def test_load_and_save(self):
params1 = np.zeros((2, 4))
weights1 = np.zeros((2, 4))
journal = Journal(0)
journal.add_parameters(params1)
journal.add_weights(weights1)
journal.save('journal_tests_testfile.pkl')
new_journal = Journal.fromFile('journal_tests_testfile.pkl')
np.testing.assert_equal(journal.parameters, new_journal.parameters)
np.testing.assert_equal(journal.weights, new_journal.weights)
def test_add_parameters(self):
params1 = np.zeros((2, 4))
params2 = np.ones((2, 4))
# test whether production mode only stores the last set of parameters
journal_prod = Journal(0)
journal_prod.add_parameters(params1)
journal_prod.add_parameters(params2)
self.assertEqual(len(journal_prod.parameters), 1)
np.testing.assert_equal(journal_prod.parameters[0], params2)
# test whether reconstruction mode stores all parameter sets
journal_recon = Journal(1)
journal_recon.add_parameters(params1)
journal_recon.add_parameters(params2)
self.assertEqual(len(journal_recon.parameters), 2)
np.testing.assert_equal(journal_recon.parameters[0], params1)
np.testing.assert_equal(journal_recon.parameters[1], params2)
def sample(self,
observations,
steps,
epsilon,
n_samples=10000,
n_samples_per_param=1,
beta=2,
delta=0.2,
v=0.3,
ar_cutoff=0.05,
resample=None,
n_update=None,
adaptcov=1,
full_output=0):
"""Samples from the posterior distribution of the model parameter given the observed
data observations.
Parameters
----------
observations : numpy.ndarray
Observed data.
steps : integer
Number of maximum iterations in the sequential algoritm ("generations")
epsilon : numpy.float
An array of proposed values of epsilon to be used at each steps.
n_samples : integer, optional
Number of samples to generate. The default value is 10000.
n_samples_per_param : integer, optional
Number of data points in each simulated data set. The default value is 1.
beta : numpy.float
Tuning parameter of SABC
delta : numpy.float
Tuning parameter of SABC
v : numpy.float, optional
Tuning parameter of SABC, The default value is 0.3.
ar_cutoff : numpy.float
Acceptance ratio cutoff, The default value is 0.5
resample: int, optional
Resample after this many acceptance, The default value if n_samples
n_update: int, optional
Number of perturbed parameters at each step, The default value if n_samples
adaptcov : boolean, optional
Whether we adapt the covariance matrix in iteration stage. The default value TRUE.
full_output: integer, optional
If full_output==1, intermediate results are included in output journal.
The default value is 0, meaning the intermediate results are not saved.
Returns
-------
abcpy.output.Journal
A journal containing simulation results, metadata and optionally intermediate results.
"""
journal = Journal(full_output)
journal.configuration["type_model"] = type(self.model)
journal.configuration["type_dist_func"] = type(self.distance)
journal.configuration["type_kernel_func"] = type(self.kernel)
journal.configuration["n_samples"] = n_samples
journal.configuration["n_samples_per_param"] = n_samples_per_param
journal.configuration["beta"] = beta
journal.configuration["delta"] = delta
journal.configuration["v"] = v
journal.configuration["ar_cutoff"] = ar_cutoff
journal.configuration["resample"] = resample
journal.configuration["n_update"] = n_update
journal.configuration["adaptcov"] = adaptcov
journal.configuration["full_output"] = full_output
accepted_parameters = np.zeros(
shape=(n_samples, len(self.model.get_parameters())))
distances = np.zeros(shape=(n_samples, ))
smooth_distances = np.zeros(shape=(n_samples, ))
accepted_weights = np.ones(shape=(n_samples, 1))
all_distances = None
accepted_cov_mat = None
if resample == None:
resample = n_samples
if n_update == None:
n_update = n_samples
sample_array = np.ones(shape=(steps, ))
sample_array[0] = n_samples
sample_array[1:] = n_update
## Acceptance counter to determine the resampling step
accept = 0
samples_until = 0
# Initialize variables that need to be available remotely
rc = _RemoteContextSABCDiffusion(self.backend, self.model, self.distance, self.kernel, \
observations, n_samples, n_samples_per_param)
for aStep in range(0, steps):
# main SABC algorithm
# print("INFO: Initialization of SABC")
seed_arr = self.rng.randint(0,
np.iinfo(np.uint32).max,
size=int(sample_array[aStep]),
dtype=np.uint32)
seed_pds = self.backend.parallelize(seed_arr)
