File size: 7,891 Bytes
bf620c6 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 |
import torch
import torch.nn.functional as F
from torch_geometric.datasets import Planetoid
from torch_geometric.nn import GCNConv
import torch_geometric.transforms as T
import numpy as np
import argparse
import networkx as nx
class GCN(torch.nn.Module):
def __init__(self, in_channels, hidden_channels, out_channels, dropout=0.5):
super().__init__()
self.conv1 = GCNConv(in_channels, hidden_channels)
self.conv2 = GCNConv(hidden_channels, out_channels)
self.dropout = dropout
def forward(self, x, edge_index):
x = self.conv1(x, edge_index)
x = F.relu(x)
x = F.dropout(x, p=self.dropout, training=self.training)
x = self.conv2(x, edge_index)
return x
def evaluate_integration_methods(dataset_name='Cora'):
"""Test different ways to integrate the L-RMC component"""
# Load data
dataset = Planetoid(root="./data", name=dataset_name, transform=T.NormalizeFeatures())
data = dataset[0]
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
data = data.to(device)
print(f"Total number of edges: {data.num_edges}")
print(f"Total number of nodes: {data.num_nodes}")
# L-RMC component (convert from 1-indexed Java IDs back to PyG indices)
lrmc_nodes_java = "1680 2883 901 617 1578 1103"
# Load the saved remap dictionary from the benchmark script
import pickle
remap_filename = f"{dataset_name.lower()}_remap.pkl"
with open(remap_filename, "rb") as f:
remap_data = pickle.load(f)
nodes_sorted = remap_data['nodes_sorted']
# Convert Java 1-indexed IDs back to PyG indices
lrmc_nodes_java_ids = [int(x) for x in lrmc_nodes_java.split()]
lrmc_nodes = [nodes_sorted[j-1] for j in lrmc_nodes_java_ids]
print(f"Loaded {len(lrmc_nodes)} L-RMC nodes with correct ID mapping")
print(f"Java IDs: {lrmc_nodes_java_ids[:5]}...")
print(f"PyG IDs: {lrmc_nodes[:5]}...")
# Verify connectivity and S(C) on correct nodes
lrmc_set = set(lrmc_nodes)
print("\nVerification:")
print(f"L-RMC component size: {len(lrmc_set)}")
# Check connectivity within the component
G = nx.Graph()
edges = data.edge_index.t().tolist()
G.add_edges_from(edges)
# Create subgraph of L-RMC component
lrmc_subgraph = G.subgraph(lrmc_nodes)
internal_edges = lrmc_subgraph.number_of_edges()
possible_edges = len(lrmc_nodes) * (len(lrmc_nodes) - 1) // 2
density = internal_edges / possible_edges if possible_edges > 0 else 0
print(f"Internal edges in L-RMC: {internal_edges}")
print(f"Possible edges: {possible_edges}")
print(f"Density: {density:.4f}")
# Calculate S(C) - sum of degrees within component
s_c = 0
for node in lrmc_nodes:
neighbors = set(G.neighbors(node))
internal_neighbors = neighbors & lrmc_set
s_c += len(internal_neighbors)
print(f"S(C): {s_c}")
print(f"Average internal degree: {s_c / len(lrmc_nodes):.2f}")
integration_methods = {
"baseline": "no_modification",
"binary_feature": "add_binary_feature",
"feature_scaling": "scale_features_2x",
"feature_scaling_strong": "scale_features_5x",
"feature_boost": "add_constant_to_features",
"edge_weights": "boost_internal_edges"
}
results = {}
for method_name, method_type in integration_methods.items():
print(f"\n{'='*50}")
print(f"Testing: {method_name}")
print(f"{'='*50}")
accuracies = []
for seed in range(42, 60): # 5 runs
torch.manual_seed(seed)
np.random.seed(seed)
# Apply integration method
if method_type == "no_modification":
x_input = data.x
edge_index = data.edge_index
elif method_type == "add_binary_feature":
component_feature = torch.zeros(data.num_nodes, 1, device=device)
component_feature[lrmc_nodes] = 1.0
x_input = torch.cat([data.x, component_feature], dim=1)
edge_index = data.edge_index
elif method_type == "scale_features_2x":
x_input = data.x.clone()
x_input[lrmc_nodes] *= 2.0
edge_index = data.edge_index
elif method_type == "scale_features_5x":
x_input = data.x.clone()
x_input[lrmc_nodes] *= 5.0
edge_index = data.edge_index
elif method_type == "add_constant_to_features":
x_input = data.x.clone()
x_input[lrmc_nodes] += 1.0 # Add constant to all features
edge_index = data.edge_index
elif method_type == "boost_internal_edges":
x_input = data.x
edge_index = data.edge_index.clone()
# Create additional "virtual" edges within the component
lrmc_set = set(lrmc_nodes)
virtual_edges = []
for i, u in enumerate(lrmc_nodes):
for v in lrmc_nodes[i+1:]:
# Add multiple copies of internal edges to boost them
for _ in range(3): # Triple the internal edges
virtual_edges.extend([[u, v], [v, u]])
if virtual_edges:
virtual_edge_tensor = torch.tensor(virtual_edges, device=device).t()
edge_index = torch.cat([edge_index, virtual_edge_tensor], dim=1)
# Train model
model = GCN(x_input.shape[1], 16, dataset.num_classes).to(device)
optimizer = torch.optim.Adam(model.parameters(), lr=0.01, weight_decay=5e-4)
best_val_acc = 0
test_acc_at_best_val = 0
for epoch in range(200):
model.train()
optimizer.zero_grad()
out = model(x_input, edge_index)
loss = F.cross_entropy(out[data.train_mask], data.y[data.train_mask])
loss.backward()
optimizer.step()
if epoch % 20 == 0:
model.eval()
with torch.no_grad():
pred = model(x_input, edge_index).argmax(dim=1)
val_correct = pred[data.val_mask] == data.y[data.val_mask]
val_acc = int(val_correct.sum()) / int(data.val_mask.sum())
if val_acc > best_val_acc:
best_val_acc = val_acc
test_correct = pred[data.test_mask] == data.y[data.test_mask]
test_acc_at_best_val = int(test_correct.sum()) / int(data.test_mask.sum())
accuracies.append(test_acc_at_best_val)
print(f" Seed {seed}: {test_acc_at_best_val:.4f}")
mean_acc = np.mean(accuracies)
std_acc = np.std(accuracies)
results[method_name] = (mean_acc, std_acc)
print(f" Mean ± Std: {mean_acc:.4f} ± {std_acc:.4f}")
# Final summary
print(f"\n{'='*60}")
print("FINAL INTEGRATION METHOD COMPARISON")
print(f"{'='*60}")
print(f"{'Method':<20} | {'Mean Accuracy':<15} | {'Std Dev':<10} | {'vs Baseline'}")
print("-" * 70)
baseline_acc = results["baseline"][0]
for method, (mean_acc, std_acc) in results.items():
improvement = mean_acc - baseline_acc
print(f"{method:<20} | {mean_acc:<15.4f} | {std_acc:<10.4f} | {improvement:+.4f}")
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Evaluate L-RMC integration methods on a specified dataset')
parser.add_argument('--dataset', type=str, default='Cora',
choices=['Cora', 'CiteSeer', 'PubMed'],
help='Dataset to use (default: Cora)')
args = parser.parse_args()
evaluate_integration_methods(args.dataset)
|