return plotfunc
if __name__ == '__main__':
num_iterations = 500
num_samples = 100
weights = [1./5, 4./5]
mus = [np.array([-4]),
np.array([4])]
sigmas = [np.eye(1),
np.eye(1)]
toy_init_particles = sample_gmm(num_samples,
[1.],
[np.array([10])],
[np.eye(1)])
# a = -10
# b = 10
# toy_init_particles = np.random.rand(100).reshape(-1, 1)*(b - a) + a
true_samples = sample_gmm(num_samples, weights, mus, sigmas)
plotfunc = get_plotfunc(true_samples)
log_lik_func = log_lik_gmm(weights, mus, sigmas)
report_metrics = metrics(log_lik_func)
toy_score = gmm_gld(weights, mus, sigmas)
def metrics(log_lik_func):
def report_metrics(particles):
print("Log Likelihood: {}".format(log_lik_func(particles)))
return report_metrics
if __name__ == '__main__':
num_iterations = 5000
num_samples = 100
weights = [1]
mus = [np.array([0, 0])]
sigmas = [np.eye(2)]
init_particles = sample_gmm(num_samples, [1.], [np.array([-7, -7])],
[np.eye(2)])
true_samples = sample_gmm(num_samples, weights, mus, sigmas)
plotfunc = get_plotfunc(true_samples)
gld = gmm_gld(weights, mus, sigmas)
svgd = SVGD(gld=gld)
log_lik_func = log_lik_gmm(weights, mus, sigmas)
report_metrics = metrics(log_lik_func)
particles = svgd.do_svgd_iterations_optimized(
init_particles=init_particles,
plt.savefig('./plots/hmc/1d_multimodal_medium_equal_other_side.png')
plt.close()
return plotfunc
if __name__ == '__main__':
num_samples = 100
weights = [1. / 5, 4. / 5]
mus = [np.array([-4]), np.array([4])]
sigmas = [np.eye(1), np.eye(1)]
true_samples = sample_gmm(num_samples, weights, mus, sigmas)
plotfunc = get_plotfunc(true_samples)
log_pdf = log_gmm_pdf_vectorized(weights, mus, sigmas)
target_energy = lambda x: -log_pdf(x)
gld = gmm_gld(weights, mus, sigmas)
target_grad = lambda x: -gld(x)
#define other parameters
params = {
'u_energy': target_energy,
'u_grad': target_grad,