Refactor similarity score calculations in stratification_function to use quantiles instead of min/max thresholds, improving accuracy in target similarity score generation.

global_func/stratification_function.py

@@ -12,8 +12,8 @@ def stratification_function(portfolio: pd.DataFrame, lineup_target: int, exclude
         concat_portfolio = concat_portfolio.sort_values(by=sorting_choice, ascending=False).reset_index(drop=True)
 
     # Calculate target similarity scores for linear progression
-    similarity_floor = concat_portfolio[sorting_choice].min() + (concat_portfolio[sorting_choice].min() * low_threshold)
-    similarity_ceiling = concat_portfolio[sorting_choice].max() - (concat_portfolio[sorting_choice].max() * high_threshold)
+    similarity_floor = concat_portfolio[sorting_choice].quantile(low_threshold / 100)
+    similarity_ceiling = concat_portfolio[sorting_choice].quantile(high_threshold / 100)
     
     # Create evenly spaced target similarity scores
     target_similarities = np.linspace(similarity_floor, similarity_ceiling, lineup_target)