AI开发之Web基础快速入门主流深度学习框架介绍PyTorch

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"#    Construct traffic samples and standard data preprocessing#    Feature format: [Request frequency, 404 count, Is robot UA, Page size MB]#    Label definition: 0 = Normal user access, 1 = Malicious attack behaviorif __name__ == "__main__":    # 1) Raw Web traffic feature matrix    X_data = np.array([        [2.1,  0.0,  0, 0.1],   # 3) Normal traffic        [95.4, 42.0, 1, 8.5],   # 4) Vulnerability scanning (Attack)        [4.5,  1.0,  0, 0.5],   # 5) Normal traffic        [1.2,  0.0,  0, 0.0],   # 6) Normal traffic        [88.0, 35.0, 1, 12.1],  # 7) Credential stuffing (Attack)    ], dtype=np.float32)    # True black and white labels (Ground truth)    y_data = np.array([[0], [1], [0], [0], [1]], dtype=np.float32)    visitor_names = ["User A", "User B", "User C", "User D", "User E"]    # Feature normalization: Scale data to a uniform magnitude    mean = X_data.mean(axis=0)    std = X_data.std(axis=0) + 1e-8    X_data_scaled = (X_data - mean) / std    # Convert NumPy arrays to PyTorch Tensors    X_tensor = torch.tensor(X_data_scaled, dtype=torch.float32)    y_tensor = torch.tensor(y_data, dtype=torch.float32)    # 2) Build the neural network layer by layer (PyTorch nn.Sequential base)    #     Sequential acts as a linear stack of layers similar to Keras    model = nn.Sequential(        # Layer 1 (Hidden Layer): 8 neurons, takes 4-dimensional input features, uses ReLU activation        nn.Linear(in_features=4, out_features=8),        nn.ReLU(),        # Layer 2 (Output Layer): 1 neuron, uses Sigmoid activation to output a risk probability between 0 and 1        nn.Linear(in_features=8, out_features=1),        nn.Sigmoid()    )    # 3) Configure loss function and optimizer    #     BCELoss stands for Binary Cross Entropy Loss    criterion = nn.BCELoss()    optimizer = optim.SGD(model.parameters(), lr=0.1)    # 4) Start the training loop to iteratively optimize parameters    print("-" * 75)    print(f"{BOLD}{YELLOW}===== AI Threat Perception System: Training PyTorch Neural Network ====={END}")    print("-" * 75)    print("PyTorch sequential architecture initialized, starting training loop...")    epochs = 1000    for epoch in range(1, epochs + 1):        # Forward pass: Compute predicted probabilities by passing X through the model        y_pred = model(X_tensor)        # Compute loss: Calculate the distance between predictions and ground truth        loss = criterion(y_pred, y_tensor)        # Backward pass: Clear old gradients, compute new gradients via backpropagation        optimizer.zero_grad()        loss.backward()        # Parameter update: Apply computed gradients to shift parameters towards global minimum        optimizer.step()    print(f"\n{BOLD}{GREEN}Model training completed! Model parameters have successfully converged.{END}\n")    # 5) Online inference and real-time Web gateway interception    #     torch.no_grad() disables gradient calculation to speed up evaluation and save memory    with torch.no_grad():        probabilities = model(X_tensor).numpy().flatten()    threshold = 0.5    print("-" * 75)    print(f"{BOLD}{YELLOW}===== AI WAF Gateway: Real-Time Live Traffic Inspection Active ====={END}")    print("-" * 75)    for name, score in zip(visitor_names, probabilities):        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)