def test_named_models_are_unsupported(self):
def normal_sim(a, b):
return np.random.normal(a, b, 1000)
with pm.Model(name="NamedModel"):
a = pm.Normal("a", mu=0, sigma=1)
b = pm.HalfNormal("b", sigma=1)
c = pm.Potential("c", pm.math.switch(a > 0, 0, -np.inf))
s = pm.Simulator(
"s", normal_sim, params=(a, b), sum_stat="sort", epsilon=1, observed=self.data
)
with pytest.raises(NotImplementedError, match="named models"):
pm.sample_smc(draws=10, kernel="ABC")
def Two_Parameter_Model(Path, N_people, SEED, Max_pump=128):
'''
Path: Path where your record file exist.
N_people: The Number of participants you have.
SEED: Set radnom seed of MCMC sampling.
Max_pump: Maxiumum trial of your BART setting. My setting is 128.
'''
traces_gamma = []
traces_beta = []
for part in range(1, N_people + 1):
with pm.Model() as total_model:
start_time = time.time()
p = 0.15
obs = BernData(part, Max_pump, Path)
#Pth participant
gamma_plus = pm.Uniform("gamma_plus", 0, 10)
beta = pm.Uniform("beta", 0, 10)
omega_k = -gamma_plus / np.log(1 - p)
for i in range(len(obs)):
for l in range(Max_pump):
theta_lk = 1 / (1 + np.exp(beta * (l - omega_k)))
prob = pm.Bernoulli("prob_{}_{}".format(i, l),
p=theta_lk,
observed=obs[i][l])
_trace = pm.sample_smc(1000, cores=6, random_seed=SEED)
print("Sampling end:", part,
"--- %s seconds ---" % (time.time() - start_time))
traces_gamma.append(_trace["gamma_plus"])
traces_beta.append(_trace["beta"])
return traces_gamma, traces_beta
def test_sample(self):
with self.SMC_test:
mtrace = pm.sample_smc(draws=self.samples)
x = mtrace["X"]
mu1d = np.abs(x).mean(axis=0)
np.testing.assert_allclose(self.muref, mu1d, rtol=0.0, atol=0.03)
def sample_model(self, method = 'log-model', field = 'deaths', **kwargs):
models = self.phenom_model(method, field)
self.traces = {}
self.traces[method] = {}
for i, country in enumerate(self.countries):
self.traces[method][country] = pm.sample_smc(model = models[method][country], **kwargs)
#marginal_likelihood = model.marginal_log_likelihood
return models, self.traces #, marginal_likelihood
def run_bayesian_inference_gaussian_error_model(loglike,
variables,
ndraws,
nburn,
njobs,
algorithm='nuts',
get_map=False,
print_summary=False):
# create our Op
if algorithm != 'nuts':
logl = LogLike(loglike)
else:
logl = LogLikeWithGrad(loglike)
# use PyMC3 to sampler from log-likelihood
with pm.Model():
# must be defined inside with pm.Model() block
pymc_variables, pymc_var_names = get_pymc_variables(
variables.all_variables())
# convert m and c to a tensor vector
theta = tt.as_tensor_variable(pymc_variables)
# use a DensityDist (use a lamdba function to "call" the Op)
pm.DensityDist('likelihood', lambda v: logl(v), observed={'v': theta})
if get_map:
map_sample_dict = pm.find_MAP()
if algorithm == 'smc':
assert njobs == 1 # njobs is always 1 when using smc
trace = pm.sample_smc(ndraws)
else:
if algorithm == 'metropolis':
step = pm.Metropolis(pymc_variables)
elif algorithm == 'nuts':
step = pm.NUTS(pymc_variables)
trace = pm.sample(ndraws,
tune=nburn,
discard_tuned_samples=True,
start=None,
cores=njobs,
step=step)
if print_summary:
print(pm.summary(trace))
samples, effective_sample_size = extract_mcmc_chain_from_pymc3_trace(
trace, pymc_var_names, ndraws, nburn, njobs)
if get_map:
map_sample = extract_map_sample_from_pymc3_dict(
map_sample_dict, pymc_var_names)
else:
map_samples = None
return samples, effective_sample_size, map_sample
def test_one_gaussian(self):
with self.SMABC_test:
trace = pm.sample_smc(draws=1000, kernel="ABC")
