autoqild.dataset_readers.open_ml_padding_dr

Reader for OpenML datasets applying padding strategies to analyze data leakage.

Classes

OpenMLPaddingDatasetReader(dataset_id, imbalance)

Reader for OpenML datasets related to leakages with respect to the error codes for each padding manipulation.

class autoqild.dataset_readers.open_ml_padding_dr.OpenMLPaddingDatasetReader(dataset_id, imbalance, create_datasets=True, random_state=None, **kwargs)

Bases: OpenMLTimingDatasetReader

Reader for OpenML datasets related to leakages with respect to the error codes for each padding manipulation.

This class extends OpenMLTimingDatasetReader and is tailored for datasets extracted from network traces exploiting error codes in the network traces to perform side-channel attacks, such as the Bleichenbacher timing attack. It reads, cleans, and processes the dataset, and provides methods to create datasets with class imbalance to simulate attack scenarios.

Parameters:

• dataset_id (int) – The ID of the OpenML dataset.

• imbalance (float) – The ratio of the number of minority class samples to the number of majority class samples. Must be between 0 and 1.

• create_datasets (bool, default=True) – If True, creates leakage datasets during initialization.

• random_state (int or RandomState instance, optional) – Random state for reproducibility.

• **kwargs (dict) – Additional keyword arguments.

logger

Logger instance for logging information.

Type:

logging.Logger

dataset

The OpenML dataset object.

Type:

openml.datasets.OpenMLDataset

data_frame_raw

The raw DataFrame containing the dataset.

Type:

pandas.DataFrame

attribute_names

List of attribute names (features) in the dataset.

Type:

list of str

dataset_dictionary

A dictionary where keys are vulnerable class labels and values are tuples of (X, y) for the respective classes.

Type:

dict

n_features

Number of features in the dataset.

Type:

int

server

The server associated with the padding attack dataset.

Type:

str

vulnerable_classes

List of class labels representing vulnerable (incorrectly formatted) messages.

Type:

list of str

correct_class

The correct class label, representing correctly formatted messages.

Type:

str

Private Methods

---------------

__read_dataset__

Reads the dataset from OpenML and extracts relevant information. This method fetches the dataset using the OpenML API, extracts the raw data, and processes the dataset description to retrieve vulnerable class labels, number of features, and server information.

__create_leakage_datasets__

Creates separate datasets for each class by selecting only the samples that belong to the correct class and one vulnerable class at a time.

__clean_up_dataset__

Cleans and preprocesses the dataset. This method encodes categorical columns, formats class labels, fills missing values, and convert class label strings to integer values.

get_data(class_label=1)

Retrieves data for a specific class label.

Parameters:

class_label (int, default=1) – The class label for which to retrieve the data.

Returns:

• X (array-like of shape (n_samples, n_features)) – Feature matrix.

• y (array-like of shape (n_samples,)) – Target vector.

get_sampled_imbalanced_data(X, y)

Creates an imbalanced dataset by sampling from the data.

Parameters:

• X (array-like of shape (n_samples, n_features)) – Feature matrix.

• y (array-like of shape (n_samples,)) – Target vector.

Returns:

• X (array-like of shape (n_samples, n_features)) – Feature matrix after applying sampling to create imbalance.

• y (array-like of shape (n_samples,)) – Target vector after applying sampling to create imbalance.