beetroots.space_transform package

Submodules

beetroots.space_transform.abstract_transform module

class beetroots.space_transform.abstract_transform.Scaler

Bases: ABC

class that defines a transition between the sampling space and a more user friendly / interpretable space

abstract from_lin_to_scaled(Theta_linscale: ndarray) → ndarray

applies the transformation from user friendly scale to sampling scale

Parameters:

Theta_linscale (np.ndarray of shape (-1, D)) – array in user friendly scale

Returns:

array in sampling scale

Return type:

np.ndarray of shape (-1, D)

abstract from_scaled_to_lin(Theta_scaled: ndarray) → ndarray

applies the transformation from sampling scale to user friendly scale

Parameters:

Theta_scaled (np.ndarray of shape (-1, D)) – array in sampling scale

Returns:

array in user friendly scale

Return type:

np.ndarray of shape (-1, D)

beetroots.space_transform.id_transform module

Contains a mock class for the scaler object, that does not transform the arrays it receives

class beetroots.space_transform.id_transform.IdScaler

Bases: Scaler

from_lin_to_scaled(Theta_linscale: ndarray) → ndarray

applies the transformation from user friendly scale to sampling scale

Parameters:

Theta_linscale (np.ndarray of shape (-1, D)) – array in user friendly scale

Returns:

array in sampling scale

Return type:

np.ndarray of shape (-1, D)

from_scaled_to_lin(Theta_scaled: ndarray) → ndarray

applies the transformation from sampling scale to user friendly scale

Parameters:

Theta_scaled (np.ndarray of shape (-1, D)) – array in sampling scale

Returns:

array in user friendly scale

Return type:

np.ndarray of shape (-1, D)

beetroots.space_transform.transform module

Contains a class that defines the transition between the sampling scale and user friendly / interpretable scale

class beetroots.space_transform.transform.MyScaler(mean_: ndarray, std_: ndarray, list_is_log: List[bool])

Bases: Scaler

Defines the scale used during sampling and the transforms to navigate from one scale to the other. The transformation is a normalization (defined with a mean mean_ and standard deviation std_) for each physical parameter, defined on the log10 scale or on the linear scale depending on list_is_log.

Note

If one of the physical parameters is the scaling factor \(\kappa\), its mean is set to 0 and its std to 1 / np.log(10), so that it is not normalized regardless to its sampling scale (log10 or linear).

The std = 1 / np.log(10) for kappa yields a scaled validity interval that is about [-2.7, 2.7] (for a [0.1, 10] true validity interval), ie comparable to that of other normalized parameters.

D

total number of physical parameters that require a standard scaler, including the scaling factor \(\kappa\)

Type:

int

LOG_10 = 2.3025850929940455

from_lin_to_scaled(Theta_linscale: ndarray) → ndarray

applies the transformation from user friendly scale to sampling scale

Parameters:

Theta_linscale (np.ndarray of shape (-1, D)) – array in user friendly scale

Returns:

array in sampling scale

Return type:

np.ndarray of shape (-1, D)

from_scaled_to_lin(Theta_scaled: ndarray) → ndarray

applies the transformation from sampling scale to user friendly scale

Parameters:

Theta_scaled (np.ndarray of shape (-1, D)) – array in sampling scale

Returns:

array in user friendly scale

Return type:

np.ndarray of shape (-1, D)

list_is_log

whether the normalization should be applied on the log10 scale or in the linear scale

Type:

list of bool of length D

mean_

mean of the D components \(\theta_d\), used in the data normalization

Type:

np.ndarray of shape (D,)

std_

standard deviation of the D components \(\theta_d\), used in the data normalization

Type:

np.ndarray of shape (D,)

Module contents

Contains tranforms between sampling scales and user friendly scales