def _iter_sample(draws,
step,
start=None,
trace=None,
chain=0,
tune=None,
model=None,
random_seed=-1):
"""Modified from :func:`pymc3.sampling._iter_sample` to be more efficient with SMC algorithm."""
model = modelcontext(model)
draws = int(draws)
if draws < 1:
raise ValueError('Argument `draws` should be above 0.')
if start is None:
start = {}
if random_seed != -1:
nr.seed(random_seed)
try:
step = pm.step_methods.CompoundStep(step)
except TypeError:
pass
point = pm.Point(start, model=model)
step.chain_index = chain
trace.setup(draws, chain)
for i in range(draws):
if i == tune:
step = pm.sampling.stop_tuning(step)
point, out_list = step.step(point)
trace.record(out_list)
yield trace
def pymc3_random_discrete(dist,
paramdomains,
valuedomain=Domain([0]),
ref_rand=None,
size=100000,
alpha=0.05,
fails=20):
model = build_model(dist, valuedomain, paramdomains)
domains = paramdomains.copy()
for pt in product(domains, n_samples=100):
pt = pm.Point(pt, model=model)
p = alpha
# Allow Chisq test to fail (i.e., the samples be different)
# a certain number of times.
f = fails
while p <= alpha and f > 0:
o = model.named_vars['value'].random(size=size, point=pt)
e = ref_rand(size=size, **pt)
o = np.atleast_1d(o).flatten()
e = np.atleast_1d(e).flatten()
observed = dict(zip(*np.unique(o, return_counts=True)))
expected = dict(zip(*np.unique(e, return_counts=True)))
for e in expected.keys():
expected[e] = (observed.get(e, 0), expected[e])
k = np.array([v for v in expected.values()])
if np.all(k[:, 0] == k[:, 1]):
p = 1.
else:
_, p = st.chisquare(k[:, 0], k[:, 1])
f -= 1
assert p > alpha, str(pt)
def _initial_population(draws, model, variables, start):
"""
Create an initial population from the prior
"""
population = []
var_info = OrderedDict()
if start is None:
init_rnd = pm.sample_prior_predictive(
draws, var_names=[v.name for v in model.unobserved_RVs], model=model
)
else:
init_rnd = start
init = model.test_point
for v in variables:
var_info[v.name] = (init[v.name].shape, init[v.name].size)
for i in range(draws):
point = pm.Point({v.name: init_rnd[v.name][i] for v in variables}, model=model)
population.append(model.dict_to_array(point))
return np.array(floatX(population)), var_info
def pymc3_random(dist,
paramdomains,
ref_rand,
valuedomain=Domain([0]),
size=10000,
alpha=0.05,
fails=10,
extra_args=None,
model_args=None):
if model_args is None:
model_args = {}
model = build_model(dist, valuedomain, paramdomains, extra_args)
domains = paramdomains.copy()
for pt in product(domains, n_samples=100):
pt = pm.Point(pt, model=model)
pt.update(model_args)
