def cluster(X, K, divergence, debug=False):
if divergence == 'KL':
dist_cls = distributions.KL
if np.any(X <= 0): # for MFCCs...
X = X - X.min() + 1e-8
X = 5. * X / X.sum(1)[:,nax]
elif divergence == 'IS':
dist_cls = distributions.ItakuraSaito
if np.any(X <= 0): # for MFCCs...
X = X - X.min() + 1e-8
# X = X / X.sum(1)[:,nax]
elif divergence == 'EU':
dist_cls = distributions.SquareDistance
else:
print 'Wrong divergence'
sys.exit(0)
assignments, centroids, _ = kmeans.kmeans_best_of_n(X, K, n_trials=10,
dist_cls=dist_cls, debug=debug)
init_pi = np.ones(K) / K
init_obs_distr = centroids
tau_em, obs_distr, pi, em_ll_train, _ = em.em(X, centroids, n_iter=10)
# tau_hmm, A, obs_distr, pi, ll_train, _ = hmm.em_hmm(X, init_pi, init_obs_distr, n_iter=10)
# seq_hmm, _ = hmm.viterbi(X, pi, A, obs_distr)
Tracer()()
return {'kmeans': assignments,
'EM': np.argmax(tau_em, axis=1),
# 'hmm_smoothing': np.argmax(tau_hmm, axis=1),
# 'hmm_viterbi': seq_hmm,
}
ll_test.append(log_likelihood(lalpha_test, lbeta_test))
if monitor:
monitor_vals.append(monitor(A, obs_distr))
Tracer()()
return seq, A, obs_distr, ll_test, monitor_vals
if __name__ == '__main__':
X = np.loadtxt('EMGaussian.data')
Xtest = np.loadtxt('EMGaussian.test')
K = 4
# Run simple EM (no HMM)
iterations = 40
assignments, centers, _ = kmeans.kmeans_best_of_n(X, K, n_trials=5)
new_centers = [distributions.Gaussian(c.mean, np.eye(2)) \
for c in centers]
tau, obs_distr, pi, gmm_ll_train, gmm_ll_test = \
em.em(X, new_centers, assignments, n_iter=iterations, Xtest=Xtest)
# example with fixed parameters
A = 1. / 6 * np.ones((K, K))
A[np.diag(np.ones(K)) == 1] = 0.5
lalpha, lbeta = alpha_beta(Xtest, pi, A, obs_distr)
log_p = smoothing(lalpha, lbeta)
p = np.exp(log_p)
def plot_traj(p):
plt.figure()
# algos to run
algs = map(int, options.algos.split(','))
# number of clusters
K = options.k
ass_plots = []
seqs = {}
results = {} # data/results obtained, indexed by algorithm
# K-means
if options.init == 'kmeans' or algos.kmeans in algs:
t = time.time()
assignments, centroids, dists = \
kmeans.kmeans_best_of_n(X, K, n_trials=4, dist_cls=distributions.KL)
print 'K-means: {}s'.format(time.time() - t)
results[algos.kmeans] = {
'seq': assignments,
'centroids': centroids,
}
seqs[algos.kmeans] = assignments
# EM
if options.init == 'em' or algos.em in algs:
iterations = 10
t = time.time()
tau_em, obs_distr, pi, em_ll_train, _ = em.em(X,
centroids,
n_iter=options.n_iter)
print 'EM: {}s'.format(time.time() - t)
for j in range(K):
x, y = np.arange(-10., 10., 0.04), np.arange(-15., 15., 0.04)
xx, yy = np.meshgrid(x, y)
sx = np.sqrt(sigmas[j][0,0])
sy = np.sqrt(sigmas[j][1,1])
sxy = sigmas[j][1,0]
z = bivariate_normal(xx, yy, sx, sy, means[j,0], means[j,1], sxy)
cs = plt.contour(xx, yy, z, [0.01])
if __name__ == '__main__':
X = np.loadtxt('EMGaussian.data')
Xtest = np.loadtxt('EMGaussian.test')
K = 4
iterations = 40
assignments, centers, _ = kmeans.kmeans_best_of_n(X, K, n_trials=5)
for k in range(K):
centers[k].sigma2 = 1.
# Isotropic
tau, obs_distr, pi, ll_train_iso, ll_test_iso = \
em(X, centers, assignments, n_iter=iterations, Xtest=Xtest)
plot_em(X, tau, obs_distr, contours=True)
plt.title('EM with covariance matrices proportional to identity')
# General
new_centers = [distributions.Gaussian(c.mean, c.sigma2*np.eye(2)) \
for c in centers]
tau, obs_distr, pi, ll_train_gen, ll_test_gen = \
em(X, new_centers, assignments, n_iter=iterations, Xtest=Xtest)
plot_em(X, tau, obs_distr, contours=True)