def rvs_score(self, theta, theta_score, n, random_state=None):
logging.debug(
'Simulating %s evolutions for theta = %s, augmenting with joint score',
n, theta)
rng = check_random_state(random_state, use_autograd=True)
all_x = []
all_t_xz = []
all_t_xz_checkpoints = []
all_histories = []
for i in range(n):
logging.debug(' Starting sample %s of %s', i + 1, n)
_, t_xz, time_series, _, t_xz_checkpoints = self._d_simulate_until_success(
theta, rng, theta_score)
all_t_xz.append(t_xz)
all_t_xz_checkpoints.append(t_xz_checkpoints)
all_histories.append(time_series)
x = self._calculate_observables(time_series, rng=rng)
all_x.append(x)
all_x = np.asarray(all_x)
all_t_xz = np.asarray(all_t_xz)
all_t_xz_checkpoints = np.asarray(all_t_xz_checkpoints)
all_histories = np.asarray(all_histories)
return all_x, all_t_xz, all_histories, all_t_xz_checkpoints
def rvs_ratio(self, theta, theta0, theta1, n, random_state=None):
logging.debug(
'Simulating %s evolutions for theta = %s, augmenting with joint ratio between %s and %s',
n, theta, theta0, theta1)
rng = check_random_state(random_state, use_autograd=True)
all_x = []
all_log_r_xz = []
all_log_r_xz_checkpoints = []
all_histories = []
for i in range(n):
logging.debug(' Starting sample %s of %s', i + 1, n)
log_p_xzs, time_series, log_p_xz_checkpoints = self._simulate_until_success(
theta, rng, [theta0, theta1])
all_log_r_xz.append(log_p_xzs[0] - log_p_xzs[1])
all_log_r_xz_checkpoints.append(log_p_xz_checkpoints[:, 0] -
log_p_xz_checkpoints[:, 1])
all_histories.append(time_series)
x = self._calculate_observables(time_series, rng=rng)
all_x.append(x)
all_x = np.asarray(all_x)
all_r_xz = np.exp(np.asarray(all_log_r_xz))
all_r_xz_checkpoints = np.exp(np.asarray(all_log_r_xz_checkpoints))
all_histories = np.asarray(all_histories)
return all_x, all_r_xz, all_histories, all_r_xz_checkpoints
def rvs(self, theta, n, random_state=None, return_histories=False):
logging.debug('Simulating %s epidemic evolutions for theta = %s', n,
theta)
rng = check_random_state(random_state)
all_x = []
histories = []
for i in range(n):
if return_histories:
_, (state,
history) = self._simulate_transmission(theta,
rng,
return_history=True)
histories.append(history)
else:
_, state = self._simulate_transmission(theta,
rng,
return_history=False)
x = self._calculate_observables(state)
all_x.append(x)
all_x = np.asarray(all_x)
if return_histories:
return all_x, histories
return all_x
def rvs_ratio_score(self,
theta, theta0, theta1, theta_score,
n, random_state=None):
""" Draws samples x according to p(x|theta), augments them with joint ratio r(x,z|theta0,theta1) and
joint score t(x,z|theta_score) """
rng = check_random_state(random_state)
all_x = []
all_log_r_xz = []
all_t_xz = []
for i in range(n):
u = rng.rand(self.n_rows)
log_p_xz_theta0, x = self.trace(theta0, u, theta_ref=theta)
log_p_xz_theta1, _ = self.trace(theta1, u, theta_ref=theta)
t_xz, _ = self.d_trace(theta_score, u, theta_ref=theta)
all_x.append(x)
all_log_r_xz.append(log_p_xz_theta0 - log_p_xz_theta1)
all_t_xz.append(t_xz)
all_x = np.array(all_x)
all_r_xz = np.exp(np.array(all_log_r_xz))
all_t_xz = np.array(all_t_xz)
return all_x, all_r_xz, all_t_xz
def rvs(self, theta, n, random_state=None, return_z_checkpoints=False, return_histories=False):
logging.debug('Simulating %s evolutions for theta = %s', n, theta)
rng = check_random_state(random_state)
all_x = []
all_z_checkpoints = []
all_histories = []
for i in range(n):
logging.debug(' Starting sample %s of %s', i + 1, n)
if return_histories:
_, time_series, _, history = self._simulate_until_success(theta, rng, return_history=True)
else:
_, time_series, _ = self._simulate_until_success(theta, rng)