np.testing.assert_almost_equal(self.data.mean(),
trace["a"].mean(),
decimal=2)
np.testing.assert_almost_equal(self.data.std(),
trace["b"].mean(),
decimal=1)
def test_start(self):
with pm.Model() as model:
a = pm.Poisson("a", 5)
b = pm.HalfNormal("b", 10)
y = pm.Normal("y", a, b, observed=[1, 2, 3, 4])
start = {
"a": np.random.poisson(5, size=500),
"b_log__": np.abs(np.random.normal(0, 10, size=500)),
}
trace = pm.sample_smc(500, start=start)
def test_ml(self):
data = np.repeat([1, 0], [50, 50])
marginals = []
a_prior_0, b_prior_0 = 1.0, 1.0
a_prior_1, b_prior_1 = 20.0, 20.0
for alpha, beta in ((a_prior_0, b_prior_0), (a_prior_1, b_prior_1)):
with pm.Model() as model:
a = pm.Beta("a", alpha, beta)
y = pm.Bernoulli("y", a, observed=data)
trace = pm.sample_smc(2000)
marginals.append(model.marginal_log_likelihood)
# compare to the analytical result
assert abs(np.exp(marginals[1] - marginals[0]) - 4.0) <= 1
def test_one_gaussian(self):
with self.SMABC_test:
trace = pm.sample_smc(draws=2000,
kernel="ABC",
epsilon=0.1,
n_steps=25)
np.testing.assert_almost_equal(
[self.data.mean() * 10, self.data.std()],
[trace["a"].mean() * 10, trace["b"].mean()],
decimal=1)
np.testing.assert_almost_equal(self.data.std(),
trace["b"].mean(),
decimal=1)
def test_sim_data_ppc(self):
with self.SMABC_test:
trace, sim_data = pm.sample_smc(draws=1000, kernel="ABC", chains=2, save_sim_data=True)
pr_p = pm.sample_prior_predictive(1000)
po_p = pm.sample_posterior_predictive(trace, 1000)
assert sim_data["s"].shape == (2, 1000, 1000)
np.testing.assert_almost_equal(self.data.mean(), sim_data["s"].mean(), decimal=2)
np.testing.assert_almost_equal(self.data.std(), sim_data["s"].std(), decimal=1)
assert pr_p["s"].shape == (1000, 1000)
np.testing.assert_almost_equal(0, pr_p["s"].mean(), decimal=1)
np.testing.assert_almost_equal(1.4, pr_p["s"].std(), decimal=1)
assert po_p["s"].shape == (1000, 1000)
np.testing.assert_almost_equal(0, po_p["s"].mean(), decimal=2)
np.testing.assert_almost_equal(1, po_p["s"].std(), decimal=1)
def test_discrete_continuous(self):
with pm.Model() as model:
a = pm.Poisson("a", 5)
b = pm.HalfNormal("b", 10)
y = pm.Normal("y", a, b, observed=[1, 2, 3, 4])
trace = pm.sample_smc()
f = pm.Bound(pm.Beta, lower=0., upper=1.0)('f', alpha=1.1, beta=1.1)
sigma = pm.HalfNormal('sigma', sd=1)
forward = th_forward_model(tauD, tauF, U, f)
cov = np.eye(2) * sigma**2
Y_obs = pm.MvNormal('Y_obs', mu=forward, cov=cov, observed=Y_sim)
startsmc = {
v.name: np.random.uniform(1e-3, 2, size=draws)
for v in TM_model.free_RVs
}
trace_TM = pm.sample_smc(draws, start=startsmc)
# %%
pm.plot_posterior(trace_TM, kind='hist', bins=30, color='seagreen')
# %%
# %% FROM SCRATCH
###############################################################################
###############################################################################
# %%
## %% TM-GLM inference
#def log_prior_theta(gamma1,gamma2,beta1,beta2):
# D, F = np.random.gamma(gamma1, gamma1)
# U, f = np.ranodm.beta(1.01, 1.01, 2)
# theta = D,F,U,f
def test_potential(self):
with self.SMABC_potential:
trace = pm.sample_smc(draws=1000, kernel="ABC")
assert np.all(trace["a"] >= 0)
def test_custom_dist_sum(self):
with self.SMABC_test2:
trace = pm.sample_smc(draws=1000, kernel="ABC")
d_I,
epsilon_I_deterministic,
rho_deterministic,
omega,
sigma_deterministic,
),
)
likelihood_model = pm.Normal("likelihood_model",
mu=fitting_model,
sigma=standard_deviation,
observed=observations_to_fit)
