mlquantify 0.0.11__py3-none-any.whl → 0.1.0__py3-none-any.whl

mlquantify/methods/threshold_optimization.py

@@ -0,0 +1,959 @@
+from abc import abstractmethod
+import numpy as np
+from sklearn.base import BaseEstimator
+
+from ..base import AggregativeQuantifier
+from ..utils.method import adjust_threshold, get_scores
+import mlquantify as mq
+
+
+
+
+class ThresholdOptimization(AggregativeQuantifier):
+    """
+    Generic Class for methods that adjust the decision boundary of the underlying classifier
+    to make the ACC (base method for threshold methods) estimation more numerically stable. 
+    Most strategies involve altering the denominator of the ACC equation.
+    
+    This class serves as a base for implementing threshold optimization techniques in classification
+    tasks. It is designed to adjust thresholds based on true positive and false positive rates, 
+    ensuring better quantification performance.
+
+    Parameters
+    ----------
+    learner : BaseEstimator
+        A scikit-learn compatible classifier to be used for threshold optimization.
+    threshold : float, optional
+        The threshold value to be used for classification decisions. Default is 0.5.
+
+    Attributes
+    ----------
+    learner : BaseEstimator
+        A scikit-learn compatible classifier.
+    threshold : float, optional
+        The optimized threshold used for classification decisions.
+    cc_output : float, optional
+        The classification count output, representing the proportion of instances classified 
+        as positive based on the threshold.
+    tpr : float, optional
+        The true positive rate corresponding to the best threshold.
+    fpr : float, optional
+        The false positive rate corresponding to the best threshold.
+
+    Notes
+    -----
+    All methods that inherit from this class will be binary quantifiers. In case of multiclass problems, it will be made One vs All.
+
+    Examples
+    --------
+    >>> from mlquantify.methods.threshold_optimization import ThresholdOptimization
+    >>> from mlquantify.utils.general import get_real_prev
+    >>> from sklearn.datasets import load_breast_cancer
+    >>> from sklearn.svm import SVC
+    >>> from sklearn.model_selection import train_test_split
+    >>> 
+    >>> class MyThrMethod(ThresholdOptimization):
+    ...     def __init__(self, learner, threshold=0.5):
+    ...         super().__init__(learner)
+    ...         self.threshold = threshold
+    ...     def best_tprfpr(self, thresholds, tpr, fpr):
+    ...         return thresholds[20], tpr[20], fpr[20]
+    >>>  
+    >>> features, target = load_breast_cancer(return_X_y=True)
+    >>> 
+    >>> X_train, X_test, y_train, y_test = train_test_split(features, target, test_size=0.2, random_state=42)
+    >>> 
+    >>> mtm = MyThrMethod(learner=SVC(probability=True), threshold=0.5)
+    >>> mtm.fit(X_train, y_train)
+    >>> y_pred = mtm.predict(X_test)
+    """
+
+    def __init__(self, learner: BaseEstimator=None):
+        self.learner = learner
+        self.threshold = None
+        self.cc_output = None
+        self.tpr = None
+        self.fpr = None
+    
+    @property
+    def is_probabilistic(self) -> bool:
+        """
+        Returns whether the method is probabilistic.
+
+        This method is used to determine whether the quantification method is probabilistic, 
+        meaning it uses class-conditional probabilities to estimate class prevalences.
+        
+        Returns
+        -------
+        bool
+            True, indicating that this method is probabilistic.
+        """
+        return True
+    
+    @property
+    def is_multiclass(self) -> bool:
+        """
+        Returns whether the method is applicable to multiclass quantification.
+
+        Threshold-based methods are typically binary classifiers, so this method 
+        returns False.
+
+        Returns
+        -------
+        bool
+            False, indicating that this method does not support multiclass quantification.
+        """
+        return False
+    
+    def _fit_method(self, X, y):
+        """
+        Fits the classifier and adjusts thresholds based on true positive rate (TPR) and false positive rate (FPR).
+
+        Parameters
+        ----------
+        X : pd.DataFrame or np.ndarray
+            The input features for training.
+        y : pd.Series or np.ndarray
+            The target labels for training.
+
+        Returns
+        -------
+        self : ThresholdOptimization
+            The fitted quantifier object with the best threshold, TPR, and FPR.
