machine learning with python

import numpy as np from sklearn.cluster import KMeans from getUsers import loadSingleData def compute_k_means(data, K=5): kmeans = KMeans(n_clusters=K, random_state=0).fit(data) return kmeans.labels_, kmeans.cluster_centers_ def get_data_clusters(cluster_assignments, cluster_centers, data): k = cluster_centers.shape[0] m = data.shape[0] return [np.asarray([data[j] for j in range(m) if cluster_assignments[j] == i]) for i in range(k)] def generate_multivariate_gaussian_attack(cluster_assignments, cluster_centers, data, counts, count_threshold=5, tile_fn=None, sample_guassian=True, sample_num=1): ''' Generate one attack per mean using multivariate gaussian ''' k = cluster_centers.shape[0] clusters = get_data_clusters(cluster_assignments, cluster_centers, data) attacks = [] for i in range(k): if counts[i] >= count_threshold: mean = cluster_centers[i] attack = [mean] if sample_guassian: cov = np.cov(clusters[i].T) # Add small epsilon to cov is cov is too small min_eig = np.min(np.real(np.linalg.eigvals(cov))) if min_eig < 0: cov -= 10 * min_eig * np.eye(*cov.shape) attack = [np.random.multivariate_normal(mean, cov) for _ in range(sample_num)] if tile_fn: attack = tile_fn(clusters[i], attack) attacks.extend(attack) return np.asarray(attacks) def tile_attacks(data, attacks): n = data.shape[1] tiled_attacks = [] for d in data: for a in attacks: example = np.concatenate((d, a)) tiled_attacks.append(example) return np.asarray(tiled_attacks) def generate_tiled_k_means_attack(K, sample_guassian=False, sample_num=1, data_file='keystroke.csv'): labels, data = loadSingleData(data_file) cluster_assignments, cluster_centers = compute_k_means(data, K=K) counts = [np.sum(cluster_assignments == i) for i in range(cluster_centers.shape[0])] attacks = generate_multivariate_gaussian_attack(cluster_assignments, cluster_centers, data, counts, tile_fn=tile_attacks, sample_guassian=sample_guassian, sample_num=sample_num) print(counts) return attacks def main(): filename = 'keystroke.csv' labels, data = loadSingleData(filename) cluster_assignments, cluster_centers = compute_k_means(data, K=5) counts = [np.sum(cluster_assignments == i) for i in range(cluster_centers.shape[0])] attacks = generate_multivariate_gaussian_attack(cluster_assignments, cluster_centers, data, counts, tile_fn=tile_attacks) print(attacks.shape) if __name__ == '__main__': main()