import numpy as np

import sklearn
import sklearn.metrics


def silhouette_score(x, labels):
    if np.unique(labels).shape[0] == 1:
        return 0

    s = np.zeros(x.shape[0])
    clusters = []

    for label in np.unique(labels):
        if (labels == label).shape[0] == 1:
            s[(labels == label)] = 0
        else:
            values = np.sum(sklearn.metrics.pairwise_distances(x, x)[:, (labels == label)], axis=1)
            s[(labels == label) * (np.sum(labels == label) > 1)] = values[(labels == label) * (np.sum(labels == label) > 1)] / (np.sum(labels == label) - 1)
            avg = values / np.sum(labels == label)
            avg[(labels == label)] = np.max(sklearn.metrics.pairwise_distances(x, x)) + 1
            clusters.append(avg)

    p = np.min(np.stack(clusters), axis=0)
    m = (p != 0) * (s != 0)

    sil = np.zeros(x.shape[0])
    sil[m] = (p - s)[m] / np.maximum(p, s)[m]
    sil_score = np.mean(sil, axis=0)
    return sil_score


def bcubed_score(true_labels, predicted_labels):
    n_objects = true_labels.shape[0]
    precision = 0
    recall = 0
    for i in range(n_objects):
        p = true_labels[(predicted_labels == predicted_labels[i])]
        precision += np.mean(p == true_labels[i])
        r = predicted_labels[(true_labels == true_labels[i])]
        recall += np.mean(r == predicted_labels[i])
    precision /= n_objects
    recall /= n_objects
    score = 2 * (precision * recall) / (precision + recall)
    return score