+        """
+        # Get predicted labels and probabilities
+        if mq.arguments["y_labels"] is not None and mq.arguments["posteriors_train"] is not None:
+            y_labels = mq.arguments["y_labels"]
+            probabilities = mq.arguments["posteriors_train"]
+        else:
+            y_labels, probabilities = get_scores(X, y, self.learner, self.cv_folds, self.learner_fitted)
+        
+        # Adjust thresholds and compute true and false positive rates
+        thresholds, tprs, fprs = adjust_threshold(y_labels, probabilities[:, 1], self.classes)
+        
+        # Find the best threshold based on TPR and FPR
+        self.threshold, self.tpr, self.fpr = self.best_tprfpr(thresholds, tprs, fprs)
+        
+        return self
+    
+    def _predict_method(self, X) -> dict:
+        """
+        Predicts class prevalences using the adjusted threshold.
+
+        Parameters
+        ----------
+        X : pd.DataFrame or np.ndarray
+            The input features for prediction.
+
+        Returns
+        -------
+        np.ndarray
+            An array of predicted prevalences for the classes.
+        """
+        # Get predicted probabilities for the positive class
+        probabilities = self.predict_learner(X)[:, 1]
+        
+        # Compute the classification count output based on the threshold
+        self.cc_output = len(probabilities[probabilities >= self.threshold]) / len(probabilities)
+        
+        # Calculate prevalence, ensuring it is within [0, 1]
+        if self.tpr - self.fpr == 0:
+            prevalence = self.cc_output
+        else:
+            # Equation of threshold methods to compute prevalence
+            prevalence = np.clip((self.cc_output - self.fpr) / (self.tpr - self.fpr), 0, 1)
+        
+        prevalences = [1 - prevalence, prevalence]
+
+        return np.asarray(prevalences)
+    
+    @abstractmethod
+    def best_tprfpr(self, thresholds: np.ndarray, tpr: np.ndarray, fpr: np.ndarray) -> float:
+        """
+        Abstract method for determining the best TPR (True Positive Rate) and FPR (False Positive Rate) 
+        to use in the equation for threshold optimization.
+
+        This method needs to be implemented by subclasses to define how the best threshold 
+        is chosen based on TPR and FPR.
+
+        Parameters
+        ----------
+        thresholds : np.ndarray
+            An array of threshold values.
+        tpr : np.ndarray
+            An array of true positive rates corresponding to the thresholds.
+        fpr : np.ndarray
+            An array of false positive rates corresponding to the thresholds.
+
+        Returns
+        -------
+        float
+            The best threshold value determined based on the true positive and false positive rates.
+        """
+        ...
+
+
+
+
+
+
+class ACC(ThresholdOptimization):
+    """
+    Adjusted Classify and Count (ACC). This method is a base approach for threshold-based 
+    quantification methods.
+
+    As described in the ThresholdOptimization base class, this method estimates the true 
+    positive rate (TPR) and false positive rate (FPR) from the training data. It then uses 
+    these values to adjust the output of the Classify and Count (CC) method, making the 
+    quantification process more accurate and stable.
+
+    Parameters
+    ----------
+    learner : BaseEstimator
+        A scikit-learn compatible classifier to be used for quantification.
+    threshold : float, optional
+        The decision threshold for classifying instances. Default is 0.5.
+    
+    Attributes
+    ----------
+    learner : BaseEstimator
+        A scikit-learn compatible classifier.
+    threshold : float
+        The decision threshold used to classify instances as positive or negative. Default is 0.5.
+
+    See Also
+    --------
+    ThresholdOptimization : Base class for threshold-based quantification methods.
+    CC : Classify and Count quantification method.
