def test_errors_and_warnings(self):
with pm.Model():
A = pm.Normal("A")
B = pm.Uniform("B")
strace = pm.sampling.NDArray(vars=[A, B])
strace.setup(10, 0)
with pytest.raises(ValueError, match="from existing MultiTrace"):
pm.sampling._choose_backend(trace=MultiTrace([strace]))
strace.record({"A": 2, "B_interval__": 0.1})
assert len(strace) == 1
with pytest.raises(ValueError, match="Continuation of traces"):
pm.sampling._choose_backend(trace=strace)
def point_list_to_multitrace(point_list: List[Dict[str, np.ndarray]],
model: Optional[Model] = None) -> MultiTrace:
"""transform point list into MultiTrace"""
_model = modelcontext(model)
varnames = list(point_list[0].keys())
with _model:
chain = NDArray(model=_model, vars=[_model[vn] for vn in varnames])
chain.setup(draws=len(point_list), chain=0)
# since we are simply loading a trace by hand, we need only a vacuous function for
# chain.record() to use. This crushes the default.
def point_fun(point):
return [point[vn] for vn in varnames]
chain.fn = point_fun
for point in point_list:
chain.record(point)
return MultiTrace([chain])
def _sample(draws,
step,
start=None,
trace=None,
chain=0,
tune=None,
progressbar=True,
model=None,
random_seed=None):
sampling = _iter_sample(draws, step, start, trace, chain, tune, model,
random_seed)
progress = progress_bar(draws)
try:
for i, trace in enumerate(sampling):
if progressbar:
progress.update(i)
except KeyboardInterrupt:
trace.close()
return MultiTrace([trace])
def sample_smc(
draws=2000,
kernel=IMH,
*,
start=None,
model=None,
random_seed=None,
chains=None,
cores=None,
compute_convergence_checks=True,
return_inferencedata=True,
idata_kwargs=None,
progressbar=True,
**kernel_kwargs,
):
r"""
Sequential Monte Carlo based sampling.
Parameters
----------
draws: int
The number of samples to draw from the posterior (i.e. last stage). And also the number of
independent chains. Defaults to 2000.
kernel: SMC Kernel used. Defaults to pm.smc.IMH (Independent Metropolis Hastings)
start: dict, or array of dict
Starting point in parameter space. It should be a list of dict with length `chains`.
When None (default) the starting point is sampled from the prior distribution.
model: Model (optional if in ``with`` context)).
random_seed: int
random seed
chains : int
The number of chains to sample. Running independent chains is important for some
convergence statistics. If ``None`` (default), then set to either ``cores`` or 2, whichever
is larger.
cores : int
The number of chains to run in parallel. If ``None``, set to the number of CPUs in the
system.
compute_convergence_checks : bool
Whether to compute sampler statistics like Gelman-Rubin and ``effective_n``.
Defaults to ``True``.
return_inferencedata : bool, default=True
Whether to return the trace as an :class:`arviz:arviz.InferenceData` (True) object or a `MultiTrace` (False)
Defaults to ``True``.
idata_kwargs : dict, optional
Keyword arguments for :func:`pymc.to_inference_data`
progressbar : bool, optional default=True
Whether or not to display a progress bar in the command line.
**kernel_kwargs: keyword arguments passed to the SMC kernel.
The default IMH kernel takes the following keywords:
threshold: float
Determines the change of beta from stage to stage, i.e. indirectly the number of stages,
the higher the value of `threshold` the higher the number of stages. Defaults to 0.5.
It should be between 0 and 1.
n_steps: int
The number of steps of each Markov Chain. If ``tune_steps == True`` ``n_steps`` will be used
for the first stage and for the others it will be determined automatically based on the
acceptance rate and `p_acc_rate`, the max number of steps is ``n_steps``.
tune_steps: bool
Whether to compute the number of steps automatically or not. Defaults to True
p_acc_rate: float
Used to compute ``n_steps`` when ``tune_steps == True``. The higher the value of
``p_acc_rate`` the higher the number of steps computed automatically. Defaults to 0.85.
It should be between 0 and 1.
Keyword arguments for other kernels should be checked in the respective docstrings
Notes
-----
SMC works by moving through successive stages. At each stage the inverse temperature
:math:`\beta` is increased a little bit (starting from 0 up to 1). When :math:`\beta` = 0
we have the prior distribution and when :math:`\beta` =1 we have the posterior distribution.
So in more general terms we are always computing samples from a tempered posterior that we can
write as:
.. math::
p(\theta \mid y)_{\beta} = p(y \mid \theta)^{\beta} p(\theta)
A summary of the algorithm is:
1. Initialize :math:`\beta` at zero and stage at zero.
2. Generate N samples :math:`S_{\beta}` from the prior (because when :math `\beta = 0` the
tempered posterior is the prior).
3. Increase :math:`\beta` in order to make the effective sample size equals some predefined
value (we use :math:`Nt`, where :math:`t` is 0.5 by default).
4. Compute a set of N importance weights W. The weights are computed as the ratio of the
likelihoods of a sample at stage i+1 and stage i.
5. Obtain :math:`S_{w}` by re-sampling according to W.
6. Use W to compute the mean and covariance for the proposal distribution, a MVNormal.
7. For stages other than 0 use the acceptance rate from the previous stage to estimate
`n_steps`.
8. Run N independent Metropolis-Hastings (IMH) chains (each one of length `n_steps`),
starting each one from a different sample in :math:`S_{w}`. Samples are IMH as the proposal
mean is the of the previous posterior stage and not the current point in parameter space.
9. Repeat from step 3 until :math:`\beta \ge 1`.
10. The final result is a collection of N samples from the posterior.
References
----------
.. [Minson2013] Minson, S. E. and Simons, M. and Beck, J. L., (2013),
Bayesian inversion for finite fault earthquake source models I- Theory and algorithm.
