Commit test

best_hyperparameters.json

-{"num_hidden_layers": 3, "hidden_layer_size": 512, "learning_rate": 0.005, "batch_size": 64, "threshold": 0.25}
\ No newline at end of file
+{"num_hidden_layers": 1, "hidden_layer_size": 48, "learning_rate": 0.01, "batch_size": 64, "threshold": 0.25}
\ No newline at end of file

final_normalization_constants.json

-{"age": {"mean": 37.26280350948786, "std": 21.842276857225034}, "latest_creatinine_value": {"mean": 166.31846085832825, "std": 94.44752591026013}, "median_previous": {"mean": 134.41235802217233, "std": 46.13848554494386}, "mean_previous": {"mean": 134.4250456736426, "std": 46.02537337310225}, "std_dev_previous": {"mean": 10.737323990705239, "std": 7.306492442412928}, "abs_percentage_diff": {"mean": 0.2897293206129817, "std": 0.40866939525398815}}
\ No newline at end of file

helper_functions.py

@@ -457,7 +457,7 @@ def objective(trial, data, device):
     # Suggest hyperparameters
     num_hidden_layers = trial.suggest_int("num_hidden_layers", 1, 5)
     hidden_layer_size = trial.suggest_categorical(
-        "hidden_layer_size", [32, 64, 128, 256, 512, 1024]
+        "hidden_layer_size", [32, 48,64, 96,128,184, 256, 512, 1024]
     )
     learning_rate = trial.suggest_categorical(
         "learning_rate", [0.001, 0.005, 0.01, 0.02, 0.05]
@@ -484,7 +484,7 @@ def objective(trial, data, device):
     return avg_f3
 
 
-def tune_hyperparameters(data: pd.DataFrame, n_trials=10):
+def tune_hyperparameters(data: pd.DataFrame, n_trials=30):
     """
     Run Optuna hyperparameter tuning on `data`, maximizing F3 via cross-validation.
 

main.py

@@ -27,7 +27,7 @@ def hyper_parameter_tuning():
     processor = DataProcessor()
     data = processor.preprocess(['training.csv'], save_constants=True)
     balanced_data = processor.handle_class_imbalance(data)
-    best_trial = tune_hyperparameters(balanced_data, n_trials=10)
+    best_trial = tune_hyperparameters(balanced_data, n_trials=100)
     return best_trial
 
 
@@ -135,6 +135,7 @@ def main():
             aki_label = 'y' if pred == 1 else 'n'
             writer.writerow([aki_label])
 
+# [I 2025-01-21 11:12:32,171] Trial 9 finished with value: 0.9955269684057108 and parameters: {'num_hidden_layers': 1, 'hidden_layer_size': 512, 'learning_rate': 0.05, 'batch_size': 1024, 'threshold': 0.625}. Best is trial 3 with value: 0.9984615847421683.
 
 if __name__ == '__main__':
     """
@@ -145,7 +146,7 @@ if __name__ == '__main__':
       3) Train final model on combined CSVs
       4) Or run main() for final predictions
     """
-    #hyper_parameter_tuning()
-    #train_model_and_test()
+    hyper_parameter_tuning()
+    train_model_and_test()
     train_final_model()
-    #main()
+    main()