+    
+    References
+    ----------
+    FORMAN, George. Quantifying counts and costs via classification. Data Mining and Knowledge Discovery, v. 17, p. 164-206, 2008. Available at: https://link.springer.com/article/10.1007/s10618-008-0097-y
+    
+    Examples
+    --------
+    >>> from mlquantify.methods.aggregative import ACC
+    >>> from mlquantify.utils.general import get_real_prev
+    >>> from sklearn.datasets import load_breast_cancer
+    >>> from sklearn.svm import SVC
+    >>> from sklearn.model_selection import train_test_split
+    >>>
+    >>> features, target = load_breast_cancer(return_X_y=True)
+    >>>
+    >>> X_train, X_test, y_train, y_test = train_test_split(features, target, test_size=0.2, random_state=42)
+    >>>
+    >>> acc = ACC(learner=SVC(probability=True), threshold=0.5)
+    >>> acc.fit(X_train, y_train)
+    >>> y_pred = acc.predict(X_test)
+    >>> y_pred
+    {0: 0.3968506555196656, 1: 0.6031493444803344}
+    >>> get_real_prev(y_test)
+    {0: 0.3991228070175439, 1: 0.6008771929824561}
+    """
+
+    def __init__(self, learner: BaseEstimator=None, threshold: float = 0.5):
+        super().__init__(learner)
+        self.threshold = threshold
+
+    def best_tprfpr(self, thresholds: np.ndarray, tprs: np.ndarray, fprs: np.ndarray) -> tuple:
+        """
+        Determines the true positive rate (TPR) and false positive rate (FPR) for the specified threshold.
+
+        This method identifies the TPR and FPR corresponding to the threshold provided 
+        during initialization. It assumes that the `thresholds`, `tprs`, and `fprs` arrays 
+        are aligned, meaning the `i-th` element of each array corresponds to the same threshold.
+
+        Parameters
+        ----------
+        thresholds : np.ndarray
+            An array of threshold values.
+        tprs : np.ndarray
+            An array of true positive rates corresponding to the thresholds.
+        fprs : np.ndarray
+            An array of false positive rates corresponding to the thresholds.
+
+        Returns
+        -------
+        tuple
+            A tuple containing the threshold, the true positive rate (TPR), and the false 
+            positive rate (FPR) for the specified threshold.
+
+        Raises
+        ------
+        IndexError
+            If the specified threshold is not found in the `thresholds` array.
+        """
+        # Get the TPR and FPR where the threshold matches the specified value
+        tpr = tprs[thresholds == self.threshold][0]
+        fpr = fprs[thresholds == self.threshold][0]
+        return (self.threshold, tpr, fpr)
+
+    
+    
+    
+    
+    
+    
+    
+
+class MAX(ThresholdOptimization):
+    """
+    Threshold MAX. This quantification method selects the threshold that maximizes 
+    the absolute difference between the true positive rate (TPR) and false positive 
+    rate (FPR). This threshold is then used in the denominator of the equation for 
+    adjusted prevalence estimation.
+
+    Parameters
+    ----------
+    learner : BaseEstimator
+        A scikit-learn compatible classifier to be used for quantification.
+
+    Attributes
+    ----------
+    learner : BaseEstimator
+        A scikit-learn compatible classifier.
+        
+    See Also
+    --------
+    ThresholdOptimization : Base class for threshold-based quantification methods.
+    ACC : Adjusted Classify and Count quantification method.
+    CC : Classify and Count quantification method.
+    
+    References
+    ----------
+    FORMAN, George. Counting positives accurately despite inaccurate classification. In: European conference on machine learning. Berlin, Heidelberg: Springer Berlin Heidelberg, 2005. p. 564-575. Available at: https://link.springer.com/chapter/10.1007/11564096_56
+    
+    Examples
+    --------
+    >>> from mlquantify.methods.aggregative import MAX
+    >>> from mlquantify.utils.general import get_real_prev
+    >>> from sklearn.datasets import load_breast_cancer
+    >>> from sklearn.svm import SVC
+    >>> from sklearn.model_selection import train_test_split
+    >>>
+    >>> features, target = load_breast_cancer(return_X_y=True)
+    >>>
+    >>> X_train, X_test, y_train, y_test = train_test_split(features, target, test_size=0.2, random_state=42)
+    >>>
+    >>> maxq = MAX(learner=SVC(probability=True))
+    >>> maxq.fit(X_train, y_train)
+    >>> y_pred = maxq.predict(X_test)
+    >>> y_pred
+    {0: 0.3920664352842359, 1: 0.6079335647157641}
+    >>> get_real_prev(y_test)
+    {0: 0.3991228070175439, 1: 0.6008771929824561}
+    """
+
+    def __init__(self, learner: BaseEstimator=None):
+        super().__init__(learner)
+
+    def best_tprfpr(self, thresholds: np.ndarray, tprs: np.ndarray, fprs: np.ndarray) -> tuple:
+        """
+        Determines the optimal threshold by maximizing the absolute difference between 
+        the true positive rate (TPR) and the false positive rate (FPR).
