to $f$ as the predictive function.

__call__(X, y, coef_, grad_per_sample=False, weights_list=None)

Evaluate the gradient from the given parameters.

Note that this function calculates the gradient as if the input data consists out of all data samples. That is, it does not incorporate coefficients for gradients of partial input data.

Parameters:

• X (List[List[SecureFixedPoint]]) – Independent variables

• y (List[SecureFixedPoint]) – Dependent variables

• coef – Current coefficients vector

• grad_per_sample (bool) – Return a list with gradient per sample instead of aggregated (summed) gradient

• weights_list (Optional[List[SecureFixedPoint]]) – List of class weights to scale the prediction error by, defaults to None

Return type:

Union[List[SecureFixedPoint], List[List[SecureFixedPoint]]]

Returns:

Gradient of objective function as specified in class docstring, evaluated from the provided parameters

__init__(predictive_func)

Constructor method.

Parameters:

predictive_func (Callable[[List[List[SecureFixedPoint]], List[SecureFixedPoint]], List[SecureFixedPoint]]) – Function that predicts the dependent variable from the independent data and the model coefficients