e_recall.py

import numpy as np


class Metric():
    def __init__(self, k, **kwargs):
        self.k = k
        self.requires = ['nearest_features@{}'.format(k), 'target_labels']
        self.name = 'e_recall@{}'.format(k)

    def __call__(self, target_labels, k_closest_classes):
        recall_at_k = np.sum([
            1 for target, recalled_predictions in zip(target_labels,
                                                      k_closest_classes)
            if target in recalled_predictions[:self.k]
        ]) / len(target_labels)
        return recall_at_k