+
+        This method identifies the index where `|TPR - FPR|` is maximized and retrieves 
+        the corresponding threshold, TPR, and FPR.
+
+        Parameters
+        ----------
+        thresholds : np.ndarray
+            An array of threshold values.
+        tprs : np.ndarray
+            An array of true positive rates corresponding to the thresholds.
+        fprs : np.ndarray
+            An array of false positive rates corresponding to the thresholds.
+
+        Returns
+        -------
+        tuple
+            A tuple containing:
+            - The threshold that maximizes `|TPR - FPR|`.
+            - The true positive rate (TPR) at the selected threshold.
+            - The false positive rate (FPR) at the selected threshold.
+
+        Raises
+        ------
+        ValueError
+            If `thresholds`, `tprs`, or `fprs` are empty or have mismatched lengths.
+        """
+        max_index = np.argmax(np.abs(tprs - fprs))
+        
+        # Retrieve the corresponding threshold, TPR, and FPR
+        threshold = thresholds[max_index]
+        tpr = tprs[max_index]
+        fpr = fprs[max_index]
+        return (threshold, tpr, fpr)
+
+    
+    
+
+
+
+
+
+
+class MS(ThresholdOptimization):
+    """
+    Median Sweep (MS). This quantification method uses an ensemble 
+    of threshold-based methods, taking the median values of the 
+    true positive rate (TPR) and false positive rate (FPR) across 
+    all thresholds to compute adjusted prevalences.
+
+    Parameters
+    ----------
+    learner : BaseEstimator
+        A scikit-learn compatible classifier to be used for quantification.
+    threshold : float, optional
+        The default threshold value to use for the quantification method. Default is 0.5.
+
+    Attributes
+    ----------
+    learner : BaseEstimator
+        A scikit-learn compatible classifier.
+    threshold : float
+        The default threshold to use for the quantification method, typically 0.5.
+
+    See Also
+    --------
+    ThresholdOptimization : Base class for threshold-based quantification methods.
+    ACC : Adjusted Classify and Count quantification method.
+    MAX : Threshold MAX quantification method.
+    CC : Classify and Count quantification method.
+    
+    References
+    ----------
+    FORMAN, George. Quantifying counts and costs via classification. Data Mining and Knowledge Discovery, v. 17, p. 164-206, 2008. Available at: https://link.springer.com/article/10.1007/s10618-008-0097-y
+    
+    Examples
+    --------
+    >>> from mlquantify.methods.aggregative import MS
+    >>> from mlquantify.utils.general import get_real_prev
+    >>> from sklearn.datasets import load_breast_cancer
+    >>> from sklearn.svm import SVC
+    >>> from sklearn.model_selection import train_test_split
+    >>>
+    >>> features, target = load_breast_cancer(return_X_y=True)
+    >>>
+    >>> X_train, X_test, y_train, y_test = train_test_split(features, target, test_size=0.2, random_state=42)
+    >>>
+    >>> ms = MS(learner=SVC(probability=True))
+    >>> ms.fit(X_train, y_train)
+    >>> y_pred = ms.predict(X_test)
+    >>> y_pred
+    {0: 0.41287676595138967, 1: 0.5871232340486103}
+    >>> get_real_prev(y_test)
+    {0: 0.3991228070175439, 1: 0.6008771929824561}
+    """
+
+    def __init__(self, learner: BaseEstimator=None, threshold: float = 0.5):
+        super().__init__(learner)
+        self.threshold = threshold
+
+    def best_tprfpr(self, thresholds: np.ndarray, tprs: np.ndarray, fprs: np.ndarray) -> tuple:
+        """
+        Determines the optimal TPR and FPR by taking the median of 
+        all TPR and FPR values across the given thresholds.
+
+        This method computes the median values of TPR and FPR to 
+        mitigate the influence of outliers and variability in the 
+        performance metrics.
+
+        Parameters
+        ----------
+        thresholds : np.ndarray
+            An array of threshold values.
+        tprs : np.ndarray
+            An array of true positive rates corresponding to the thresholds.
+        fprs : np.ndarray
+            An array of false positive rates corresponding to the thresholds.
