AI开发之基础快速入门神经网络

import numpy as npimport torchimport torch.nn as nnimport torch.optim as optimRED    = "\033[31m"GREEN  = "\033[32m"YELLOW = "\033[33m"BOLD   = "\033[1m"END    = "\033[0m"# =====================================================================# 1) Prepare Sample Data (Web Traffic Features)# Feature shape details: [Request Frequency, 404 Count, Contains Sensitive Keywords (0/1), Average Response Size]# =====================================================================X_data = np.array([    [2.1,  0.0,  0, 0.1],   # 1) Normal traffic    [95.4, 42.0, 1, 8.5],   # 2) Vulnerability scanning (Attack)    [4.5,  1.0,  0, 0.5],   # 3) Normal traffic    [1.2,  0.0,  0, 0.0],   # 4) Normal traffic    [88.0, 35.0, 1, 12.1],  # 5) Credential stuffing (Attack)], dtype=np.float32)# Ground truth labels: 0.0 for normal traffic, 1.0 for malicious attacky_data = np.array([[0.0], [1.0], [0.0], [0.0], [1.0]], dtype=np.float32)visitor_names = ["User A", "User B", "User C", "User D", "User E"]# Convert NumPy arrays to PyTorch TensorsX_tensor = torch.tensor(X_data)y_tensor = torch.tensor(y_data)# =====================================================================# 2) Define Network Structure & Activation Functions# =====================================================================class WafModel(nn.Module):    def __init__(self):        super(WafModel, self).__init__()        # Hidden Layer: 4 input features mapped to 8 hidden abstract dimensions        self.hidden = nn.Linear(in_features=4, out_features=8)        # Output Layer: 8 hidden features mapped to 1 final probability output        self.output = nn.Linear(in_features=8, out_features=1)    def forward(self, x):        # Forward propagation data workflow        x = self.hidden(x)        x = torch.relu(x)     # Standard activation for hidden layers: ReLU        x = self.output(x)        x = torch.sigmoid(x)  # Final activation: Sigmoid to compress output into a 0~1 probability range        return x# Instantiate the WAF neural network modelmodel = WafModel()# =====================================================================# 3) Define Loss Function & Optimizer# =====================================================================# Binary Cross-Entropy Loss: The industry standard for binary classification (Malicious vs Normal)criterion = nn.BCELoss()# Adam Optimizer: Handles gradient updates dynamically with a learning rate of 0.01optimizer = optim.Adam(model.parameters(), lr=0.01)# =====================================================================# 4) Model Training Loop (Forward & Backward Optimization)# =====================================================================epochs = 100  # Number of times the AI loops through the entire training datasetprint("-" * 75)print(f"{BOLD}{YELLOW}===== AI Threat Perception System: Training PyTorch nn.Module Model ====={END}")print("-" * 75)print("Starting Mini-WAF Malicious Traffic Model Training...")for epoch in range(epochs):    # --- The Core Optimization Loop ---    # Forward Pass — Let the AI make predictions    predictions = model(X_tensor)    # Compute Loss — Measure how far off the predictions are from reality    loss = criterion(predictions, y_tensor)    # Backward Pass — Clear old gradients and backpropagate the error using chain rule    optimizer.zero_grad()  # Reset residual gradients from the previous iteration    loss.backward()        # Calculate how much weights (W) and biases (b) need to change    # Optimization Step — Update parameters based on calculated gradients    optimizer.step()    # Print training logs every 10 epochs    if (epoch + 1) % 10 == 0:        print(f"Epoch [{epoch + 1:03d}/{epochs}] | Current Loss: {loss.item():.4f}")print(f"\n{BOLD}{GREEN}Model training completed! Model parameters have successfully converged.{END}\n")# =====================================================================# 5) Verify AI Defense Performance# =====================================================================# Disable gradient computation since we are running evaluation, not trainingwith torch.no_grad():    final_preds = model(X_tensor).numpy().flatten()threshold = 0.5  # Risk control interception thresholdprint("-" * 75)print(f"{BOLD}{YELLOW}===== AI WAF Gateway: Real-Time Live Traffic Inspection Active ====={END}")print("-" * 75)for i, (name, score) in enumerate(zip(visitor_names, final_preds)):    if score > threshold:        action = f"{BOLD}{RED}[BLOCK & DROP] - Intercepted by AI Neural Network! (Risk Prob: {score*100:.2f}%){END}"    else:        action = f"{BOLD}{GREEN}[PASS] - Clear Traffic. (Risk Prob: {score*100:.2f}%){END}"    print(f"Live Flow: {name:<12} -> Predicted: {action}")print("-" * 75)