Geophysical Journal International, 2013, 194(3), pp.1701-1726,
`link <https://gji.oxfordjournals.org/content/194/3/1701.full>`__
.. [Ching2007] Ching, J. and Chen, Y. (2007).
Transitional Markov Chain Monte Carlo Method for Bayesian Model Updating, Model Class
Selection, and Model Averaging. J. Eng. Mech., 10.1061/(ASCE)0733-9399(2007)133:7(816),
816-832. `link <http://ascelibrary.org/doi/abs/10.1061/%28ASCE%290733-9399
%282007%29133:7%28816%29>`__
"""
if isinstance(kernel, str) and kernel.lower() in ("abc", "metropolis"):
warnings.warn(
f'The kernel string argument "{kernel}" in sample_smc has been deprecated. '
f"It is no longer needed to distinguish between `abc` and `metropolis`",
FutureWarning,
stacklevel=2,
)
kernel = IMH
if kernel_kwargs.pop("save_sim_data", None) is not None:
warnings.warn(
"save_sim_data has been deprecated. Use pm.sample_posterior_predictive "
"to obtain the same type of samples.",
FutureWarning,
stacklevel=2,
)
if kernel_kwargs.pop("save_log_pseudolikelihood", None) is not None:
warnings.warn(
"save_log_pseudolikelihood has been deprecated. This information is "
"now saved as log_likelihood in models with Simulator distributions.",
FutureWarning,
stacklevel=2,
)
parallel = kernel_kwargs.pop("parallel", None)
if parallel is not None:
warnings.warn(
"The argument parallel is deprecated, use the argument cores instead.",
FutureWarning,
stacklevel=2,
)
if parallel is False:
cores = 1
if cores is None:
cores = _cpu_count()
if chains is None:
chains = max(2, cores)
else:
cores = min(chains, cores)
if random_seed == -1:
raise FutureWarning(
f"random_seed should be a non-negative integer or None, got: {random_seed}"
"This will raise a ValueError in the Future")
random_seed = None
if isinstance(random_seed, int) or random_seed is None:
rng = np.random.default_rng(seed=random_seed)
random_seed = list(rng.integers(2**30, size=chains))
elif isinstance(random_seed, Iterable):
if len(random_seed) != chains:
raise ValueError(
f"Length of seeds ({len(seeds)}) must match number of chains {chains}"
)
else:
raise TypeError(
"Invalid value for `random_seed`. Must be tuple, list, int or None"
)
model = modelcontext(model)
_log = logging.getLogger("pymc")
_log.info("Initializing SMC sampler...")
_log.info(f"Sampling {chains} chain{'s' if chains > 1 else ''} "
f"in {cores} job{'s' if cores > 1 else ''}")
params = (
draws,
kernel,
start,
model,
)
t1 = time.time()
if cores > 1:
pbar = progress_bar((), total=100, display=progressbar)
pbar.update(0)
pbars = [pbar] + [None] * (chains - 1)
pool = mp.Pool(cores)
# "manually" (de)serialize params before/after multiprocessing
params = tuple(cloudpickle.dumps(p) for p in params)
kernel_kwargs = {
key: cloudpickle.dumps(value)
for key, value in kernel_kwargs.items()
}
results = _starmap_with_kwargs(
pool,
_sample_smc_int,
[(*params, random_seed[chain], chain, pbars[chain])
for chain in range(chains)],
repeat(kernel_kwargs),
)
results = tuple(cloudpickle.loads(r) for r in results)
pool.close()
pool.join()
else:
results = []
pbar = progress_bar((), total=100 * chains, display=progressbar)
pbar.update(0)
for chain in range(chains):
pbar.offset = 100 * chain
pbar.base_comment = f"Chain: {chain+1}/{chains}"
results.append(
_sample_smc_int(*params, random_seed[chain], chain, pbar,
**kernel_kwargs))
(
traces,
sample_stats,
sample_settings,
) = zip(*results)
trace = MultiTrace(traces)
idata = None
# Save sample_stats
_t_sampling = time.time() - t1
sample_settings_dict = sample_settings[0]
sample_settings_dict["_t_sampling"] = _t_sampling
sample_stats_dict = sample_stats[0]
if chains > 1:
# Collect the stat values from each chain in a single list
for stat in sample_stats[0].keys():
value_list = []
for chain_sample_stats in sample_stats:
value_list.append(chain_sample_stats[stat])
sample_stats_dict[stat] = value_list
if not return_inferencedata:
for stat, value in sample_stats_dict.items():
setattr(trace.report, stat, value)
for stat, value in sample_settings_dict.items():
setattr(trace.report, stat, value)
else:
for stat, value in sample_stats_dict.items():
if chains > 1:
# Different chains might have more iteration steps, leading to a
# non-square `sample_stats` dataset, we cast as `object` to avoid
# numpy ragged array deprecation warning
sample_stats_dict[stat] = np.array(value, dtype=object)
else:
sample_stats_dict[stat] = np.array(value)
sample_stats = dict_to_dataset(
sample_stats_dict,
attrs=sample_settings_dict,
library=pymc,
)
ikwargs = dict(model=model)
if idata_kwargs is not None:
ikwargs.update(idata_kwargs)
idata = to_inference_data(trace, **ikwargs)
idata = InferenceData(**idata, sample_stats=sample_stats)
if compute_convergence_checks:
if draws < 100:
warnings.warn(
"The number of samples is too small to check convergence reliably.",
stacklevel=2,
)
else:
if idata is None:
idata = to_inference_data(trace, log_likelihood=False)
trace.report._run_convergence_checks(idata, model)
trace.report._log_summary()
return idata if return_inferencedata else trace