Emcee batch fix

flepimop/gempyor_pkg/src/gempyor/calibrate.py

@@ -171,6 +171,8 @@ def calibrate(
 
     # Draw/get initial parameters:
     backend = emcee.backends.HDFBackend(filename)
+    # save the dataframe of parameter names (this should be added to the h5 file)
+    gempyor_inference.inferpar.get_parameter_df().to_csv(filename.replace(".h5", ".csv"), index=True)
     if resume or resume_location is not None:
         # Normally one would put p0 = None to get the last State from the sampler, but that poses problems when the likelihood change
         # and then acceptances are not guaranted, see issue #316. This solves this issue and greates a new chain with llik evaluation

flepimop/gempyor_pkg/src/gempyor/inference.py

@@ -382,14 +382,15 @@ def __init__(
         self.already_built = False  # whether we have already built the costly objects that need just one build
         self.autowrite_seir = autowrite_seir
 
-        self.static_sim_arguments = get_static_arguments(self.modinf)
-
         ## Inference Stuff
         self.do_inference = False
         if config["inference"].exists():
             from . import inference_parameter, logloss
 
             if config["inference"]["method"].get("default") == "emcee":
+                # Generates the static_sim_arguments only if this is a config argument.
+                self.static_sim_arguments = get_static_arguments(self.modinf)
+
                 self.do_inference = True
                 self.inference_method = "emcee"
                 self.inferpar = inference_parameter.InferenceParameters(

flepimop/gempyor_pkg/src/gempyor/inference_parameter.py

@@ -173,6 +173,15 @@ def __len__(self):
         so one can use the built-in python len function
         """
         return len(self.pnames)
+
+    def get_parameter_df(self):
+        data = {
+            "ptypes": self.ptypes,
+            "subpops": self.subpops,
+            "pnames": self.pnames
+        }
+        df = pd.DataFrame(data)
+        return df
 
     def draw_initial(self, n_draw=1):
         """
@@ -235,6 +244,8 @@ def inject_proposal(
         hnpi_df_mod = hnpi_df.copy(deep=True)
 
         # Ideally this should lie in each submodules, e.g NPI.inject, parameter.inject
+        if self.get_dim() != len(proposal):
+            raise ValueError(f"Inference object stores {self.get_dim()} parameters, but received a proposal of lenght {len(proposal)} to be injected.")
 
         for p_idx in range(self.get_dim()):
             if self.ptypes[p_idx] == "seir_modifiers":

flepimop/gempyor_pkg/src/gempyor/outcomes.py

@@ -198,6 +198,8 @@ def read_parameters_from_config(modinf: model_info.ModelInfo):
                         parameters[new_comp]["delay::npi_param_name"] = f"{new_comp}::delay".lower()
                 else:
                     logging.critical(f"No delay for outcome {new_comp}, using a 0 delay")
+                    # FIXME: We should not here modify outcomes_config := modinf.outcomes_config["outcomes"]
+                    # because this changes the order in which the outcomes the next time this function is called.
                     outcomes_config[new_comp]["delay"] = {"value": 0}
                     parameters[new_comp]["delay"] = outcomes_config[new_comp]["delay"]["value"]
                     parameters[new_comp]["delay::npi_param_name"] = f"{new_comp}::delay".lower()