tapas.report.utils.plot_roc_curve

tapas.report.utils.plot_roc_curve(data, names, title, output_path, suffix='', eff_epsilon=None, zoom_in=1, low_corner=True)

Parameters

• data (list of pairs (labels, scores), both np.arrays of same lengths) – The true labels and the scores of each attack.

• names (list of str of the same length) – The label for each curve.

• title (str) – Title to display on the figure.

• output_path (str) – Path to the folder where the ROC curve should be saved.

• eff_epsilon (positive float, or None) – If not None, the value of the effective epsilon for this ROC curve, for which the TP/FP and (1-FP)/(1-TP) curves are plotted.

• zoom_in (float, default 1) – Maximum value of TP and FP shown on the plot. The default of 1 shows the full ROC curve, but this can be used to “zoom in” to the TPR at low FPR, an important quantity for privacy analysis.

• low_corner (bool, default True) – Whether to zoom in near (0,0) (True), or (1,1) (False).