Merge pull request #17 from gdmnl/zulun

add the degree output

examples/run_grid_search.py

@@ -4,6 +4,8 @@
 import logging
 from itertools import product
 import os
+import json
+import argparse
 import pandas as pd
 from trainer import DatasetLoader, ModelLoader
 from pathlib import Path
@@ -72,7 +74,9 @@ def update_results_csv(result_path, model_name, dataset_name, new_result):
 
     print(f"Updated CSV file at {result_path} with new results for model '{model_name}' and dataset '{dataset_name}'.")
 
-def search_hyperparameters(parser, method, model):
+def search_hyperparameters(args, method, model):
+    setattr(args, 'grid_search_flag', True)
+    setattr(args, 'degree_flag', False)
     # Ensure method and model are valid
     if method in method_model_configs and model in method_model_configs[method]:
         # Get hyperparameters for the specific method and model
@@ -96,8 +100,30 @@ def search_hyperparameters(parser, method, model):
     else:
         print("Invalid method or model configuration.")
 
+def get_degree_accuracy(args, method, model):
+    setattr(args, 'grid_search_flag', False)
+    setattr(args, 'degree_flag', True)
+    args.logpath = setup_logpath(
+        folder_args=(args.data, args.model, args.flag),
+        quiet=args.quiet)
+    # Ensure method and model are valid
+    if method in method_model_configs and model in method_model_configs[method]:
+        optimal_path = os.path.join(args.logpath, args.conv+ "_optimal/config.json")
+        hyperparams = method_model_configs[method][model]
+        with open(optimal_path, 'r') as file:
+            config_dict = json.load(file)
+
+            # Convert the dictionary to a namespace object
+            for param in hyperparams:
+                setattr(args, param, int(config_dict[param])) if isinstance(config_dict[param], int) else setattr(args, param, float(config_dict[param]))
+            # args = argparse.Namespace(**config_dict)
+            print('args:',args.data)
+            main(args)
 
 
+    else:
+        print("Invalid method or model configuration.")
+
 def main(args):
     # ========== Run configuration
     args.logpath = setup_logpath(
@@ -129,12 +155,17 @@ def main(args):
     logger.log(logging.LRES, f"[res]: {res_logger}")
     current_accuracy = float(res_logger.get_str("f1micro_test", 0).split(":")[1])
     print("result:", current_accuracy)
+    f1micro_high = float(res_logger.get_str("f1micro_high", 0).split(":")[1])
+    f1micro_low = float(res_logger.get_str("f1micro_low", 0).split(":")[1])
+    print("f1micro_high:", f1micro_high, " f1micro_low", f1micro_low)
+
 
     res_logger.save()
     clear_logger(logger)
     global best_accuracy
     table_path = Path('../log/sum_table.csv')
-    if current_accuracy>best_accuracy:
+    degree_table_path = Path('../log/degree_table.csv')
+    if current_accuracy>best_accuracy and args.grid_search_flag:
         setattr(args, 'optimal_accuracy', current_accuracy)
         optimal_path = os.path.join(args.logpath, args.conv+ "_optimal")
         if not os.path.exists(optimal_path):
@@ -143,16 +174,24 @@ def main(args):
         best_accuracy = current_accuracy
         update_results_csv(table_path, args.model + "-" + args.conv+ "-" +args.theta, args.data, best_accuracy)
 
-
+    if args.degree_flag:
+        #Just output the degree accuracy using the optimal config
+        update_results_csv(degree_table_path, args.model + "-" + args.conv+ "-" +args.theta, args.data+'-'+'high', f1micro_high) 
+        update_results_csv(degree_table_path, args.model + "-" + args.conv+ "-" +args.theta, args.data+'-'+'low', f1micro_low)
 
 if __name__ == '__main__':
     parser = setup_argparse()
     # Experiment-specific arguments
     # parser.add_argument()
     args = setup_args(parser)
 
-    # Example usage
-    search_hyperparameters(args, args.model, args.conv)
+    # Example usage for grid search
+    # search_hyperparameters(args, args.model, args.conv)
+
+    # Get the degree accuracy when you have an optimal config
+    get_degree_accuracy(args, args.model, args.conv)
+
+
 
 
 

examples/scripts/grid_search_all.sh

@@ -1,5 +1,6 @@
 # Define the list of datasets
-# datasets=("Cora" "CiteSeer" "PubMed")
+# datasets=("Cora" "CiteSeer" "PubMed" "Texas" "Squirrel" "Chameleon")
+
 datasets=("texas" "squirrel" "chameleon")
 
 # Loop through each dataset and run the Python script with the parameters
@@ -8,15 +9,15 @@ for graph in "${datasets[@]}"; do
     #fixed-linear
     python run_grid_search.py --model IterGNN --conv FixLinSumAdj --data $graph --theta khop
     #fixed-Impulse 
-    python run_grid_search.py --model PreDecMLP --conv FixSumAdj --data $graph --theta khop
-    #fixed-Monomial
-    python run_grid_search.py --model PreDecMLP --conv FixSumAdj --data $graph --theta mono
-    #fixed-ppr
-    python run_grid_search.py --model PostMLP --conv FixSumAdj --data $graph --theta appr
-    #fixed-heart kernel
-    python run_grid_search.py --model PreDecMLP --conv FixSumAdj --data $graph --theta hk
-    #fixed-guassian
-    python run_grid_search.py --model PreDecMLP --conv FixSumAdj --data $graph --theta gaussian
+    # python run_grid_search.py --model PreDecMLP --conv FixSumAdj --data $graph --theta khop
+    # #fixed-Monomial
+    # python run_grid_search.py --model PreDecMLP --conv FixSumAdj --data $graph --theta mono
+    # #fixed-ppr
+    # python run_grid_search.py --model PreDecMLP --conv FixSumAdj --data $graph --theta appr
+    # #fixed-heart kernel
+    # python run_grid_search.py --model PreDecMLP --conv FixSumAdj --data $graph --theta hk
+    # #fixed-guassian
+    # python run_grid_search.py --model PreDecMLP --conv FixSumAdj --data $graph --theta gaussian
 
     #var-Linear-unfinished
     # python run_grid_search.py --model IterGNN --conv VarSumAdj --data $graph --theta khop
@@ -28,7 +29,7 @@ for graph in "${datasets[@]}"; do
     python run_grid_search.py --model PostMLP --conv ChebBase --data $graph
     #var-Chebyshev2
     python run_grid_search.py --model PostMLP --conv ChebConv2 --data $graph
-    #var-Bernstein
+    var-Bernstein
     python run_grid_search.py --model PostMLP --conv BernConv --data $graph
 
     #bank-linear

examples/trainer/fullbatch.py

@@ -135,7 +135,7 @@ def run(self) -> ResLogger:
         res_test = self.test()
         res_run.merge(res_test)
 
-        self.test_deg()
+        res_run.merge(self.test_deg())
 
         return self.res_logger.merge(res_run)