Version_4/train.py

import json
import torch
import torch.nn as nn
import torch.optim as optim
from models.gnn import PhishingGNN_Model
from pipeline.graph_engine import TopologicalGraphEngine
from config import *

def run_training_pipeline():
    print("[*] Launching Production Defender V5 Training Run...")
    
    # 1. Try to load real data, fallback to mock if missing
    try:
        with open("network_telemetry.json", "r") as f:
            dataset = json.load(f)
            print(f"[+] Loaded {len(dataset)} real telemetry logs.")
    except FileNotFoundError:
        print("[-] network_telemetry.json not found. Falling back to simulated context.")
        dataset = [
            {'ip': '185.220.101.5', 'domain': 'shadow-dns-bypass.net', 'asn': 44050, 'is_malicious': 1.0},
            {'ip': '193.56.28.14', 'asn': 57099, 'is_malicious': 1.0},
            {'ip': '8.8.8.8', 'domain': 'dns.google', 'asn': 15169, 'is_malicious': 0.0}
        ]

    # 2. Extract logs and structure target threat labels
    engine = TopologicalGraphEngine()
    x_dict, edge_index_dict = engine.extract_and_build(dataset)
    
    labels_list = []
    for ip_str in engine.ip_map.keys():
        matching_logs = [log for log in dataset if log.get('ip') == ip_str]
        label = matching_logs[0].get('is_malicious', 0.0) if matching_logs else 0.0
        labels_list.append([label])
        
    labels = torch.tensor(labels_list, dtype=torch.float32)

    # 3. Model setup
    in_channels_dict = {'ip': 16, 'domain': 32, 'asn': 8, 'cert': 16}
    model = PhishingGNN_Model(
        metadata=GRAPH_METADATA,
        in_channels_dict=in_channels_dict,
        hidden_channels=HIDDEN_CHANNELS,
        num_heads=NUM_HEADS,
        num_layers=NUM_LAYERS,
        dropout_rate=DROPOUT_RATE
    )
    
    optimizer = optim.AdamW(model.parameters(), lr=0.0005, weight_decay=1e-3)
    criterion = nn.BCELoss()

    # 4. Optimization Loop
    model.train()
    for epoch in range(100):
        optimizer.zero_grad()
        predictions = model(x_dict, edge_index_dict)
        
        # Only calculate loss on the nodes we have labels for
        valid_preds = predictions[:len(labels)]
        loss = criterion(valid_preds, labels)
        
        loss.backward()
        optimizer.step()
        
        if (epoch + 1) % 20 == 0:
            print(f"Epoch {epoch+1:03d}/100 | Topological Loss: {loss.item():.5f}")

    model.safe_save(MODEL_SAVE_PATH)
    print(f"[+] Operational weights successfully frozen at: {MODEL_SAVE_PATH}")

if __name__ == "__main__":
    run_training_pipeline()