seirdpq_trace_calibration = pm.sample_smc(draws=draws,
n_steps=25,
parallel=True,
cores=int(os.cpu_count()),
progressbar=True,
random_seed=seed)
duration = time.time() - start_time
print(f"-- Monte Carlo simulations done in {duration / 60:.3f} minutes")
# %%
print("-- Arviz post-processing:")
import warnings
warnings.filterwarnings("ignore")
start_time = time.time()
plot_step = 1
def run_bayesian_inference_gaussian_error_model(loglike,
variables,
ndraws,
nburn,
njobs,
algorithm='nuts',
get_map=False,
print_summary=False,
loglike_grad=None,
seed=None):
r"""
Draw samples from the posterior distribution using Markov Chain Monte
Carlo for data that satisfies
.. math:: y=f(z)+\epsilon
where :math:`y` is a vector of observations, :math:`z` are the
parameters of a function which are to be inferred, and :math:`\epsilon`
is Gaussian noise.
Parameters
----------
loglike : pyapprox_dev.bayesian_inference.markov_chain_monte_carlo.GaussianLogLike
A log-likelihood function associated with a Gaussian error model
variables : pya.IndependentMultivariateRandomVariable
Object containing information of the joint density of the inputs z.
This is used to generate random samples from this join density
ndraws : integer
The number of posterior samples
nburn : integer
The number of samples to discard during initialization
njobs : integer
The number of prallel chains
algorithm : string
The MCMC algorithm should be one of
- 'nuts'
- 'metropolis'
- 'smc'
get_map : boolean
If true return the MAP
print_summary : boolean
If true print summary statistics about the posterior samples
loglike_grad : callable
Function with signature
``loglikegrad(z) -> np.ndarray (nvars)``
where ``z`` is a 2D np.ndarray with shape (nvars,nsamples
random_seed : int or list of ints
A list is accepted if ``cores`` is greater than one. PyMC3 does not
produce consistent results by setting numpy.random.seed instead
seed must be passed in
"""
# create our Op
if algorithm != 'nuts':
logl = LogLike(loglike)
else:
logl = LogLikeWithGrad(loglike, loglike_grad)
# use PyMC3 to sampler from log-likelihood
with pm.Model():
# must be defined inside with pm.Model() block
pymc_variables, pymc_var_names = get_pymc_variables(
variables.all_variables())
# convert m and c to a tensor vector
theta = tt.as_tensor_variable(pymc_variables)
# use a DensityDist (use a lamdba function to "call" the Op)
pm.DensityDist('likelihood', lambda v: logl(v), observed={'v': theta})
if get_map:
map_sample_dict = pm.find_MAP()
if algorithm == 'smc':
assert njobs == 1 # njobs is always 1 when using smc
trace = pm.sample_smc(ndraws)
else:
if algorithm == 'metropolis':
step = pm.Metropolis(pymc_variables)
elif algorithm == 'nuts':
step = pm.NUTS(pymc_variables)
trace = pm.sample(ndraws,
tune=nburn,
discard_tuned_samples=True,
start=None,
cores=njobs,
step=step,
compute_convergence_checks=False,
random_seed=seed)
# compute_convergence_checks=False avoids bugs in theano
if print_summary:
try:
print(pm.summary(trace))
except:
print('could not print summary. likely issue with theano')
samples, effective_sample_size = extract_mcmc_chain_from_pymc3_trace(
trace, pymc_var_names, ndraws, nburn, njobs)
if get_map:
map_sample = extract_map_sample_from_pymc3_dict(
map_sample_dict, pymc_var_names)
else:
map_samples = None
return samples, effective_sample_size, map_sample
def test_name_is_string_type(self):
with self.SMABC_potential:
assert not self.SMABC_potential.name
trace = pm.sample_smc(draws=10, kernel="ABC")
assert isinstance(trace._straces[0].name, str)
