Quick to start

Pytorch vs Tensorflow

  • Pytorch + Caffine2 :
    1. FaceBook開發
    2. 一邊推一邊算
    3. 靈活，方便調適，主要學術界
  • Tensorflow2.0+Karis :
    1. Google開發
    2. 先寫公式，在代數字
    3. 開發時間早，生態好，主要工業界

創建 Models

繼承 nn.Module
def __init__(self,) : 定義網路 每層結構(layer)
def forward(self) : 資料如何在網路中傳遞
super() = super(className,self)，找到MOR裡，claaaName後，最先有__init__
ex:

python
1
2
3
4
5
6
class MyNetWork(nn.Module):
  super().__init__()
  self.model = nn.Sequential(
    models...
  )
  def forward(self, ...):

查看

plaintext
1
2
model = MyNetWork()
print(model)

Optimizing(最佳化) Model Parameters

訓練模型需要:

  • Loss Function
  • Optimizer
python
1
2
loss_fn = nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(model.paramet)

在單個訓練循環中，模型對訓練數據集（分批輸入）進行預測，並反向傳播預測誤差以調整模型的參數。

py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
def train(dataloader, model, loss_fn, optimizer):
  size = len(dataloader.dataset)
  model.train()
  for batch, (X, y) in enumerate(dataloader):
    X, y = X.to(device), y.to(device)

    # Compute prediction error
    pred = model(X)
    loss = loss_fn(pred, y)

    # Backpropagation
    optimizer.zero_grad()
    loss.backward()
    optimizer.step()

    if batch % 100 == 0:
        loss, current = loss.item(), (batch + 1) * len(X)
        print(f"loss: {loss:>7f}  [{current:>5d}/{size:>5d}]")

訓練可在cpu或gpu

py
1
2
device = "cuda" if torch.cuda.is_available() else "cpu"
print(f"using {device}")

we can put instance to cpu or gpu

plaintext
1
model = className().to(device)

對比模型在test dataset以確保學習

py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
def test(dataloader, model, loss_fn):
    size = len(dataloader.dataset)
    num_batches = len(dataloader)
    model.eval()
    test_loss, correct = 0, 0
    with torch.no_grad():
        for X, y in dataloader:
            X, y = X.to(device), y.to(device)
            pred = model(X)
            test_loss += loss_fn(pred, y).item()
            correct += (pred.argmax(1) == y).type(torch.float).sum().item()
    test_loss /= num_batches
    correct /= size
    print(f"Test Error: \n Accuracy: {(100*correct):>0.1f}%, Avg loss: {test_loss:>8f} \n")

每個epoch 模型
我們會打印模型accuracy和loss

py
1
2
3
4
5
6
epochs = 5
for t in range(epochs):
    print(f"Epoch {t+1}\n-------------------------------")
    train(train_dataloader, model, loss_fn, optimizer)
    test(test_dataloader, model, loss_fn)
print("Done!")

保存模型

py
1
2
torch.save(model.state_dict(), "model.pth")
print("Saved PyTorch Model State to model.pth")

加載模型

py
1
2
model = NeuralNetwork()
model.load_state_dict(torch.load("model.pth"))