Release 0.10.12

doc/releasenotes.rst

@@ -8,6 +8,13 @@ Release Notes
 ...........
 
 
+0.10.12 (2021-01-20)
+--------------------
+
+* Check components for their adaptivity for correct application of look-ahead
+  mode (#397).
+
+
 0.10.11 (2021-01-02)
 --------------------
 

pyabc/acceptor/acceptor.py

@@ -22,7 +22,7 @@
 from ..distance import Distance, SCALE_LIN, StochasticKernel
 from ..epsilon import Epsilon
 from ..parameters import Parameter
-from .pdf_norm import pdf_norm_max_found
+from .pdf_norm import pdf_norm_from_kernel, pdf_norm_max_found
 from ..storage import save_dict_to_json
 
 
@@ -171,6 +171,20 @@ def __call__(self,
         """
         raise NotImplementedError()
 
+    def requires_calibration(self) -> bool:
+        """
+        Whether the class requires an initial calibration, based on
+        samples from the prior. Default: False.
+        """
+        return False
+
+    def is_adaptive(self) -> bool:
+        """
+        Whether the class is dynamically updated after each generation,
+        based on the last generation's available data. Default: False.
+        """
+        return False
+
     # pylint: disable=R0201
     def get_epsilon_config(self, t: int) -> dict:
         """
@@ -187,7 +201,7 @@ def get_epsilon_config(self, t: int) -> dict:
         config: dict
             The relevant information.
         """
-        return None
+        return {}
 
 
 class SimpleFunctionAcceptor(Acceptor):
@@ -373,6 +387,16 @@ def __init__(
         self.kernel_scale = None
         self.kernel_pdf_max = None
 
+    def requires_calibration(self) -> bool:
+        # this check is rather superficial and may be improved by a re-design
+        #  of `pdf_norm_method`
+        return self.pdf_norm_method != pdf_norm_from_kernel
+
+    def is_adaptive(self) -> bool:
+        # this check is rather superficial and may be improved by a re-design
+        #  of `pdf_norm_method`
+        return self.pdf_norm_method != pdf_norm_from_kernel
+
     def initialize(
             self,
             t: int,

pyabc/distance/base.py

@@ -126,6 +126,20 @@ def __call__(
             observed data.
         """
 
+    def requires_calibration(self) -> bool:
+        """
+        Whether the class requires an initial calibration, based on
+        samples from the prior. Default: False.
+        """
+        return False
+
+    def is_adaptive(self) -> bool:
+        """
+        Whether the class is dynamically updated after each generation,
+        based on the last generation's available data. Default: False.
+        """
+        return False
+
     def get_config(self) -> dict:
         """
         Return configuration of the distance.
@@ -176,7 +190,7 @@ def __call__(self,
                  t: int = None,
                  par: dict = None) -> float:
         raise AssertionError(
-            f"{self.__class__.__name__} is not intended to be called.")
+            f"Distance {self.__class__.__name__} should not be called.")
 
 
 class AcceptAllDistance(Distance):

pyabc/distance/distance.py

@@ -222,6 +222,12 @@ def configure_sampler(self,
         if self.adaptive:
             sampler.sample_factory.record_rejected = True
 
+    def requires_calibration(self) -> bool:
+        return self.initial_weights is None
+
+    def is_adaptive(self) -> bool:
+        return self.adaptive
+
     def initialize(self,
                    t: int,
                    get_all_sum_stats: Callable[[], List[dict]],
@@ -233,7 +239,7 @@ def initialize(self,
         self.x_0 = x_0
 
         # initial weights pre-defined
-        if self.initial_weights is not None:
+        if not self.requires_calibration():
             self.weights[t] = self.initial_weights
             return
 
@@ -249,7 +255,7 @@ def update(self,
         """
         Update weights.
         """
-        if not self.adaptive:
+        if not self.is_adaptive():
             return False
 
         # execute function
@@ -407,6 +413,12 @@ def __init__(
         self.weights = weights
         self.factors = factors
 
+    def requires_calibration(self) -> bool:
+        return any(d.requires_calibration() for d in self.distances)
+
+    def is_adaptive(self) -> bool:
+        return any(d.is_adaptive() for d in self.distances)
+
     def initialize(
             self,
             t: int,
@@ -552,6 +564,14 @@ def __init__(
         self.scale_function = scale_function
         self.log_file = log_file
 
+    def requires_calibration(self) -> bool:
+        return (self.initial_weights is None
+                or any(d.requires_calibration() for d in self.distances))
+
+    def is_adaptive(self) -> bool:
+        return (self.adaptive
+                or any(d.is_adaptive() for d in self.distances))
+
     def initialize(self,
                    t: int,
                    get_all_sum_stats: Callable[[], List[dict]],
@@ -716,6 +736,9 @@ def _calculate_whitening_transformation_matrix(self, sum_stats):
         self._whitening_transformation_matrix = (
             v.dot(np.diag(1. / np.sqrt(w))).dot(v.T))
 
+    def requires_calibration(self) -> bool:
+        return True
+
     def initialize(self,
                    t: int,
                    get_all_sum_stats: Callable[[], List[dict]],
@@ -812,6 +835,9 @@ def _calculate_normalization(self, sum_stats):
                               - self.lower(measures[measure])
                               for measure in self.measures_to_use}
 
+    def requires_calibration(self) -> bool:
+        return True
+
     def initialize(self,
                    t: int,
                    get_all_sum_stats: Callable[[], List[dict]],

pyabc/epsilon/base.py

@@ -122,6 +122,20 @@ def __call__(self,
             The epsilon for population t.
         """
 
+    def requires_calibration(self) -> bool:
+        """
+        Whether the class requires an initial calibration, based on
+        samples from the prior. Default: False.
+        """
+        return False
+
+    def is_adaptive(self) -> bool:
+        """
+        Whether the class is dynamically updated after each generation,
+        based on the last generation's available data. Default: False.
+        """
+        return False
+
     def get_config(self):
         """
         Return configuration of the distance function.

pyabc/epsilon/epsilon.py

@@ -135,13 +135,19 @@ def get_config(self):
 
         return config
 
+    def requires_calibration(self) -> bool:
+        return self._initial_epsilon == 'from_sample'
+
+    def is_adaptive(self) -> bool:
+        return True
+
     def initialize(self,
                    t: int,
                    get_weighted_distances: Callable[[], pd.DataFrame],
                    get_all_records: Callable[[], List[dict]],
                    max_nr_populations: int,
                    acceptor_config: dict):
-        if self._initial_epsilon != 'from_sample':
+        if not self.requires_calibration():
             # safety check in __call__
             return