Merge pull request #318 from SubstraFoundation/aggregation_fix

Homogenize aggregation and aggregation_weighting

Author: arthurPignet

mplc/contributivity.py

@@ -977,7 +977,7 @@ def PVRL(self, learning_rate):
             logger.info(f"Partners selected for the next epoch: {[p.id for p in mpl.partners_list]}")
 
             # apply one epoch with the selected partner to the previous model/ do the action
-            mpl.aggregator = self.scenario._aggregation(mpl)  # we have to reset the weight of _aggregation
+            mpl.aggregator = self.scenario._aggregation_weighting(mpl)  # we have to reset the weight of _aggregation
             mpl.fit_epoch()
             loss = mpl.history.history['mpl_model']['val_loss'][mpl.epoch_index, -1]
             mpl.epoch_index += 1

mplc/multi_partner_learning.py

@@ -54,8 +54,8 @@ def __init__(self, scenario, **kwargs):
         self.minibatch_count = scenario.minibatch_count
         self.is_early_stopping = scenario.is_early_stopping
 
-        # Attributes related to the _aggregation approach
-        self.aggregation_method = scenario._aggregation
+        # Attributes related to the _aggregation_weighting approach
+        self.aggregation_method = scenario._aggregation_weighting
 
         # Attributes to store results
         self.save_folder = scenario.save_folder

mplc/scenario.py

@@ -94,7 +94,7 @@ def __init__(
         params_known += [
             "contributivity_methods",
             "multi_partner_learning_approach",
-            "aggregation",
+            "aggregation_weighting",
         ]  # federated learning related
         params_known += [
             "partners_count",
@@ -232,9 +232,9 @@ def __init__(
 
         # Define how federated learning aggregation steps are weighted...
         # ... Toggle between 'uniform' (default) and 'data_volume'
-        self.aggregation = aggregation_weighting
+        self.aggregation_weighting = aggregation_weighting
         try:
-            self._aggregation = AGGREGATORS[aggregation_weighting]
+            self._aggregation_weighting = AGGREGATORS[aggregation_weighting]
         except KeyError:
             raise ValueError(f"aggregation approach '{aggregation_weighting}' is not a valid approach. ")
 
@@ -380,7 +380,7 @@ def copy(self, **kwargs):
         for key in ['partners_list',
                     'mpl',
                     '_multi_partner_learning_approach',
-                    '_aggregation',
+                    '_aggregation_weighting',
                     'use_saved_weights',
                     'contributivity_list',
                     'scenario_name',
@@ -408,7 +408,7 @@ def log_scenario_description(self):
         logger.info(f"   Number of partners defined: {self.partners_count}")
         logger.info(f"   Data distribution scenario chosen: {self.splitter}")
         logger.info(f"   Multi-partner learning approach: {self.multi_partner_learning_approach}")
-        logger.info(f"   Weighting option: {self.aggregation}")
+        logger.info(f"   Weighting option: {self.aggregation_weighting}")
         logger.info(f"   Iterations parameters: "
                     f"{self.epoch_count} epochs > "
                     f"{self.minibatch_count} mini-batches > "
@@ -515,7 +515,7 @@ def to_dataframe(self):
 
         # Multi-partner learning approach parameters
         dict_results["multi_partner_learning_approach"] = self.multi_partner_learning_approach
-        dict_results["aggregation"] = self.aggregation
+        dict_results["aggregation_weighting"] = self.aggregation_weighting
         dict_results["epoch_count"] = self.epoch_count
         dict_results["minibatch_count"] = self.minibatch_count
         dict_results["gradient_updates_per_pass_count"] = self.gradient_updates_per_pass_count

tests/unit_tests.py

@@ -80,7 +80,7 @@ def create_MultiPartnerLearning(create_all_datasets):
         epoch_count=2,
         minibatch_count=2,
         dataset=data,
-        aggregation=UniformAggregator,
+        aggregation_weighting=UniformAggregator,
         is_early_stopping=True,
         is_save_data=False,
     )
@@ -137,7 +137,7 @@ def create_Scenario(request):
         {
             "contributivity_methods": ["Shapley values", "Independent scores"],
             "multi_partner_learning_approach": "fedavg",
-            "aggregation": "uniform",
+            "aggregation_weighting": "uniform",
         }
     )
     params.update(