+
+        Returns
+        -------
+        tuple
+            A tuple containing:
+            - The default threshold value (float).
+            - The median true positive rate (float).
+            - The median false positive rate (float).
+
+        Raises
+        ------
+        ValueError
+            If `thresholds`, `tprs`, or `fprs` are empty or have mismatched lengths.
+        """
+        # Compute median TPR and FPR
+        tpr = np.median(tprs)
+        fpr = np.median(fprs)
+
+        return (self.threshold, tpr, fpr)
+
+    
+
+
+
+
+
+
+class MS2(ThresholdOptimization):
+    """
+    Median Sweep 2 (MS2). This method is an extension of the 
+    Median Sweep strategy, but it focuses only on cases where 
+    the difference between the true positive rate (TPR) and the 
+    false positive rate (FPR) exceeds a threshold (0.25). The 
+    method computes the median values of TPR, FPR, and thresholds 
+    for these selected cases.
+
+    Parameters
+    ----------
+    learner : BaseEstimator
+        A scikit-learn compatible classifier to be used for quantification.
+
+    Attributes
+    ----------
+    learner : BaseEstimator
+        A scikit-learn compatible classifier.
+
+    References
+    ----------
+    FORMAN, George. Quantifying counts and costs via classification. Data Mining and Knowledge Discovery, v. 17, p. 164-206, 2008. Available at: https://link.springer.com/article/10.1007/s10618-008-0097-y
+
+    See Also
+    --------
+    ThresholdOptimization : Base class for threshold-based quantification methods.
+    ACC : Adjusted Classify and Count quantification method.
+    MS : Median Sweep quantification method.
+    CC : Classify and Count quantification method.
+    
+    Examples
+    --------
+    >>> from mlquantify.methods.aggregative import MS2
+    >>> from mlquantify.utils.general import get_real_prev
+    >>> from sklearn.datasets import load_breast_cancer
+    >>> from sklearn.svm import SVC
+    >>> from sklearn.model_selection import train_test_split
+    >>>
+    >>> features, target = load_breast_cancer(return_X_y=True)
+    >>>
+    >>> X_train, X_test, y_train, y_test = train_test_split(features, target, test_size=0.2, random_state=42)
+    >>>
+    >>> ms2 = MS2(learner=SVC(probability=True))
+    >>> ms2.fit(X_train, y_train)
+    >>> y_pred = ms2.predict(X_test)
+    >>> y_pred
+    {0: 0.41287676595138967, 1: 0.5871232340486103}
+    >>> get_real_prev(y_test)
+    {0: 0.3991228070175439, 1: 0.6008771929824561}
+    """
+
+    def __init__(self, learner: BaseEstimator=None):
+        super().__init__(learner)
+
+    def best_tprfpr(self, thresholds: np.ndarray, tprs: np.ndarray, fprs: np.ndarray) -> tuple:
+        """
+        Determines the optimal threshold, TPR, and FPR by focusing only on 
+        cases where the absolute difference between TPR and FPR is greater 
+        than 0.25. For these cases, the method computes the median values.
+
+        Parameters
+        ----------
+        thresholds : np.ndarray
+            An array of threshold values.
+        tprs : np.ndarray
+            An array of true positive rates corresponding to the thresholds.
+        fprs : np.ndarray
+            An array of false positive rates corresponding to the thresholds.
+
+        Returns
+        -------
+        tuple
+            A tuple containing:
+            - The median threshold value for cases meeting the condition (float).
+            - The median true positive rate for cases meeting the condition (float).
+            - The median false positive rate for cases meeting the condition (float).
+
+        Raises
+        ------
+        ValueError
+            If no cases satisfy the condition `|TPR - FPR| > 0.25` or if the 
+            input arrays are empty or have mismatched lengths.
+        """
+        # Identify indices where the condition is satisfied
+        indices = np.where(np.abs(tprs - fprs) > 0.25)[0]
+        if len(indices) == 0:
+            raise ValueError("No cases meet the condition |TPR - FPR| > 0.25.")
+
+        # Compute medians for the selected cases
+        threshold = np.median(thresholds[indices])
+        tpr = np.median(tprs[indices])
+        fpr = np.median(fprs[indices])
+
+        return (threshold, tpr, fpr)
+
+    
+    
+    
+    
+
+
+
+
+
+class PACC(ThresholdOptimization):
+    """
+    Probabilistic Adjusted Classify and Count (PACC). 