p = alpha
# Allow KS test to fail (i.e., the samples be different)
# a certain number of times. Crude, but necessary.
f = fails
while p <= alpha and f > 0:
s0 = model.named_vars['value'].random(size=size, point=pt)
s1 = ref_rand(size=size, **pt)
_, p = st.ks_2samp(
np.atleast_1d(s0).flatten(),
np.atleast_1d(s1).flatten())
f -= 1
assert p > alpha, str(pt)
def __init__(self,
vars=None,
covariance=None,
scaling=1.,
n_chains=100,
tune=True,
tune_interval=100,
model=None,
check_bound=True,
likelihood_name='like',
proposal_dist=MvNPd,
coef_variation=1.,
**kwargs):
model = pm.modelcontext(model)
if vars is None:
vars = model.vars
vars = pm.inputvars(vars)
if covariance is None:
self.covariance = np.eye(sum(v.dsize for v in vars))
self.scaling = np.atleast_1d(scaling)
self.tune = tune
self.check_bnd = check_bound
self.tune_interval = tune_interval
self.steps_until_tune = tune_interval
self.proposal_dist = proposal_dist(self.covariance)
self.proposal_samples_array = self.proposal_dist(n_chains)
self.stage_sample = 0
self.accepted = 0
self.beta = 0
self.stage = 0
self.coef_variation = coef_variation
self.n_chains = n_chains
self.likelihoods = []
self.likelihood_name = likelihood_name
self.discrete = np.concatenate(
[[v.dtype in pm.discrete_types] * (v.dsize or 1) for v in vars])
self.any_discrete = self.discrete.any()
self.all_discrete = self.discrete.all()
# create initial population
self.population = []
self.array_population = np.zeros(n_chains)
for i in range(self.n_chains):
dummy = pm.Point({v.name: v.random() for v in vars}, model=model)
self.population.append(dummy)
shared = make_shared_replacements(vars, model)
self.logp_forw = logp_forw(model.logpt, vars, shared)
self.check_bnd = logp_forw(model.varlogpt, vars, shared)
self.delta_logp = pm.metropolis.delta_logp(model.logpt, vars, shared)
super(ATMCMC, self).__init__(vars, shared)
def _initial_population(draws, model, variables):
"""
Create an initial population from the prior
"""
population = []
var_info = {}
start = model.test_point
init_rnd = pm.sample_prior_predictive(draws, model=model)
for v in variables:
var_info[v.name] = (start[v.name].shape, start[v.name].size)
for i in range(draws):
point = pm.Point({v.name: init_rnd[v.name][i] for v in variables}, model=model)
population.append(model.dict_to_array(point))
return np.array(floatX(population)), var_info
def too_slow(self):
model = self.build_model()
with model:
start = pm.Point({
'groupmean': self.obs_means.mean(),
'groupsd_interval__': 0,
'sd_interval__': 0,
'means': np.array(self.obs_means),
'u_m': np.array([.72]),
'floor_m': 0.,
})
start = pm.find_MAP(start, model.vars[:-1])
H = model.fastd2logp()
h = np.diag(H(start))
step = pm.HamiltonianMC(model.vars, h)
pm.sample(50, step=step, start=start)
def too_slow(self):
model = self.build_model()
with model:
start = pm.Point({
"groupmean": self.obs_means.mean(),
"groupsd_interval__": 0,
"sd_interval__": 0,
"means": np.array(self.obs_means),
"u_m": np.array([0.72]),
"floor_m": 0.0,
})
start = pm.find_MAP(start, model.vars[:-1])
H = model.fastd2logp()
h = np.diag(H(start))
step = pm.HamiltonianMC(model.vars, h)
pm.sample(50, step=step, start=start)
def _initial_population(samples, chains, model, variables):
"""
Create an initial population from the prior
"""
population = []
init_rnd = {}
start = model.test_point
for v in variables:
if pm.util.is_transformed_name(v.name):
trans = v.distribution.transform_used.forward_val
init_rnd[v.name] = trans(v.distribution.dist.random(size=chains, point=start))
else:
init_rnd[v.name] = v.random(size=chains, point=start)
for i in range(chains):
population.append(pm.Point({v.name: init_rnd[v.name][i] for v in variables}, model=model))
return population
def __init__(self,
vars=None,
out_vars=None,
n_chains=100,
scaling=1.,
covariance=None,
likelihood_name='like',
proposal_name='MultivariateNormal',
tune=True,
tune_interval=100,
coef_variation=1.,
check_bound=True,
model=None,
random_seed=-1):
warnings.warn(EXPERIMENTAL_WARNING)
if random_seed != -1:
nr.seed(random_seed)
model = modelcontext(model)
if vars is None:
vars = model.vars
vars = inputvars(vars)
if out_vars is None:
out_vars = model.unobserved_RVs
out_varnames = [out_var.name for out_var in out_vars]
