Fix small issues in aggregation script (#553)

- No debug mode by default
- Ignore number of evaluation workers in equivalence check
- Change logic for size check to not throw an exception

analytics/tools/aggregate_runs/core_aggregation.py

@@ -9,7 +9,7 @@
 from modyn.supervisor.internal.grpc.enums import PipelineStage
 from modyn.supervisor.internal.pipeline_executor.models import PipelineLogs, SingleEvaluationInfo
 
-DEBUGGING_MODE = True
+DEBUGGING_MODE = False
 """if True, the the process will halt on breakpoints to allow for manual verification"""
 
 
@@ -75,7 +75,7 @@ def aggregate_eval_metrics(df_eval_single: pd.DataFrame, logs: list[PipelineLogs
     )
 
     sizes = groups.agg(size=("model_idx", "size")).reset_index()
-    if len(sizes["size"].unique()) != 1 or int(sizes[0]) != len(logs):
+    if len(sizes["size"].unique()) != 1 or int(sizes["size"].unique()[0]) != len(logs):
         logging.warning(f"\n{sizes[sizes['size'] != len(logs)]}")
         logging.warning(
             "The number of records in every group is not equal to the number of logs. "

analytics/tools/aggregate_runs/pipeline_equivalence.py

@@ -16,6 +16,8 @@ def assert_pipeline_equivalence(logs: list[PipelineLogs]) -> None:
         candidate.config.pipeline.training.seed = candidates[0].config.pipeline.training.seed
         candidate.config.pipeline.training.device = candidates[0].config.pipeline.training.device
         candidate.config.pipeline.evaluation.device = candidates[0].config.pipeline.evaluation.device
+        candidate.config.pipeline.evaluation.after_pipeline_evaluation_workers = candidates[0].config.pipeline.evaluation.after_pipeline_evaluation_workers
+        candidate.config.pipeline.evaluation.after_training_evaluation_workers = candidates[0].config.pipeline.evaluation.after_training_evaluation_workers
 
         if isinstance(candidate.config.pipeline.selection_strategy, CoresetStrategyConfig) and isinstance(
             candidate.config.pipeline.selection_strategy.downsampling_config, RHOLossDownsamplingConfig