rocelib.tasks package

Submodules

rocelib.tasks.ClassificationTask module

class rocelib.tasks.ClassificationTask.ClassificationTask(model, dataset, mm_models=None)[source]

Bases: Task

A specific task type for classification problems that extends the base Task class.

This class provides methods for training the model and retrieving positive instances from the training data.

model: The model to be trained and used for predictions.

_dataset: The dataset used for training the model.

add_evaluation_metric(metric_name, metric_class)[source]

add_recourse_method(method_name, method_class)[source]

convert_datatype(data, target_type)[source]

Converts a Pandas DataFrame to the specified data type.

@param data: pd.DataFrame - The input DataFrame. @param target_type: str - The target data type: “DataFrame”, “NPArray”, or “TTensor”. @return: Converted data in the specified format.

evaluate(methods=None, evaluations=None, visualisation=False, **kwargs)[source]

Evaluates the generated counterfactual explanations using specified evaluation metrics.

@param methods: List of recourse methods to evaluate. @param evaluations: List of evaluation metrics to apply. @return: Dictionary containing evaluation results per method and metric.

Return type:: Dict[str, Dict[str, Any]]

generate(methods=None, type='DataFrame', **kwargs)[source]

Generates counterfactual explanations for the specified methods and stores the results.

@param methods: List of recourse methods (by name) to use for counterfactual generation. If not provided, then counterfactuals will be generated for all methods @param type: The datatype your instances are in e.g. dataframe, nparray, tensor @return: A dictionary from recourse method to a tuple of (Pandas dataframe holding the counterfactual, time taken to generate the counterfactual)

Return type:: Dict[str, Tuple[DataFrame, float]]

generate_for_model_method(model_name, method, type, **kwargs)[source]

Return type:: Tuple[DataFrame, float]

generate_mm(methods=None, type='DataFrame', **kwargs)[source]

Generates counterfactual explanations for the specified methods for each of the stored models and stores the results.

@param methods: List of recourse methods (by name) to use for counterfactual generation. @return: A nested dictionary from recourse method to model name to a tuple of (Pandas dataframe holding the counterfactual, time taken to generate the counterfactual)

Return type:: Dict[str, Dict[str, Tuple[DataFrame, float]]]

get_random_positive_instance(neg_value, column_name='target')[source]

Retrieves a random positive instance from the training data that does not have the specified negative value.

This method continues to sample from the training data until a positive instance is found whose predicted label is not equal to the negative value.

@param neg_value: The value considered negative in the target variable. @param column_name: The name of the target column used to identify positive instances. @return: A Pandas Series representing a random positive instance.

Return type:: Series

save_evals_as_zip(table_data, headers, csv_filename, graph_filename)[source]

rocelib.tasks.Task module

class rocelib.tasks.Task.Task(model, dataset, mm_models=None)[source]

Bases: ABC

An abstract base class representing a general task that involves training a model on a specific dataset.

_dataset

The dataset used for training the model.

Type:: DatasetLoader

__model

The model to be trained and used for predictions.

Type:: TrainableModel

property CEs

property ces

Property to access the training data.

@return: The training data loaded from DatasetLoader.

property dataset

Property to access the training data.

@return: The training data loaded from DatasetLoader.

evaluate(methods, evaluations)[source]

Return type:: Dict[str, Dict[str, Any]]

generate(methods)[source]

Return type:: Dict[str, Tuple[DataFrame, float]]

generate_mm(methods)[source]

Return type:: Dict[str, Tuple[DataFrame, float]]

get_evaluation_metrics()[source]

Return type:: List[str]

get_random_positive_instance(neg_value, column_name='target')[source]

Abstract method to retrieve a random positive instance from the training data.

@param neg_value: The value considered negative in the target variable. @param column_name: The name of the target column. @return: A Pandas Series representing a random positive instance.

Return type:: Series

get_recourse_methods()[source]

Return type:: List[str]

property mm_CEs

property mm_models

Property to access the model.

@return: The model instance that extends TrainableModel

property model