pyanno4rt.optimization.components._machine_learning_component_class

Machine learning component template.

Overview

Classes

MachineLearningComponentClass

Machine learning component template class.

Classes

class pyanno4rt.optimization.components._machine_learning_component_class.MachineLearningComponentClass(name, parameter_name, parameter_category, model_parameters, embedding, weight, rank, bounds, link, identifier, display)

Machine learning component template class.

Parameters:

• name (str) – Name of the component class.

• parameter_name (tuple) – Name of the component parameters.

• parameter_category (tuple) – Category of the component parameters.

• model_parameters (dict) –

  Dictionary with the data handling & learning model parameters:

  • model_labelstr

    Label for the machine learning model.

  • model_folder_pathNone or str, default=None

    Path to a folder for loading an external machine learning model.

  • data_pathstr

    Path to the data set used for fitting the machine learning model.

  • feature_filterdict, default={‘features’: [], ‘filter_mode’: ‘remove’}

    Dictionary with a list of feature names and a value from {‘retain’, ‘remove’} as an indicator for retaining/removing the features prior to model fitting.

  • label_namestr

    Name of the label variable.

  • label_boundslist, default=[1, 1]

    Bounds for the label values to binarize into positive (value lies inside the bounds) and negative class (value lies outside the bounds).

  • time_variable_nameNone or str, default=None

    Name of the time-after-radiotherapy variable (unit should be days).

  • label_viewpoint{‘early’, ‘late’, ‘long-term’, ‘longitudinal’, ‘profile’}, default=’longitudinal’

    Time of observation for the presence of tumor control and/or normal tissue complication events. The options can be described as follows:

    • ’early’ : event between 0 and 6 months after treatment

    • ’late’ : event between 6 and 15 months after treatment

    • ’long-term’ : event between 15 and 24 months after treatment

    • ’longitudinal’ : no period, time after treatment as covariate

    • ’profile’ : TCP/NTCP profiling over time, multi-label scenario with one label per month (up to 24 labels in total).

  • fuzzy_matchingbool, default=True

    Indicator for the use of fuzzy string matching to generate the feature map (if False, exact string matching is applied).

  • preprocessing_stepslist, default=[‘Identity’]

    Sequence of labels associated with preprocessing algorithms to preprocess the input features.

    The following preprocessing steps are currently available:

    • ’Identity’ Identity

    • ’StandardScaler’ StandardScaler

    • ’Whitening’ Whitening

  • architecture{‘input-convex’, ‘standard’}, default=’input-convex’

    Type of architecture for the neural network model.

  • max_hidden_layersint, default=2

    Maximum number of hidden layers for the neural network model.

  • tune_spacedict, default={}

    Search space for the Bayesian hyperparameter optimization.

  • tune_evaluationsint, default=50

    Number of evaluation steps (trials) for the Bayesian hyperparameter optimization.

  • tune_score{‘AUC’, ‘Brier score’, ‘Logloss’}, default=’Logloss’

    Scoring function for the evaluation of the hyperparameter set candidates.

  • tune_splitsint, default=5

    Number of splits for the stratified cross-validation within each hyperparameter optimization step.

  • inspect_modelbool, default=False

    Indicator for the inspection of the machine learning model.

  • evaluate_modelbool, default=False

    Indicator for the evaluation of the machine learning model.

  • oof_splitsint, default=5

    Number of splits for the stratified cross-validation within the out-of-folds evaluation step.

  • write_featuresbool, default=True

    Indicator for writing the iteratively calculated feature vectors into a feature history.

  • display_optionsdict, default={‘graphs’: [‘AUC-ROC’, ‘AUC-PR’, ‘F1’], ‘kpis’: [‘Logloss’, ‘Brier score’, ‘Subset accuracy’, ‘Cohen Kappa’, ‘Hamming loss’, ‘Jaccard score’, ‘Precision’, ‘Recall’, ‘F1 score’, ‘MCC’, ‘AUC’]}

    Dictionary with the graph and KPI display options.

• embedding ({'active', 'passive'}) – Mode of embedding for the component. In ‘passive’ mode, the component value is computed and tracked, but not considered in the optimization problem, unlike in ‘active’ mode.

• weight (int or float) – Weight of the component function.

• rank (int, default=1) – Rank of the component in the lexicographic order.

• bounds (None or list) – Constraint bounds for the component.

• link (None or list) – Other segments used for joint evaluation.

• identifier (None or str) – Additional string for naming the component.

• display (bool) – Indicator for the display of the component.

name

See ‘Parameters’.

Type:

str

parameter_name

See ‘Parameters’.

Type:

tuple

parameter_category

See ‘Parameters’.

Type:

tuple

parameter_value

Value of the component parameters.

Type:

list

embedding

See ‘Parameters’.

Type:

{‘active’, ‘passive’}

weight

See ‘Parameters’.

Type:

float

rank

See ‘Parameters’.

Type:

int

bounds

See ‘Parameters’.

Type:

list

link

See ‘Parameters’.

Type:

list

identifier

See ‘Parameters’.

Type:

None or str

display

See ‘Parameters’.

Type:

bool

model_parameters

See ‘Parameters’.

Type:

dict

data_model_handler

Initial variable for the object used to handle the dataset, the feature map generation and the feature (re-)calculation.

Type:

None

model

Initial variable for the object used to preprocess, tune, train, inspect and evaluate the machine learning model.

Type:

None

adjusted_parameters

Indicator for the adjustment of the parameters due to fractionation.

Type:

bool

RETURNS_OUTCOME

Indicator for the outcome focus of the component.

Type:

bool

DEPENDS_ON_MODEL

Indicator for the model dependency of the component.

Type:

bool

Overview

Methods

get_parameter_value()	Get the value of the parameters.
set_parameter_value(*args)	Set the value of the parameters.
get_weight_value()	Get the value of the weight.
set_weight_value(*args)	Set the value of the weight.
add_model()	abc Add the machine learning model to the component.
compute_value(*args)	abc Compute the component value.
compute_gradient(*args)	abc Compute the component gradient.

Members

get_parameter_value()

Get the value of the parameters.

Returns:

Value of the parameters.

Return type:

list

set_parameter_value(*args)

Set the value of the parameters.

Parameters:

*args (tuple) – Keyworded parameters. args[0] should give the value to be set.

get_weight_value()

Get the value of the weight.

Returns:

Value of the weight.

Return type:

float

set_weight_value(*args)

Set the value of the weight.

Parameters:

*args (tuple) – Keyworded parameters. args[0] should give the value to be set.

abstract add_model()

Add the machine learning model to the component.

abstract compute_value(*args)

Compute the component value.

abstract compute_gradient(*args)

Compute the component gradient.