def generate_samples():
model = HMM.from_fixed_params()
T = 10
Z, X = model.sample(T)
print(model)
print(f"X: {X}, True Z: {Z}")
smcmc = SMCMC(model, T)
samples = smcmc.sample(N=100, x_sequence=X)
print(f"Last L samples: \n{samples[:, -1]}")
# print(np.unique(samples[:, 0], return_counts=True)[1] / len(samples))
print(f"Mean sample: {samples.mean((0, 1)).round(1)}")
Z_samples = np.empty((N, T), dtype=int)
Z_samples[0] = self.model.sample_Z(T)
for n in range(1, N):
z_proposal = self.model.sample_Z(T)
log_alpha = compute_log_alpha(z_proposal, Z_samples[n - 1], x_sequence, self.model)
self.acceptance_probs.append(np.exp(min(0., log_alpha)))
if np.random.rand() <= self.acceptance_probs[-1]:
Z_samples[n] = z_proposal
else:
Z_samples[n] = Z_samples[n - 1]
return Z_samples
def plot_acceptance(self):
fig, ax = plt.subplots(1, 1)
ax.plot(self.acceptance_probs)
plt.show()
if __name__ == '__main__':
model = HMM.from_fixed_params()
T = 5
Z, X = model.sample(10, 10, None)
mh = MH(model, T)
print(model)
print(X, Z)
samples = mh.sample(X)
print(samples[-10:])
# mh.plot_acceptance()
# print(np.array(mh.acceptance_probs))
print(samples.mean(0).round())
