机器学习 - 神经网络中的训练模型

接着上一篇机器学习-创建一个PyTorch classification model做进一步陈述。

训练模型的步骤:

  1. Forward pass : The model goes through all of the training data once, performing its forward() function calculations (model(x_train))
  2. Calculate the loss : 使用 loss = loss_fn(y_pred, y_train)
  3. Zero gradients : optimizer.zero_grad()
  4. Perform backpropagation on the loss : Computes the gradient of the loss with respect for every model parameter to be updated (each parameter with requires_grad=True). This is known as backpropagation, hence "backwards" (loss.backward())
  5. Step the optimizer (gradient descent) : Update the parameters with requires_grad = True with respect to the loss gradients in order to improve them (optimizer.step())

python 复制代码
# View the first 5 outputs of the forward pass on the test data
y_logits = model_0(X_test.to("cpu"))[:5]
print(y_logits)

# Use sigmoid on model logits 
y_pred_probs = torch.sigmoid(y_logits)
print(y_pred_probs)

# Find the predicted labels (round the prediction probabilities)
y_preds = torch.round(y_pred_probs)

# In full 
y_pred_labels = torch.round(torch.sigmoid(model_0(X_test))[:5])

# Check for equality 
print(torch.eq(y_preds.squeeze(), y_pred_labels.squeeze()))

print(y_preds.squeeze())

# 结果如下
tensor([[ 0.3798],
        [ 0.2257],
        [ 0.4383],
        [ 0.3647],
        [-0.1101]], grad_fn=<SliceBackward0>)
tensor([[0.5938],
        [0.5562],
        [0.6078],
        [0.5902],
        [0.4725]], grad_fn=<SigmoidBackward0>)
tensor([True, True, True, True, True])
tensor([1., 1., 1., 1., 0.], grad_fn=<SqueezeBackward0>)

The use of the sigmoid activation function is often only for binary classification logits.

The use of the sigmoid activation function is not required when passing the model's raw outputs to the nn.BCEWithLogitsLoss (the "logits" in logits loss is because it works on the model's raw logits output), this is because it has a sigmoid function built-in.


创建 training 和 testing loop

python 复制代码
# 创建一个 loss function
loss_fn = nn.BCEWithLogitsLoss()

def accuracy_fn(y_true, y_pred):
  correct = torch.eq(y_true, y_pred).sum().item()
  acc = (correct / len(y_pred)) * 100
  return acc

# Build a train and test loop 

torch.manual_seed(42)

# Set the number of epochs
epochs = 100

# Put data into device
X_train, y_train = X_train.to("cpu"), y_train.to("cpu")
X_test, y_test = X_test.to("cpu"), y_test.to("cpu")

# Build training and evaluation loop
for epoch in range(epochs):
  ### Training
  model_0.train()

  # 1. Forward pass (model outputs raw logits)
  y_logits = model_0(X_train).squeeze()
  y_pred = torch.round(torch.sigmoid(y_logits)) # turn logits -> pred probs -> pred labls

  # 2. Calculate loss/accuracy
  loss = loss_fn(y_logits,
                 y_train)
  acc = accuracy_fn(y_true = y_train,
                    y_pred = y_pred)
  
  # 3. Optimizer zero grad 
  optimizer.zero_grad()

  # 4. Loss backwards
  loss.backward()

  # 5. Optimizer step 
  optimizer.step() 

  ### Testing 
  model_0.eval()
  with torch.inference_mode():
    # 1. Forward pass
    test_logits = model_0(X_test).squeeze()
    test_pred = torch.round(torch.sigmoid(test_logits))
    # 2. Caculate loss/accuracy
    test_loss = loss_fn(test_logits,
                        y_test)
    test_acc = accuracy_fn(y_true = y_test,
                           y_pred = test_pred)
  
  if epoch % 10 == 0:
    print(f"Epoch: {epoch} | Loss: {loss:.5f}, Accuracy: {acc:.2f}% | Test loss: {test_loss:.5f}, Test acc: {test_acc:.2f}%")

# 输出结果
Epoch: 0 | Loss: 0.70758, Accuracy: 50.25% | Test loss: 0.70294, Test acc: 56.00%
Epoch: 10 | Loss: 0.70192, Accuracy: 50.25% | Test loss: 0.69895, Test acc: 52.50%
Epoch: 20 | Loss: 0.69892, Accuracy: 50.00% | Test loss: 0.69713, Test acc: 50.00%
Epoch: 30 | Loss: 0.69716, Accuracy: 49.75% | Test loss: 0.69626, Test acc: 51.50%
Epoch: 40 | Loss: 0.69603, Accuracy: 49.75% | Test loss: 0.69582, Test acc: 51.50%
Epoch: 50 | Loss: 0.69527, Accuracy: 49.75% | Test loss: 0.69561, Test acc: 51.00%
Epoch: 60 | Loss: 0.69474, Accuracy: 49.25% | Test loss: 0.69551, Test acc: 52.50%
Epoch: 70 | Loss: 0.69435, Accuracy: 49.00% | Test loss: 0.69547, Test acc: 51.00%
Epoch: 80 | Loss: 0.69406, Accuracy: 49.75% | Test loss: 0.69545, Test acc: 51.00%
Epoch: 90 | Loss: 0.69384, Accuracy: 49.25% | Test loss: 0.69545, Test acc: 51.50%

看到这了,给个赞呗~

相关推荐
sanxing三省4 分钟前
DeepSeek Harness 到底是什么?标题里最重的词其实不是「DeepSeek」
前端·人工智能·deepseek
合尘猫1 小时前
用 6GB 显卡搭短视频产线:MiniMax H3 + ffmpeg 全流程实录(55 秒片 / 40 分钟)
人工智能·音视频
ZONGYUAN-ROOT1 小时前
【自愈治理】自愈式数据管道:大规模公开数据采集的治理方法
人工智能
Starry-sky(jing)1 小时前
Hermes Agent 升级 v0.23.2:先辨更新渠道,再看 CLI 退出残留修复
人工智能·python·ai agent
oil欧哟2 小时前
AI Agent Skill 入门指南:文件夹本质、渐进式加载与挑选方法
人工智能·ai编程
百度Geek说2 小时前
从代码生成到需求交付:DarkFactory 的工程实践
人工智能
Baihai_IDP2 小时前
使用场景不同,模型“推理速度慢”的原因也不同
人工智能·面试·llm
挖掘狂人2 小时前
AI 三分钟写完的代码,你凭什么敢直接上线?
人工智能·程序员·ai编程
墨林陌2 小时前
AI 热点日报(2026-11-20):苹果自研 AI 智能体曝光,Anthropic 上市首日大涨,智能体支付落地
人工智能