evaluation.py

'''
Created on Jan 21, 2020

@author: maria
'''

import numpy as np
from collections import defaultdict

import sklearn.metrics as metrics
from sklearn.utils.linear_assignment_ import linear_assignment

def evaluate_predictions(y_true, y_pred, scoring={'accuracy','precision','recall','nmi','adj_rand','f1_score','adj_mi'}):
    """
    Evaluation of predictions given true labels.
    y_true: true cluster labels
    y_pred: predicted cluster labels
    """
    scores = defaultdict(list)
    
    y_true, y_pred = hungarian_match(y_true, y_pred)
    set_scores(scores, y_true, y_pred, scoring)
    
    return scores
    

def set_scores(scores, y_true, y_pred, scoring):
    labels=list(set(y_true))
    
    for metric in scoring:
        if metric=='accuracy':
            scores[metric] = metrics.accuracy_score(y_true, y_pred)
        elif metric=='precision':
            scores[metric] = metrics.precision_score(y_true, y_pred, labels, average='macro')
        elif metric=='recall':
            scores[metric] = metrics.recall_score(y_true, y_pred, labels, average='macro')
        elif metric=='f1_score':
            scores[metric] = metrics.f1_score(y_true, y_pred, labels, average='macro')
        elif metric=='nmi':
            scores[metric] = metrics.normalized_mutual_info_score(y_true, y_pred)
        elif metric=='adj_mi':
            scores[metric] = metrics.adjusted_mutual_info_score(y_true, y_pred)
        elif metric=='adj_rand':
            scores[metric] = metrics.adjusted_rand_score(y_true, y_pred)
       

def hungarian_match(self, y_true, y_pred):
    """Matches predicted labels to original using hungarian algorithm."""
    
    y_true = self.adjust_range(y_true)
    y_pred = self.adjust_range(y_pred)
    
    D = max(y_pred.max(), y_true.max()) + 1
    w = np.zeros((D, D), dtype=np.int64)
    # Confusion matrix.
    for i in range(y_pred.size):
        w[y_pred[i], y_true[i]] += 1
    ind = linear_assignment(-w)
    d = {i:j for i, j in ind}
    y_pred = np.array([d[v] for v in y_pred])
    
    return y_true, y_pred


def adjust_range(self, y):
    """Assures that the range of indices if from 0 to n-1."""
    y = np.array(y, dtype=np.int64)
    val_set = set(y)
    mapping = {val:i for  i,val in enumerate(val_set)}
    y = np.array([mapping[val] for val in y], dtype=np.int64)
    return y