+    This method extends the Adjusted Classify and Count (AC) approach 
+    by leveraging the average class-conditional confidences obtained 
+    from a probabilistic classifier instead of relying solely on true 
+    positive and false positive rates.
+
+    Parameters
+    ----------
+    learner : BaseEstimator
+        A scikit-learn compatible classifier to be used for quantification.
+    threshold : float, optional
+        The decision threshold for classification. Default is 0.5.
+
+    Attributes
+    ----------
+    learner : BaseEstimator
+        A scikit-learn compatible classifier.
+    threshold : float
+        Decision threshold for classification. Default is 0.5.
+    tpr : float
+        True positive rate computed during the fitting process.
+    fpr : float
+        False positive rate computed during the fitting process.
+        
+    See Also
+    --------
+    ThresholdOptimization : Base class for threshold-based quantification methods.
+    ACC : Adjusted Classify and Count quantification method.
+    CC : Classify and Count quantification method.
+    
+    References
+    ----------
+    A. Bella, C. Ferri, J. Hernández-Orallo and M. J. Ramírez-Quintana, "Quantification via Probability Estimators," 2010 IEEE International Conference on Data Mining, Sydney, NSW, Australia, 2010, pp. 737-742, doi: 10.1109/ICDM.2010.75. Available at: https://ieeexplore.ieee.org/abstract/document/5694031
+    
+    Examples
+    --------
+    >>> from mlquantify.methods.aggregative import PACC
+    >>> from mlquantify.utils.general import get_real_prev
+    >>> from sklearn.datasets import load_breast_cancer
+    >>> from sklearn.svm import SVC
+    >>> from sklearn.model_selection import train_test_split
+    >>>
+    >>> features, target = load_breast_cancer(return_X_y=True)
+    >>>
+    >>> X_train, X_test, y_train, y_test = train_test_split(features, target, test_size=0.2, random_state=42)
+    >>>
+    >>> pacc = PACC(learner=SVC(probability=True))
+    >>> pacc.fit(X_train, y_train)
+    >>> y_pred = pacc.predict(X_test)
+    >>> y_pred
+    {0: 0.4664886119311328, 1: 0.5335113880688672}
+    >>> get_real_prev(y_test)
+    {0: 0.3991228070175439, 1: 0.6008771929824561}
+    """
+
+    def __init__(self, learner: BaseEstimator=None, threshold: float = 0.5):
+        super().__init__(learner)
+        self.threshold = threshold
+
+    def _predict_method(self, X):
+        """
+        Predicts the class prevalence using the mean class-conditional 
+        probabilities from a probabilistic classifier.
+
+        Parameters
+        ----------
+        X : array-like or sparse matrix of shape (n_samples, n_features)
+            The input data for prediction.
+
+        Returns
+        -------
+        dict
+            A dictionary with class labels as keys and their respective 
+            prevalence estimates as values.
+
+        Notes
+        -----
+        The prevalence is adjusted using the formula:
+            prevalence = |mean_score - FPR| / (TPR - FPR), 
+        where mean_score is the average probability for the positive class.
+
+        Raises
+        ------
+        ZeroDivisionError
+            If `TPR - FPR` equals zero, indicating that the classifier's 
+            performance does not vary across the threshold range.
+        """
+        prevalences = {}
+
+        # Calculate probabilities for the positive class
+        probabilities = self.predict_learner(X)[:, 1]
+
+        # Compute the mean score for the positive class
+        mean_scores = np.mean(probabilities)
+
+        # Adjust prevalence based on TPR and FPR
+        if self.tpr - self.fpr == 0:
+            prevalence = mean_scores
+        else:
+            prevalence = np.clip(abs(mean_scores - self.fpr) / (self.tpr - self.fpr), 0, 1)
+
+        # Map the computed prevalence to the class labels
+        prevalences[self.classes[0]] = 1 - prevalence
+        prevalences[self.classes[1]] = prevalence
+
+        return prevalences
+
+    def best_tprfpr(self, thresholds: np.ndarray, tprs: np.ndarray, fprs: np.ndarray) -> tuple:
+        """
+        Finds the true positive rate (TPR) and false positive rate (FPR) 
+        corresponding to the specified decision threshold.
