具有 sigmoid 激活的 Pytorch 模型仅输出 0 或 1 进行预测，而不是实际概率

Question

我的 pytorch 神经网络只给我 1 和 0，而不是实际概率，即使我使用的是 sigmoid 函数

X_train, X_val, y_train, y_val = train_test_split(scaled_X, y, 
stratify=y, test_size=0.3, random_state=42)

X_train_tensor = torch.tensor(X_train, dtype=torch.float32)
y_train_tensor = torch.tensor(y_train, 
dtype=torch.float32).reshape(-1, 1)
X_val_tensor = torch.tensor(X_val, dtype=torch.float32)
y_val_tensor = torch.tensor(y_val.values).reshape(-1, 1)

test_tensor = torch.tensor(test.values, dtype=torch.float32)

class Network(nn.Module):
def __init__(self):
    super(Network, self).__init__()
    
    self.fc1 = nn.Linear(13, 2024)
    self.relu1 = nn.ReLU()
    self.dropout1 = nn.Dropout(0.3)
    self.bn1 = nn.BatchNorm1d(2024)
    
    self.fc2 = nn.Linear(2024, 2024)
    self.relu2 = nn.ReLU()
    self.dropout2 = nn.Dropout(0.3)
    self.bn2 = nn.BatchNorm1d(2024)
    
    self.fc3 = nn.Linear(2024, 2024)
    self.relu3 = nn.ReLU()
    self.dropout3 = nn.Dropout(0.3)
    self.bn3 = nn.BatchNorm1d(2024)
    
    self.fc5 = nn.Linear(2024, 1)
    self.sigmoid = nn.Sigmoid()

 def forward(self, x):
    x = self.fc1(x)
    x = self.relu1(x)
    x = self.dropout1(x)
    x = self.bn1(x)
    
    x = self.fc2(x)
    x = self.relu2(x)
    x = self.dropout2(x)
    x = self.bn2(x)
    
    x = self.fc3(x)
    x = self.relu3(x)
    x = self.dropout3(x)
    x = self.bn3(x)
    
    x = self.fc5(x)
    x = self.sigmoid(x)
    
    return x

def model_train(model, X_train, y_train, X_val, y_val):
# loss function and optimizer
    loss_fn = nn.BCELoss()  # binary cross entropy
    optimizer = optim.Adam(model.parameters(), lr=0.0001)

    n_epochs = 50   # number of epochs to run
    batch_size = 500  # size of each batch
    batch_start = torch.arange(0, len(X_train), batch_size)

    # Hold the best model
    best_acc = - np.inf   # init to negative infinity
    best_weights = None

    for epoch in range(n_epochs):
        model.train()
        with tqdm.tqdm(batch_start, unit="batch", mininterval=0, 
        disable=False) as bar:
            bar.set_description(f"Epoch {epoch}")
            for start in bar:
                # take a batch
                X_batch = X_train[start:start+batch_size]
                y_batch = y_train[start:start+batch_size]
                # forward pass
                y_pred = model(X_batch)
                loss = loss_fn(y_pred, y_batch)
            # backward pass
            optimizer.zero_grad()
            loss.backward()
            # update weights
            optimizer.step()
            # print progress
            acc = (y_pred.round() == y_batch).float().mean()
            bar.set_postfix(
                loss=float(loss),
                acc=float(ACC)
            )
        # evaluate accuracy at end of each epoch
        model.eval()
    y_pred = model(X_val)
        acc = (y_pred.round() == y_val).float().mean()
        acc = float(ACC)
        if acc > best_acc:
            best_acc = ACC
            best_weights = copy.deepcopy(model.state_dict())
    # restore model and return best accuracy
    model.load_state_dict(best_weights)
    return best_acc

best_validation_accuracy = model_train(model, X_train_tensor, 
y_train_tensor, X_val_tensor, y_val_tensor)
print("Best Validation Accuracy:", best_validation_accuracy)

with torch.no_grad():
    model.eval()
    y_pred_prob = model(test_tensor).numpy().flatten()

问题是预测给我的唯一值是零，而不是实际概率；这与我的预期不同。我需要理解它为什么这样做。我的代码很容易理解：

我尝试通过kaggle查看其他笔记本是否有同样的问题。我尝试修改我的神经网络的架构，改变批量大小、纪元大小和学习率。我询问了 chatGPT 并在 Google 上进行了搜索，但没有找到我的问题的明确答案。

Answer 1

零或一已经表示模型输出的概率，只是模型过度拟合或过度自信。