# 0: update remotely required variables
# print("INFO: Broadcasting parameters.")
rc.epsilon = epsilon
rc._update_broadcasts(self.backend, accepted_parameters,
accepted_cov_mat, smooth_distances,
all_distances)
# 1: Calculate parameters
# print("INFO: Initial accepted parameter parameters")
params_and_dists_pds = self.backend.map(rc._accept_parameter,
seed_pds)
params_and_dists = self.backend.collect(params_and_dists_pds)
new_parameters, new_distances, new_all_parameters, new_all_distances, index, acceptance = [
list(t) for t in zip(*params_and_dists)
]
new_parameters = np.array(new_parameters)
new_distances = np.array(new_distances)
new_all_distances = np.concatenate(new_all_distances)
index = np.array(index)
acceptance = np.array(acceptance)
# print('parallel stuff finsihed')
# Reading all_distances at Initial step
if aStep == 0:
index = np.linspace(0, n_samples - 1,
n_samples).astype(int).reshape(
n_samples, )
accept = 0
all_distances = new_all_distances
# print(index[acceptance == 1])
# Initialize/Update the accepted parameters and their corresponding distances
accepted_parameters[index[acceptance == 1], :] = new_parameters[
acceptance == 1, :]
distances[index[acceptance == 1]] = new_distances[acceptance == 1]
# 2: Smoothing of the distances
smooth_distances[index[acceptance == 1]] = self._smoother_distance(
distances[index[acceptance == 1]], all_distances)
# 3: Initialize/Update U, epsilon and covariance of perturbation kernel
if aStep == 0:
U = self._avergae_redefined_distance(
self._smoother_distance(all_distances, all_distances),
epsilon)
else:
U = np.mean(smooth_distances)
epsilon = self._schedule(U, v)
# if accepted_parameters.shape[1] > 1:
# accepted_cov_mat = beta*np.cov(np.transpose(accepted_parameters)) + \
# 0.0001*np.trace(np.cov(np.transpose(accepted_parameters)))*np.eye(accepted_parameters.shape[1])
# else:
accepted_cov_mat = beta* np.cov(np.transpose(accepted_parameters[:, [0, 1]])) + \
0.0001 * (np.cov(np.transpose(accepted_parameters[:, [0, 1]]))) * np.eye(1)
## 4: Show progress and if acceptance rate smaller than a value break the iteration
# print("INFO: Saving intermediate configuration to output journal.")
if full_output == 1:
journal.add_parameters(accepted_parameters)
journal.add_weights(accepted_weights)
if aStep > 0:
accept = accept + np.sum(acceptance)
samples_until = samples_until + sample_array[aStep]
acceptance_rate = accept / samples_until
print('updates: ', np.sum(sample_array[1:aStep + 1]) / np.sum(sample_array[1:]) * 100, ' epsilon: ',
epsilon, \
'u.mean: ', U, 'acceptance rate: ', acceptance_rate)
if acceptance_rate < ar_cutoff:
break
# 5: Resampling if number of accepted particles greater than resample
if accept >= resample and U > 1e-100:
## Weighted resampling:
weight = np.exp(-smooth_distances * delta / U)
weight = weight / sum(weight)
index_resampled = self.rng.choice(np.arange(n_samples),
n_samples,
replace=1,
p=weight)
accepted_parameters = accepted_parameters[index_resampled, :]
smooth_distances = smooth_distances[index_resampled]
## Update U and epsilon:
epsilon = epsilon * (1 - delta)
U = np.mean(smooth_distances)
epsilon = self._schedule(U, v)
## Print effective sampling size
print('Resampling: Effective sampling size: ',
1 / sum(pow(weight / sum(weight), 2)))
accept = 0
samples_until = 0
# Add epsilon_arr, number of final steps and final output to the journal
# print("INFO: Saving final configuration to output journal.")
if full_output == 0:
journal.add_parameters(accepted_parameters)
journal.add_weights(accepted_weights)
journal.configuration["steps"] = aStep + 1
journal.configuration["epsilon"] = epsilon
return journal