if return_z_checkpoints or return_histories:
all_z_checkpoints.append(time_series)
if return_histories:
all_histories.append(history)
x = self._calculate_observables(time_series, rng=rng)
all_x.append(x)
all_x = np.asarray(all_x)
all_z_checkpoints = np.asarray(all_z_checkpoints)
if return_histories:
all_z_checkpoints = np.asarray(all_z_checkpoints)
return all_x, all_z_checkpoints, all_histories
if return_z_checkpoints:
return all_x, all_z_checkpoints
return all_x
def rvs(self,
theta,
n, random_state=None):
""" Draws samples x according to p(x|theta) """
rng = check_random_state(random_state)
all_x = []
for i in range(n):
u = rng.rand(self.n_rows)
_, x = self.trace(theta, u)
all_x.append(x)
return all_x
def rvs_ratio_score(self, theta, theta0, theta1, theta_score, n, random_state=None, return_histories=False):
logging.debug('Simulating %s evolutions for theta = %s, augmenting with joint ratio between %s and %s and joint'
' score at %s', n, theta, theta0, theta1, theta_score)
rng = check_random_state(random_state, use_autograd=True)
all_x = []
all_log_r_xz = []
all_t_xz = []
all_t_xz_checkpoints = []
all_log_r_xz_checkpoints = []
all_z_checkpoints = []
all_histories = []
for i in range(n):
logging.debug(' Starting sample %s of %s', i + 1, n)
results = self._d_simulate_until_success(theta, rng, theta_score, [theta0, theta1],
return_history=return_histories)
if return_histories:
log_p_xzs, t_xz, time_series, log_p_xz_checkpoints, t_xz_checkpoints, history = results
else:
log_p_xzs, t_xz, time_series, log_p_xz_checkpoints, t_xz_checkpoints = results
all_t_xz.append(t_xz)
all_log_r_xz.append(log_p_xzs[0] - log_p_xzs[1])
all_t_xz_checkpoints.append(t_xz_checkpoints)
all_log_r_xz_checkpoints.append(log_p_xz_checkpoints[:, 0] - log_p_xz_checkpoints[:, 1])
all_z_checkpoints.append(time_series)
if return_histories:
all_histories.append(history)
x = self._calculate_observables(time_series, rng=rng)
all_x.append(x)
all_x = np.asarray(all_x)
all_t_xz = np.asarray(all_t_xz)
all_r_xz = np.exp(np.asarray(all_log_r_xz))
all_r_xz_checkpoints = np.exp(np.asarray(all_log_r_xz_checkpoints))
all_t_xz_checkpoints = np.asarray(all_t_xz_checkpoints)
all_z_checkpoints = np.asarray(all_z_checkpoints)
if return_histories:
return all_x, all_r_xz, all_t_xz, all_z_checkpoints, all_r_xz_checkpoints, all_t_xz_checkpoints, all_histories
return all_x, all_r_xz, all_t_xz, all_z_checkpoints, all_r_xz_checkpoints, all_t_xz_checkpoints
def rvs(self, theta, n, random_state=None, return_histories=False):
logging.debug('Simulating %s evolutions for theta = %s', n, theta)
rng = check_random_state(random_state)
all_x = []
histories = []
for i in range(n):
logging.debug(' Starting sample %s of %s', i + 1, n)
_, time_series, _ = self._simulate_until_success(theta, rng)
if return_histories:
histories.append(time_series)
x = self._calculate_observables(time_series, rng=rng)
all_x.append(x)
all_x = np.asarray(all_x)
if return_histories:
return all_x, histories
return all_x
def rvs_score(self,
theta,
theta_score,
n,
random_state=None,
return_histories=False):
logging.debug(
'Simulating %s epidemic evolutions for theta = %s, augmenting with joint score',
n, theta)
rng = check_random_state(random_state)
all_x = []
all_t_xz = []
histories = []
for i in range(n):
if return_histories:
t_xz, (state, history) = self._d_simulate_transmission(
theta, rng, return_history=True)
histories.append(history)
else:
t_xz, state = self._d_simulate_transmission(
theta, rng, return_history=False)
x = self._calculate_observables(state)
all_x.append(x)
all_t_xz.append(t_xz)
all_x = np.asarray(all_x)
all_t_xz = np.asarray(all_t_xz)
if return_histories:
return all_x, all_t_xz, histories
return all_x, all_t_xz