+
+        Parameters
+        ----------
+        thresholds : np.ndarray
+            An array of threshold values.
+        tprs : np.ndarray
+            An array of true positive rates corresponding to the thresholds.
+        fprs : np.ndarray
+            An array of false positive rates corresponding to the thresholds.
+
+        Returns
+        -------
+        tuple
+            A tuple containing the specified threshold, TPR, and FPR.
+
+        Raises
+        ------
+        IndexError
+            If the specified threshold is not found in the `thresholds` array.
+        """
+        # Locate TPR and FPR for the specified threshold
+        tpr = tprs[thresholds == self.threshold][0]
+        fpr = fprs[thresholds == self.threshold][0]
+        return (self.threshold, tpr, fpr)
+
+    
+    
+    
+    def best_tprfpr(self, thresholds:np.ndarray, tprs: np.ndarray, fprs: np.ndarray) -> tuple:
+        tpr = tprs[thresholds == self.threshold][0]
+        fpr = fprs[thresholds == self.threshold][0]
+        return (self.threshold, tpr, fpr)
+    
+    
+    
+
+
+
+
+
+class T50(ThresholdOptimization):
+    """
+    Threshold 50 (T50). This method adjusts the decision threshold 
+    to the point where the true positive rate (TPR) is approximately 
+    equal to 0.5. This approach is particularly useful for balancing 
+    sensitivity and specificity in binary classification tasks.
+
+    Parameters
+    ----------
+    learner : BaseEstimator
+        A scikit-learn compatible classifier to be used for quantification.
+
+    Attributes
+    ----------
+    learner : BaseEstimator
+        A scikit-learn compatible classifier.
+    threshold : float
+        Decision threshold determined during training.
+    tpr : float
+        True positive rate corresponding to the selected threshold.
+    fpr : float
+        False positive rate corresponding to the selected threshold.
+
+    See Also
+    --------
+    ThresholdOptimization : Base class for threshold-based quantification methods.
+    ACC : Adjusted Classify and Count quantification method.
+    CC : Classify and Count quantification method.
+    
+    References
+    ----------
+    FORMAN, George. Quantifying counts and costs via classification. Data Mining and Knowledge Discovery, v. 17, p. 164-206, 2008. Available at: https://link.springer.com/article/10.1007/s10618-008-0097-y
+    
+    Examples
+    --------
+    >>> from mlquantify.methods.aggregative import T50
+    >>> from mlquantify.utils.general import get_real_prev
+    >>> from sklearn.datasets import load_breast_cancer
+    >>> from sklearn.svm import SVC
+    >>> from sklearn.model_selection import train_test_split
+    >>>
+    >>> features, target = load_breast_cancer(return_X_y=True)
+    >>>
+    >>> X_train, X_test, y_train, y_test = train_test_split(features, target, test_size=0.2, random_state=42)
+    >>>
+    >>> t50 = T50(learner=SVC(probability=True))
+    >>> t50.fit(X_train, y_train)
+    >>> y_pred = t50.predict(X_test)
+    >>> y_pred
+    {0: 0.49563196626070505, 1: 0.504368033739295}
+    >>> get_real_prev(y_test)
+    {0: 0.3991228070175439, 1: 0.6008771929824561}
+    """
+
+    def __init__(self, learner: BaseEstimator=None):
+        super().__init__(learner)
+
+    def best_tprfpr(self, thresholds: np.ndarray, tprs: np.ndarray, fprs: np.ndarray) -> tuple:
+        """
+        Determines the threshold, true positive rate (TPR), and false positive 
+        rate (FPR) where TPR is closest to 0.5.
+
+        Parameters
+        ----------
+        thresholds : np.ndarray
+            An array of threshold values.
+        tprs : np.ndarray
+            An array of true positive rates corresponding to the thresholds.
+        fprs : np.ndarray
+            An array of false positive rates corresponding to the thresholds.
+
+        Returns
+        -------
+        tuple
+            A tuple containing the selected threshold, TPR, and FPR.
+
+        Notes
+        -----
+        - The method identifies the index where the absolute difference 
+          between TPR and 0.5 is minimized.
+        - This ensures that the selected threshold represents a balance 
+          point in the ROC space.
