update paper, size up number of epochs for accuracy surrogate and add preview information about ensemble accuracy on random architectures

Author: Demoren1

Udeneev2025Surrogate.pdf

@@ -11,7 +11,7 @@
     GraphNorm,
 )
 from torch_geometric.utils import dense_to_sparse
-from torch.optim.lr_scheduler import StepLR
+from torch.optim.lr_scheduler import StepLR, CosineAnnealingLR
 from tqdm.notebook import tqdm
 import matplotlib.pyplot as plt
 import numpy as np
@@ -347,10 +347,11 @@ def train_model_diversity(
     num_epochs,
     device="cpu",
     developer_mode=False,
+    final_lr=0.001
 ):
     model.to(device)
     train_losses, valid_losses = [], []
-    scheduler = StepLR(optimizer, step_size=5, gamma=0.5)
+    scheduler = CosineAnnealingLR(optimizer, num_epochs, eta_min=final_lr)
 
     for epoch in tqdm(range(num_epochs), desc="Training Progress"):
         # --------------------
@@ -424,7 +425,8 @@ def train_model_diversity(
         except ImportError:
             pass
 
-        plt.figure(figsize=(10, 5))
+        plt.figure(figsize=(12, 6))
+        plt.rc('font', size=20)
         plt.plot(range(1, len(train_losses)+1), train_losses, marker='o', label='Train Loss')
         plt.plot(range(1, len(valid_losses)+1), valid_losses, marker='s', label='Valid Loss')
         plt.xlabel('Epoch')
@@ -449,12 +451,13 @@ def train_model_accuracy(
     num_epochs,
     device="cpu",
     developer_mode=False,
+    final_lr=0.001,
 ):
     model.to(device)
     train_losses = []
     valid_losses = []
 
-    scheduler = StepLR(optimizer, step_size=5, gamma=0.5)
+    scheduler = CosineAnnealingLR(optimizer, num_epochs, eta_min=final_lr)
 
     for epoch in tqdm(range(num_epochs), desc="Training Progress"):
         model.train()
@@ -511,6 +514,7 @@ def train_model_accuracy(
             pass
 
         plt.figure(figsize=(12, 6))
+        plt.rc('font', size=20)
         tmp_train_losses = np.array(train_losses)
         tmp_valid_losses = np.array(valid_losses)
         plt.plot(range(1, len(tmp_train_losses) + 1), tmp_train_losses * 1e4, marker="o", label="Train Loss")

code/GCN.py

@@ -0,0 +1,25 @@
+========================================================================
+Ensemble Accuracy: 93.62%
+Model 1 Accuracy: 90.95%
+Model 2 Accuracy: 90.92%
+Model 3 Accuracy: 90.57%
+Model 4 Accuracy: 91.02%
+Model 5 Accuracy: 90.76%
+Model 6 Accuracy: 90.38%
+========================================================================
+Ensemble Accuracy: 93.80%
+Model 1 Accuracy: 91.24%
+Model 2 Accuracy: 91.25%
+Model 3 Accuracy: 91.36%
+Model 4 Accuracy: 91.89%
+Model 5 Accuracy: 91.64%
+Model 6 Accuracy: 90.58%
+========================================================================
+Ensemble Accuracy: 94.18%
+Model 1 Accuracy: 91.64%
+Model 2 Accuracy: 91.51%
+Model 3 Accuracy: 91.24%
+Model 4 Accuracy: 92.18%
+Model 5 Accuracy: 91.99%
+Model 6 Accuracy: 91.70%
+========================================================================

code/best_models.zip

@@ -0,0 +1,8 @@
+import math
+
+def binom(n, m):
+    return math.factorial(n) // (math.factorial(n - m) * math.factorial(m))
+
+result = binom(2, 2) * binom(3, 2) * binom(4, 2) * binom(5, 2) * 7 ** 8
+
+print(result // 1e9)