Module trustML.metrics.verifiederrortree

Expand source code 
from trustML.metrics.metric import Metric
from art.metrics import RobustnessVerificationTreeModelsCliqueMethod
from art.estimators.classification import SklearnClassifier
import pandas as pd

class VerifiedErrorSKLTree(Metric):
    """Verified error of a decision-tree sklearn-based model on the provided dataset (dataX, dataY) 
    using the ART package.

    This metric is typically used to verify the robustness of the classifier on the provided dataset. 
    Although it is a cost function, its assessment is inverted so it can be treated as the rest of metrics 
    (i.e., as a percentage).
   
    ADDITIONAL PROPERTIES:
    None

    Args:
        Metric (Class): Metric abstract class
    """
    
    def __init__(self):
        super().__init__()

    def assess(self, trained_model, data_x, data_y):
        print("Computing verified error...")
        rf_skmodel = SklearnClassifier(model=trained_model)
        rt = RobustnessVerificationTreeModelsCliqueMethod(classifier=rf_skmodel)        
        average_bound, verified_error = rt.verify(x=data_x.values, y=pd.get_dummies(data_y).values, eps_init=0.001,
         nb_search_steps=1, max_clique=2, max_level=1)

        self.score = (1-verified_error) # INVERT THE ERROR TO GET THE SCORE

Classes

class VerifiedErrorSKLTree

Verified error of a decision-tree sklearn-based model on the provided dataset (dataX, dataY) using the ART package.

This metric is typically used to verify the robustness of the classifier on the provided dataset. Although it is a cost function, its assessment is inverted so it can be treated as the rest of metrics (i.e., as a percentage).

ADDITIONAL PROPERTIES: None

Args

Metric : Class
Metric abstract class
Expand source code 
class VerifiedErrorSKLTree(Metric):
    """Verified error of a decision-tree sklearn-based model on the provided dataset (dataX, dataY) 
    using the ART package.

    This metric is typically used to verify the robustness of the classifier on the provided dataset. 
    Although it is a cost function, its assessment is inverted so it can be treated as the rest of metrics 
    (i.e., as a percentage).
   
    ADDITIONAL PROPERTIES:
    None

    Args:
        Metric (Class): Metric abstract class
    """
    
    def __init__(self):
        super().__init__()

    def assess(self, trained_model, data_x, data_y):
        print("Computing verified error...")
        rf_skmodel = SklearnClassifier(model=trained_model)
        rt = RobustnessVerificationTreeModelsCliqueMethod(classifier=rf_skmodel)        
        average_bound, verified_error = rt.verify(x=data_x.values, y=pd.get_dummies(data_y).values, eps_init=0.001,
         nb_search_steps=1, max_clique=2, max_level=1)

        self.score = (1-verified_error) # INVERT THE ERROR TO GET THE SCORE

Ancestors

Inherited members