metric_wrappers.py

"""Useful performance metric
"""

# Author: Xiaolu Xiong <beardeer@gmail.com>

from abc import ABCMeta, abstractmethod
from scipy import stats
from sklearn import metrics

class MetricWrapper(object):
    """The abstract class to build performance metrics
    """
    __metaclass__ = ABCMeta

    @classmethod
    @abstractmethod
    def measure(cls, label, prediction):
        """The abstract class method to measure the performance of predictions

        Parameters
        ----------
        label : array
            Lable data
        prediction : array
            Prediction data

        Raises
        ------
        NotImplementedError
        """
        raise NotImplementedError()


class RSquare(MetricWrapper):
    """The implementation of simple coefficient of determination.

    It is the square of pearson correlation.
    """
    @classmethod
    def measure(cls, label, prediction):
        """Measure the simple coefficient of determination.

        Parameters
        ----------
        label : array
            Lable data
        prediction : array
            Prediction data

        Returns
        -------
        float
            The square of pearson correlation between label and prediction
        """
        pearson, _ = stats.pearsonr(label, prediction)
        return pearson**2

class AUC(MetricWrapper):
    """The implementation of Area under the ROC curve.
    """
    @classmethod
    def measure(cls, label, prediction):
        """Measure the AUC.

        Parameters
        ----------
        label : array
            Lable data
        prediction : array
            Prediction data

        Returns
        -------
        float
            The AUC value between label and prediction
        """
        return metrics.roc_auc_score(label, prediction)


class RMSE(MetricWrapper):
    """The implementation of root-mean-square error.
    """
    @classmethod
    def measure(cls, label, prediction):
        """Measure the RMSE

        Parameters
        ----------
        label : array
            Lable data
        prediction : array
            Prediction data

        Returns
        -------
        float
            The RMSE value of lable and prediction
        """
        return metrics.mean_squared_error(label, prediction)**0.5