if covariance is None and proposal_name == 'MultivariateNormal':
self.covariance = np.eye(sum(v.dsize for v in vars))
scale = self.covariance
elif covariance is None:
scale = np.ones(sum(v.dsize for v in vars))
else:
scale = covariance
self.scaling = np.atleast_1d(scaling)
self.tune = tune
self.check_bnd = check_bound
self.tune_interval = tune_interval
self.steps_until_tune = tune_interval
self.proposal_name = proposal_name
self.proposal_dist = choose_proposal(self.proposal_name, scale=scale)
self.proposal_samples_array = self.proposal_dist(n_chains)
self.stage_sample = 0
self.accepted = 0
self.beta = 0
self.stage = 0
self.chain_index = 0
self.resampling_indexes = np.arange(n_chains)
self.coef_variation = coef_variation
self.n_chains = n_chains
self.likelihoods = np.zeros(n_chains)
self.likelihood_name = likelihood_name
self._llk_index = out_varnames.index(likelihood_name)
self.discrete = np.concatenate(
[[v.dtype in discrete_types] * (v.dsize or 1) for v in vars])
self.any_discrete = self.discrete.any()
self.all_discrete = self.discrete.all()
# create initial population
self.population = []
self.array_population = np.zeros(n_chains)
for _ in range(self.n_chains):
self.population.append(
pm.Point({v.name: v.random()
for v in vars}, model=model))
self.chain_previous_lpoint = copy.deepcopy(self.population)
shared = make_shared_replacements(vars, model)
self.logp_forw = logp_forw(out_vars, vars, shared)
self.check_bnd = logp_forw([model.varlogpt], vars, shared)
super(SMC, self).__init__(vars, out_vars, shared)
def __init__(self,
vars=None,
out_vars=None,
samples=1000,
n_chains=100,
n_steps=25,
scaling=1.,
covariance=None,
likelihood_name='l_like__',
proposal_name='MultivariateNormal',
tune_interval=10,
threshold=0.5,
check_bound=True,
model=None,
random_seed=-1):
warnings.warn(EXPERIMENTAL_WARNING)
if random_seed != -1:
nr.seed(random_seed)
model = modelcontext(model)
if vars is None:
vars = model.vars
vars = inputvars(vars)
if out_vars is None:
if not any(likelihood_name == RV.name
for RV in model.unobserved_RVs):
pm._log.info('Adding model likelihood to RVs!')
with model:
llk = pm.Deterministic(likelihood_name, model.logpt)
else:
pm._log.info('Using present model likelihood!')
out_vars = model.unobserved_RVs
out_varnames = [out_var.name for out_var in out_vars]
if covariance is None and proposal_name == 'MultivariateNormal':
self.covariance = np.eye(sum(v.dsize for v in vars))
scale = self.covariance
elif covariance is None:
scale = np.ones(sum(v.dsize for v in vars))
else:
scale = covariance
self.proposal_name = proposal_name
self.proposal_dist = choose_proposal(self.proposal_name, scale=scale)
self.scaling = np.atleast_1d(scaling)
self.check_bnd = check_bound
self.tune_interval = tune_interval
self.steps_until_tune = tune_interval
self.proposal_samples_array = self.proposal_dist(n_chains)
self.samples = samples
self.n_steps = n_steps
self.stage_sample = 0
self.accepted = 0
self.beta = 0
self.sjs = 1
self.stage = 0
self.chain_index = 0
self.resampling_indexes = np.arange(n_chains)
self.threshold = threshold
self.n_chains = n_chains
self.likelihoods = np.zeros(n_chains)
self.likelihood_name = likelihood_name
self._llk_index = out_varnames.index(likelihood_name)
self.discrete = np.concatenate(
[[v.dtype in discrete_types] * (v.dsize or 1) for v in vars])
self.any_discrete = self.discrete.any()
self.all_discrete = self.discrete.all()
# create initial population
self.population = []
self.array_population = np.zeros(n_chains)
start = model.test_point
init_rnd = {}
for v in vars:
if pm.util.is_transformed_name(v.name):
trans = v.distribution.transform_used.forward
rnd = trans(
v.distribution.dist.random(size=self.n_chains,
point=start))
init_rnd[v.name] = rnd.eval()
else:
init_rnd[v.name] = v.random(size=self.n_chains, point=start)
for i in range(self.n_chains):
self.population.append(
pm.Point({v.name: init_rnd[v.name][i]
for v in vars},
model=model))
self.chain_previous_lpoint = copy.deepcopy(self.population)
shared = make_shared_replacements(vars, model)
self.logp_forw = logp_forw(out_vars, vars, shared)
self.check_bnd = logp_forw([model.varlogpt], vars, shared)
super(SMC, self).__init__(vars, out_vars, shared)
