Dynesty warnings added (#1324)

* added warnings to dynesty and support for posterior

* adapted objective in test to likelihood

* added string constants

* added objective type to dynesty in notebook

doc/example/sampler_study.ipynb

@@ -590,7 +590,7 @@
    "metadata": {},
    "outputs": [],
    "source": [
-    "sampler = sample.DynestySampler()\n",
+    "sampler = sample.DynestySampler(objective_type=\"negloglike\")\n",
     "result = sample.sample(\n",
     "    problem=problem,\n",
     "    n_samples=None,\n",

pypesto/C.py

@@ -70,6 +70,8 @@ class EnsembleType(Enum):
 RES = "res"  # residual
 SRES = "sres"  # residual sensitivities
 RDATAS = "rdatas"  # returned simulated data sets
+OBJECTIVE_NEGLOGPOST = "neglogpost"  # objective is negative log-posterior
+OBJECTIVE_NEGLOGLIKE = "negloglike"  # objective is negative log-likelihood
 
 TIME = "time"  # time
 N_FVAL = "n_fval"  # number of function evaluations

pypesto/sample/dynesty.py

@@ -30,6 +30,7 @@
 
 import numpy as np
 
+from ..C import OBJECTIVE_NEGLOGLIKE, OBJECTIVE_NEGLOGPOST
 from ..problem import Problem
 from ..result import McmcPtResult
 from .sampler import Sampler, SamplerImportError
@@ -62,6 +63,7 @@ def __init__(
         sampler_args: dict = None,
         run_args: dict = None,
         dynamic: bool = True,
+        objective_type: str = OBJECTIVE_NEGLOGPOST,
     ):
         """
         Initialize sampler.
@@ -76,6 +78,9 @@ def __init__(
             method of the dynesty sampler.
         dynamic:
             Whether to use dynamic or static nested sampling.
+        objective_type:
+            The objective to optimize (as defined in  pypesto.problem). Either "neglogpost" or
+            "negloglike". If "neglogpost", x_priors have to be defined in the problem.
         """
         # check dependencies
         import dynesty
@@ -94,6 +99,13 @@ def __init__(
             run_args = {}
         self.run_args: dict = run_args
 
+        if objective_type not in [OBJECTIVE_NEGLOGPOST, OBJECTIVE_NEGLOGLIKE]:
+            raise ValueError(
+                f"Objective has to be either '{OBJECTIVE_NEGLOGPOST}' or '{OBJECTIVE_NEGLOGLIKE}' "
+                f"as defined in pypesto.problem."
+            )
+        self.objective_type = objective_type
+
         # set in initialize
         self.problem: Union[Problem, None] = None
         self.sampler: Union[
@@ -112,8 +124,6 @@ def prior_transform(self, prior_sample: np.ndarray) -> np.ndarray:
         ----------
         prior_sample:
             The prior sample, provided by dynesty.
-        problem:
-            The pyPESTO problem.
 
         Returns
         -------
@@ -130,7 +140,13 @@ def loglikelihood(self, x):
         if any(x < self.problem.lb) or any(x > self.problem.ub):
             return -np.inf
         # invert sign
-        # TODO this is possibly the posterior if priors are defined
+        if self.objective_type == OBJECTIVE_NEGLOGPOST:
+            # problem.objective returns negative log-posterior
+            # compute log-likelihood by subtracting log-prior
+            return -1.0 * (
+                self.problem.objective(x) - self.problem.x_priors(x)
+            )
+        # problem.objective returns negative log-likelihood
         return -1.0 * self.problem.objective(x)
 
     def initialize(
@@ -149,6 +165,28 @@ def initialize(
         if self.dynamic:
             sampler_class = dynesty.DynamicNestedSampler
 
+        # check if objective fits to the pyPESTO problem
+        if self.objective_type == OBJECTIVE_NEGLOGPOST:
+            if self.problem.x_priors is None:
+                # objective is the negative log-posterior, but no priors are defined
+                # sampler needs the likelihood
+                raise ValueError(
+                    f"x_priors have to be defined in the problem if objective is '{OBJECTIVE_NEGLOGPOST}'."
+                )
+        else:
+            # if objective is the negative log likelihood, we will ignore x_priors even if they are defined
+            if self.problem.x_priors is not None:
+                logger.warning(
+                    f"Assuming '{OBJECTIVE_NEGLOGLIKE}' as objective. "
+                    f"'x_priors' defined in the problem will be ignored."
+                )
+
+        # if priors are uniform, we can use the default prior transform (assuming that bounds are set correctly)
+        logger.warning(
+            "Assuming 'prior_transform' is correctly specified. If 'x_priors' is not uniform, 'prior_transform'"
+            " has to be adjusted accordingly."
+        )
+
         # initialize sampler
         self.sampler = sampler_class(
             loglikelihood=self.loglikelihood,

test/sample/test_sample.py

@@ -13,6 +13,7 @@
 import pypesto.optimize as optimize
 import pypesto.petab
 import pypesto.sample as sample
+from pypesto.C import OBJECTIVE_NEGLOGLIKE, OBJECTIVE_NEGLOGPOST
 from pypesto.sample.pymc import PymcSampler
 
 
@@ -190,7 +191,7 @@ def sampler(request):
     elif request.param == "Emcee":
         return sample.EmceeSampler(nwalkers=10)
     elif request.param == "Dynesty":
-        return sample.DynestySampler()
+        return sample.DynestySampler(objective_type="negloglike")
 
 
 @pytest.fixture(params=["gaussian", "gaussian_mixture", "rosenbrock"])
@@ -722,13 +723,13 @@ def test_samples_cis():
         assert (diff == 0).all()
         # check if lower bound is smaller than upper bound
         assert (lb < ub).all()
-        # check if dimmensions agree
+        # check if dimensions agree
         assert lb.shape == ub.shape
 
 
 def test_dynesty_mcmc_samples():
     problem = gaussian_problem()
-    sampler = sample.DynestySampler()
+    sampler = sample.DynestySampler(objective_type=OBJECTIVE_NEGLOGLIKE)
 
     result = sample.sample(
         problem=problem,
@@ -746,6 +747,46 @@ def test_dynesty_mcmc_samples():
     assert not (np.diff(mcmc_sample_result.trace_neglogpost) <= 0).all()
 
 
+def test_dynesty_posterior():
+    # define negative log posterior
+    posterior_fun = pypesto.Objective(fun=negative_log_posterior)
+
+    # define negative log prior
+    prior_fun = pypesto.Objective(fun=negative_log_prior)
+
+    # define pypesto prior object
+    prior_object = pypesto.NegLogPriors(objectives=[prior_fun])
+
+    # define pypesto problem using prior object
+    test_problem = pypesto.Problem(
+        objective=posterior_fun,
+        x_priors_defs=prior_object,
+        lb=-10,
+        ub=10,
+        x_names=["x"],
+    )
+
+    # define sampler
+    sampler = sample.DynestySampler(
+        objective_type=OBJECTIVE_NEGLOGPOST
+    )  # default
+
+    result = sample.sample(
+        problem=test_problem,
+        sampler=sampler,
+        n_samples=None,
+        filename=None,
+    )
+
+    original_sample_result = sampler.get_original_samples()
+    mcmc_sample_result = result.sample_result
+
+    # Nested sampling function values are monotonically increasing
+    assert (np.diff(original_sample_result.trace_neglogpost) <= 0).all()
+    # MCMC samples are not
+    assert not (np.diff(mcmc_sample_result.trace_neglogpost) <= 0).all()
+
+
 def test_thermodynamic_integration():
     # test thermodynamic integration
     problem = gaussian_problem()