+
+        Raises
+        ------
+        ValueError
+            If the arrays `thresholds`, `tprs`, or `fprs` are empty or 
+            misaligned in length.
+        """
+        # Find the index where TPR is closest to 0.5
+        min_index = np.argmin(np.abs(tprs - 0.5))
+
+        # Retrieve the corresponding threshold, TPR, and FPR
+        threshold = thresholds[min_index]
+        tpr = tprs[min_index]
+        fpr = fprs[min_index]
+
+        return (threshold, tpr, fpr)
+
+    
+    
+    
+    
+
+
+
+
+
+class X_method(ThresholdOptimization):
+    """
+    Threshold X. This method identifies the decision threshold where the 
+    false positive rate (FPR) is approximately equal to 1 - true positive rate (TPR). 
+    This criterion is useful for identifying thresholds that align with a balance 
+    point on the ROC curve.
+
+    Parameters
+    ----------
+    learner : BaseEstimator
+        A scikit-learn compatible classifier to be used for quantification.
+
+    Attributes
+    ----------
+    learner : BaseEstimator
+        A scikit-learn compatible classifier.
+    threshold : float
+        Decision threshold determined during training.
+    tpr : float
+        True positive rate corresponding to the selected threshold.
+    fpr : float
+        False positive rate corresponding to the selected threshold.
+
+    See Also
+    --------
+    ThresholdOptimization : Base class for threshold-based quantification methods.
+    ACC : Adjusted Classify and Count quantification method.
+    CC : Classify and Count quantification method.
+    
+    References
+    ----------
+    FORMAN, George. Quantifying counts and costs via classification. Data Mining and Knowledge Discovery, v. 17, p. 164-206, 2008. Available at: https://link.springer.com/article/10.1007/s10618-008-0097-y
+    
+    Examples
+    --------
+    >>> from mlquantify.methods.aggregative import X_method
+    >>> from mlquantify.utils.general import get_real_prev
+    >>> from sklearn.datasets import load_breast_cancer
+    >>> from sklearn.svm import SVC
+    >>> from sklearn.model_selection import train_test_split
+    >>>
+    >>> features, target = load_breast_cancer(return_X_y=True)
+    >>>
+    >>> X_train, X_test, y_train, y_test = train_test_split(features, target, test_size=0.2, random_state=42)
+    >>>
+    >>> x_method = X_method(learner=SVC(probability=True))
+    >>> x_method.fit(X_train, y_train)
+    >>> y_pred = x_method.predict(X_test)
+    >>> y_pred
+    {0: 0.40523495782808205, 1: 0.594765042171918}
+    >>> get_real_prev(y_test)
+    {0: 0.3991228070175439, 1: 0.6008771929824561}
+    """
+
+    def __init__(self, learner: BaseEstimator=None):
+        super().__init__(learner)
+
+    def best_tprfpr(self, thresholds: np.ndarray, tprs: np.ndarray, fprs: np.ndarray) -> tuple:
+        """
+        Determines the threshold, true positive rate (TPR), and false positive 
+        rate (FPR) where FPR is closest to 1 - TPR.
+
+        Parameters
+        ----------
+        thresholds : np.ndarray
+            An array of threshold values.
+        tprs : np.ndarray
+            An array of true positive rates corresponding to the thresholds.
+        fprs : np.ndarray
+            An array of false positive rates corresponding to the thresholds.
+
+        Returns
+        -------
+        tuple
+            A tuple containing the selected threshold, TPR, and FPR.
+
+        Notes
+        -----
+        - The method identifies the index where the absolute difference 
+          between FPR and 1 - TPR is minimized.
+        - This ensures that the selected threshold corresponds to a balance 
+          point based on the given criterion.
+
+        Raises
+        ------
+        ValueError
+            If the arrays `thresholds`, `tprs`, or `fprs` are empty or 
+            misaligned in length.
+        """
+        # Find the index where FPR is closest to 1 - TPR
+        min_index = np.argmin(np.abs(1 - (tprs + fprs)))
+
+        # Retrieve the corresponding threshold, TPR, and FPR
+        threshold = thresholds[min_index]
+        tpr = tprs[min_index]
+        fpr = fprs[min_index]
+
+        return (threshold, tpr, fpr)