auc

auc.py

import numpy as np
from sklearn.metrics import roc_auc_score
actual = [1, 0, 0, 1, 1, 0, 0, 1, 0, 1]
predicted = [0.8, 0.2, 0.6, 0.3, 0.1, 0.2, 0.3, 0.9, 0.2, 0.7]
y_true = np.array(actual)
y_pred = np.array(predicted)
heaviside_new = np.vectorize(lambda x : 0.0 if x<0 else (.5 if x == 0 else 1.0))
heaviside = np.vectorize(lambda x : 0 if x<0 else .5 if x == 0 else 1)  #buggy

def Broadcast_AUC(y_true, y_pred):
    #the predictions of our classifier for all the positive labeled data
    pos_pred = y_pred[ y_true == 1]

    #the predictions of our classifier for all the negative labeled data
    neg_pred = y_pred[ y_true == 0]
    #creates a matrix that have pairwise difference between all the observations
    pairwise_matrix = pos_pred[:, np.newaxis] - neg_pred
    transform = heaviside(pairwise_matrix)
    return transform.mean(), transform

def Broadcast_new_AUC(y_true, y_pred):
    #the predictions of our classifier for all the positive labeled data
    pos_pred = y_pred[ y_true == 1]

    #the predictions of our classifier for all the negative labeled data
    neg_pred = y_pred[ y_true == 0]
    #creates a matrix that have pairwise difference between all the observations
    pairwise_matrix = pos_pred[:, np.newaxis] - neg_pred
    transform = heaviside_new(pairwise_matrix)
    return transform.mean(), transform

def Loop_AUC(y_true, y_pred):
    #the predictions of our classifier for all the positive labeled data
    pos_pred = y_pred[ y_true == 1]
    P = len(pos_pred) #the number of the positive population

    #the predictions of our classifier for all the negative labeled data
    neg_pred = y_pred[ y_true == 0]
    N = len(neg_pred) #the number of the negative population

    AUC = 0
    matrix = np.ones((len(pos_pred), len(neg_pred)))
    for p_idx, pos in enumerate(pos_pred):
        for n_idx, neg in enumerate(neg_pred):
            AUC +=  heaviside(pos - neg)
            matrix[p_idx, n_idx] = heaviside(pos - neg)

    return AUC/(P*N), matrix

loop_auc, loop_matrix = Loop_AUC(y_true, y_pred)
b_auc, b_matrix = Broadcast_AUC(y_true, y_pred)
b_new_auc, b_new_matrix = Broadcast_new_AUC(y_true, y_pred)
print 'sklearn', roc_auc_score(y_true, y_pred)
print 'loop_auc', loop_auc
print 'b_auc', b_auc #buggy output
print 'b_new_auc', b_new_auc